## Thinking Path
> - Paperclip is the open source app people use to manage AI agents for
work
> - Agents execute through adapters; the codex_local adapter runs the
Codex CLI and reports each run's outcome, including an error message
when the CLI exits nonzero
> - When no error can be parsed from the CLI's JSONL output, `toResult`
in `packages/adapters/codex-local/src/server/execute.ts` falls back to
the first non-empty stderr line as the run error
> - The adapter itself passes the approvals-bypass flag, so the CLI's
first stderr line is always the benign startup warning "YOLO mode is
enabled. All tool calls will be automatically approved."
> - Failed runs therefore record that warning as their error, hiding the
real cause (for example an OpenAI API error further down in stderr) and
making failures hard to diagnose from the run record
> - This pull request derives the fallback error from the first
meaningful stderr line, skipping a conservative set of known benign
lines, and keeps the existing behavior when every line is benign
> - The benefit is that failed Codex runs surface the actual failure
reason instead of a harmless startup warning, without ever producing an
emptier message than before
## Linked Issues or Issue Description
No public issue exists for the codex_local case. The same bug class was
fixed for gemini-local in Refs #5099 and Refs #3476; this PR applies the
equivalent fix to codex_local.
**What happened?**
On a multi-tenant cloud deployment of Paperclip, several codex_local
runs failed and their run records showed `error_code=adapter_failed`
with the error text "YOLO mode is enabled. All tool calls will be
automatically approved." That is a benign Codex CLI startup warning,
printed on every run because the adapter passes the approvals-bypass
flag itself. The real failure (an OpenAI API error printed later in
stderr) was never surfaced.
**Expected behavior**
When the Codex CLI exits nonzero and no error was parsed from its JSONL
output, the run error should be the first stderr line that actually
explains the failure, not a startup warning the adapter itself provoked.
**Steps to reproduce**
1. Configure a codex_local agent and make the underlying Codex CLI
invocation fail after startup (for example, configure a model id the
active credentials cannot use).
2. Run the agent so the CLI exits nonzero with no parsed JSONL error.
3. Inspect the run's error message: it shows the YOLO approvals warning
(the first stderr line) instead of the real error printed further down
in stderr.
## What Changed
- Added `firstMeaningfulStderrLine` next to `firstNonEmptyLine` in
`packages/adapters/codex-local/src/server/execute.ts`, with a
conservative benign-line predicate covering the YOLO approvals warning
and `[paperclip] ...` diagnostic lines the adapter injected (for example
ACP fallback notes).
- Used it only in the `toResult` fallback error derivation. If every
stderr line is benign, the existing chain still applies (first non-empty
line, then `Codex exited with code N`), so the message never gets
emptier than today. Logging is unchanged.
- Added
`packages/adapters/codex-local/src/server/execute.stderr-error.test.ts`:
four end-to-end cases through `execute()` with a mocked CLI process,
plus unit coverage for the new helper. Tests were written first and
confirmed failing before the fix.
## Verification
- `pnpm exec vitest run
packages/adapters/codex-local/src/server/execute.stderr-error.test.ts`
(7 tests pass; 5 failed before the fix as expected)
- `pnpm exec vitest run packages/adapters/codex-local` (21 files, 188
tests pass)
- `pnpm run typecheck` in `packages/adapters/codex-local` (clean)
## Risks
Low risk. Only the derived fallback `errorMessage` changes, and only
when a benign line would otherwise have been picked; parsed JSONL
errors, logging, retry/quota/auth classification inputs, and the
empty-stderr exit-code fallback are untouched. The benign-line list is
deliberately conservative (exact prefixes) so real errors are never
skipped.
## Model Used
Claude Fable 5 (claude-fable-5), extended thinking, via Claude Code
## Checklist
- [x] I have included a thinking path that traces from project context
to this change
- [x] I have specified the model used (with version and capability
details)
- [x] I have checked ROADMAP.md and confirmed this PR does not duplicate
planned core work
- [x] I have searched GitHub for duplicate or related PRs and linked
them above
- [x] I have either (a) linked existing issues with `Fixes: #` / `Closes
#` / `Refs #` OR (b) described the issue in-PR following the relevant
issue template
- [x] I have not referenced internal/instance-local Paperclip issues or
links (only public GitHub `#NNN` / `github.com/paperclipai/paperclip`
URLs)
- [x] My branch name describes the change (e.g. `docs/...`, `fix/...`)
and contains no internal Paperclip ticket id or instance-derived details
- [x] I have run tests locally and they pass
- [x] I have added or updated tests where applicable
- [x] I have updated relevant documentation to reflect my changes
- [x] I have considered and documented any risks above
- [x] All Paperclip CI gates are green
- [x] Greptile is 5/5 with no open P2s, recommendations, or follow-ups
- [x] I will address all Greptile and reviewer comments before
requesting merge
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
## Thinking Path
> - Paperclip helps people manage AI agents for work.
> - Agent adapters connect Paperclip to tools such as the Codex command
line tool.
> - A sandboxed Codex agent may start without a credential.
> - The operator needs a safe sign-in flow that does not expose
credentials to the shared package or the sandbox.
> - This pull request adds a company-scoped device-login flow with a
temporary Daytona sandbox.
> - The flow promotes the credential only after readiness checks pass
and removes the temporary sandbox after use.
> - The result lets an operator sign in to a sandboxed Codex agent from
the agent form.
## Linked Issues or Issue Description
**Subsystem affected**
Cross-cutting (multiple of the above)
**Problem or motivation**
A Codex adapter that runs in a sandbox cannot authenticate when the
company has no pre-provisioned Codex credential.
**Proposed solution**
Add a company-scoped device-login session. Start a temporary sandbox,
run `codex login --device-auth`, stream the code and URL, verify
readiness, promote the credential, and delete the sandbox.
**Alternatives considered**
Pre-provisioning a credential does not support first-time sandbox login.
Keeping the credential in the login sandbox does not provide a durable
company credential.
**Roadmap alignment**
This supports the roadmap item for cloud and sandbox agents.
**Additional context**
The flow uses a five-minute cleanup reaper, compare-and-set status
changes, and a PostgreSQL advisory lock to protect promotion and
cleanup.
## What Changed
- Add the adapter login-session contract, database table, and migration.
- Add company-scoped server routes and a service for sandbox device
login.
- Add credential promotion, readiness checks, and cleanup after login.
- Add restart-safe cleanup for abandoned login sandboxes.
- Add sandbox login controls to the agent creation and edit forms.
- Keep device-login and vendor identifiers out of public shared and
adapter UI symbols.
## Verification
- `pnpm --filter @paperclipai/adapter-codex-local exec vitest run`
passed with 310 tests at the submitted commit.
- The server login route, service, and reaper tests passed with 45 tests
at the submitted commit.
- The agent form render tests passed with 26 tests at the submitted
commit.
- The public-symbol leak check passed at the submitted commit.
- A live Daytona sign-in flow still requires confirmation by a user with
a live sandbox.
## Risks
The migration adds a new company-scoped table. A promotion or cleanup
race could remove a credential or leave a sandbox active, so the service
uses claims, compare-and-set transitions, and an advisory lock. The live
Daytona flow needs operator confirmation because local tests do not
provide a real browser sign-in.
## Model Used
OpenAI Codex, GPT-5, tool use and code execution, extended reasoning.
## Checklist
- [x] I have included a thinking path that traces from project context
to this change
- [x] I have specified the model used (with version and capability
details)
- [x] I have checked ROADMAP.md and confirmed this PR does not duplicate
planned core work
- [x] I have searched GitHub for duplicate or related PRs and linked
them above
- [x] I have either (a) linked existing issues with `Fixes: #` /
`Closes: #` / `Refs: #` OR (b) described the issue in-PR following the
relevant issue template
- [x] I have not referenced internal/instance-local Paperclip issues or
links (only public GitHub `#NNN` / `github.com/paperclipai/paperclip`
URLs)
- [x] My branch name describes the change (e.g. `docs/...`, `fix/...`)
and contains no internal Paperclip ticket id or instance-derived details
- [x] I have run tests locally and they pass
- [x] I have added or updated tests where applicable
- [x] I have updated relevant documentation to reflect my changes
- [x] I have considered and documented any risks above
- [x] All Paperclip CI gates are green
- [x] Greptile is 5/5 with no open P2s, recommendations, or follow-ups
- [x] I will address all Greptile and reviewer comments before
requesting merge
---------
Co-authored-by: Paperclip <noreply@paperclip.ing>
## Thinking Path
> - Paperclip connects AI agents to local and remote runtimes.
> - The Codex local adapter needs a safe device-login flow.
> - A future sandbox integration needs strict prompt validation, secret
protection, cleanup, and private credential storage.
> - This pull request adds tested building blocks for that flow.
> - The result gives a later Daytona integration a clear security
boundary.
## Linked Issues or Issue Description
No public issue covers this change.
**Problem or motivation**
The Codex local adapter has no safe, reusable flow to prove device login
inside an isolated sandbox.
**Proposed solution**
Add parser, runner, credential export, and proof helpers. Validate the
prompt, protect login data, store credentials in a private run-scoped
home, and dispose all sandbox resources.
**Alternatives considered**
Do not connect a production Daytona driver in this change. Use an
injected sandbox driver and focused tests first. This keeps the security
controls testable before live provider integration.
**Roadmap alignment**
The change extends the Codex local adapter. It does not add a core
Paperclip route or duplicate a planned core feature.
**Additional context**
The flow keeps the login URL, code, and token out of logs, results, and
errors. The proof home uses a company-scoped root and a run-scoped
private directory.
## What Changed
- Add a pure parser for the exact Codex device-login URL and one-time
code shape.
- Add a sandbox runner with prompt handling, timeout, cancellation, and
disposal.
- Add a credential export step with company scoping, path checks,
payload checks, private modes, locking, and cleanup.
- Add redacted device-login fixtures and focused tests for parsing,
secret redaction, runner outcomes, credential export, and cleanup.
## Verification
- Run `pnpm --filter @paperclipai/adapter-codex-local exec vitest run`.
- Run `pnpm --filter @paperclipai/adapter-codex-local exec tsc
--noEmit`.
- Review tests for strict URL and code validation, timeout,
cancellation, disposal, secret redaction, path safety, payload safety,
file modes, and cleanup.
## Risks
- This change provides building blocks, not a live Daytona proof.
- A later integration must connect the runner to a concrete sandbox
driver.
- Credential export depends on existing Codex authentication cache
helpers.
- Incorrect path or payload assumptions can reject valid credentials.
## Model Used
Codex, GPT-5, tool use, code execution, and repository review. The
Paperclip runtime controls the exact context window and reasoning mode.
## Checklist
- [x] I have included a thinking path that traces from project context
to this change
- [x] I have specified the model used (with version and capability
details)
- [x] I have checked ROADMAP.md and confirmed this PR does not duplicate
planned core work
- [x] I have searched GitHub for duplicate or related PRs and linked
them above
- [x] I have either (a) linked existing issues with `Fixes: #` /
`Closes: #` / `Refs #` OR (b) described the issue in-PR following the
relevant issue template
- [x] I have not referenced internal/instance-local Paperclip issues or
links (only public GitHub `#NNN` / `github.com/paperclipai/paperclip`
URLs)
- [x] My branch name describes the change and contains no internal
Paperclip ticket id or instance-derived details
- [x] I have run tests locally and they pass
- [x] I have added or updated tests where applicable
- [x] I have updated relevant documentation to reflect my changes
- [x] I have considered and documented any risks above
- [ ] All Paperclip CI gates are green
- [ ] Greptile is 5/5 with no open P2s, recommendations, or follow-ups
- [x] I will address all Greptile and reviewer comments before
requesting merge
---------
Co-authored-by: Paperclip <noreply@paperclip.ing>
## Thinking Path
> - Paperclip is the open source app people use to manage AI agents for
work
> - The agent Test action builds adapter config from the form.
> - The build-config parser kept plain and secret_ref bindings.
> - It dropped user_secret_ref bindings on the create path.
> - This PR shares one parser that keeps every binding shape.
> - The test path now sends the same env binding set that a real run
sees.
> - The benefit is one fix across every adapter build-config path.
## Linked Issues or Issue Description
**What happened?**
The agent Test action dropped a user-scoped env binding in create mode.
The same agent config worked in a real run. Related public PRs: #10115,
#9321, #9921, #8825.
**Expected behavior**
The Test action should keep user-scoped env bindings and resolve them
like a real run.
**Steps to reproduce**
1. Set a user-scoped env binding on an agent config form.
2. Run Test in create mode.
3. The probe runs without the variable.
**Paperclip version or commit**
c09d2509e3
**Deployment mode**
Local dev (pnpm dev)
**Agent adapter(s) involved**
Not adapter-specific (core bug)
**Database mode**
Embedded PGlite (default — DATABASE_URL unset)
**Additional context**
This change is not Claude-specific.
## What Changed
- Added a shared env binding parser in `@paperclipai/adapter-utils`.
- Replaced the eight adapter build-config copies with the shared helper.
- Kept `plain`, `secret_ref`, and `user_secret_ref` bindings intact in
create mode and edit mode.
- Preserved the runtime merge behavior from the earlier env merge
change.
## Verification
- Author-recorded test run:
`packages/adapter-utils/src/env-bindings.test.ts`
- Author-recorded test run:
`packages/adapters/claude-local/src/ui/build-config.test.ts`
- Author-recorded test run: six adapter build-config test files
- Author-recorded typecheck: `tsc --noEmit` for adapter-utils and the
eight adapter packages
- GitHub checks: all required PR checks pass on PR #10926.
- Greptile review: 5/5 with no open comments.
## Risks
- The change touches adapter config assembly.
- A wrong binding shape would change test-time probe input.
- Tests cover the binding types and the create-mode path.
> For core feature work, check [`ROADMAP.md`](ROADMAP.md) first and
discuss it in `#dev` before opening the PR. Feature PRs that overlap
with planned core work may need to be redirected — check the roadmap
first. See `CONTRIBUTING.md`.
## Model Used
OpenAI GPT-5, code execution and repo inspection.
## Checklist
- [x] I have included a thinking path that traces from project context
to this change
- [x] I have specified the model used (with version and capability
details)
- [x] I have checked ROADMAP.md and confirmed this PR does not duplicate
planned core work
- [x] I have searched GitHub for duplicate or related PRs and linked
them above
- [x] I have either (a) linked existing issues with `Fixes: #` /
`Closes: #` / `Refs: #` OR (b) described the issue in-PR following the
relevant issue template
- [x] I have not referenced internal/instance-local Paperclip issues or
links (only public GitHub `#NNN` / `github.com/paperclipai/paperclip`
URLs)
- [x] My branch name describes the change and contains no internal
Paperclip ticket id or instance-derived details
- [x] I have run tests locally and they pass
- [x] I have added or updated tests where applicable
- [x] I have updated relevant documentation to reflect my changes or
confirmed no documentation update is needed
- [x] I have considered and documented any risks above
- [x] All Paperclip CI gates are green
- [x] Greptile is 5/5 with no open P2s, recommendations, or follow-ups
- [x] I will address all Greptile and reviewer comments before
requesting merge
---------
Co-authored-by: Paperclip <noreply@paperclip.ing>
## Thinking Path
> - Paperclip manages agent work as issues and pull requests.
> - This change lives in the Codex local adapter path.
> - The host auth flow needs a stable cache per identity.
> - The cache must not change the copy-back path or the default store
overwrite.
> - This pull request adds that cache and keeps the existing flow
intact.
> - The benefit is repeatable host auth with safer identity scoping.
## Linked Issues or Issue Description
**Subsystem affected**
- packages/adapters: Codex local adapter and host auth flow
**Problem or motivation**
- The host auth flow needs one usable credential per identity.
- The current flow does not keep that identity state in a separate
cache.
**Proposed solution**
- Add an identity-keyed host credential cache.
- Keep the cache company scoped.
- Add an opt-in seed mode for the merge decision helper.
- Write the cache at copy-back time without changing the default store
overwrite.
**Alternatives considered**
- Store the cache in the instance-global root.
- Seed the host default store through an environment flag.
- Both choices weaken isolation or caller control, so I did not use
them.
**Roadmap alignment**
- I checked `ROADMAP.md`.
- I found no direct overlap with an active roadmap item.
- The change fits the core auth and secrets direction.
**Additional context**
- Local tests passed before I opened this pull request.
- I found no direct duplicate pull request for this branch.
## What Changed
- Added `codex-auth-cache.ts` for company scoped cache storage and
identity anchored vending.
- Added `codex-auth-merge-decision.cjs` support for a leading
`--seed-if-dest-absent` flag.
- Updated `codex-auth-copyback.ts` to write the cache at teardown under
the merge lock.
- Wired `execute.ts` to use the identity anchored vend before the
managed home seed step.
- Added `CODEX-AUTH-CACHE.md` for the cache rules, directions, state
matrix, off switch, and clear action.
- Added and updated tests for cache behavior, merge decision flow, and
copy-back flow.
## Verification
- `pnpm --filter @paperclipai/adapter-codex-local exec vitest run`.
- `node_modules/.bin/vitest run --project @paperclipai/adapter-utils
workspace-restore-merge`.
- `tsc --noEmit` in `packages/adapters/codex-local`.
- `git log --oneline
origin/master..origin/feat/codex-auth-identity-cache` shows the expected
commits.
## Risks
- This change touches credential storage.
- A mistake in the cache path could break host auth for one identity.
- The tests reduce this risk, but the surface stays sensitive.
## Model Used
- OpenAI Codex, GPT-5, tool use enabled.
## Checklist
- [x] I have included a thinking path that traces from project context
to this change
- [x] I have specified the model used with version and capability
details
- [x] I have checked ROADMAP.md and confirmed this PR does not duplicate
planned core work
- [x] I have searched GitHub for duplicate or related PRs and linked
them above
- [x] I have either linked existing issues with `Fixes:` / `Closes:` /
`Refs:` or described the issue in the pull request body
- [x] I have not referenced internal or instance-local Paperclip issues
or links
- [x] My branch name describes the change and contains no internal
Paperclip ticket id or instance-derived details
- [x] I have run tests locally and they pass
- [x] I have added or updated tests where applicable
- [x] I have updated relevant documentation to reflect my changes
- [x] I have considered and documented any risks above
- [x] All Paperclip CI gates are green
- [x] Greptile is 5/5 with no open P2s, recommendations, or follow-ups
- [x] I will address all Greptile and reviewer comments before
requesting merge
---------
Co-authored-by: Paperclip <noreply@paperclip.ing>
## Thinking Path
> - Paperclip is the open source app people use to manage AI agents for
work
> - The codex-local adapter runs Codex agents through the ACP lane and
offers a "Test" button on the agent configuration page to validate the
environment
> - The environment test used its own ad-hoc credential probe, while
real dispatch uses the shared `evaluateCodexCredentialReadiness`
predicate in `codex-home.ts`
> - The two paths disagreed: a user with valid Codex subscription auth
in the shared, server-visible Codex home still saw "No Codex ACP
credentials were detected"
> - The old warning also suggested `codex login` without explaining that
a `/login` in a separate Codex or chat session does not authenticate the
Paperclip server process
> - This pull request makes `testCodexAcpEnvironment` use the same
shared readiness predicate as real dispatch and rewords the warning to
name the server credential boundary
> - The benefit is that the Test button now agrees with what dispatch
will actually do, and the warning tells the user exactly which process
needs the credentials
## Linked Issues or Issue Description
No public GitHub issue exists for this bug. Related credential-handling
work, not duplicates: Refs #10160 (classifies OpenAI invalid-key 401s at
dispatch time) and Refs #9598 (classifies Codex refresh auth failures).
Both cover dispatch-time failures; this PR fixes the pre-dispatch
environment test.
Description follows the bug report template:
**What happened?**
A user configured a Codex ACP agent and had already authenticated Codex
(subscription auth present in the shared Codex home visible to the
Paperclip server). Clicking "Test" on the agent configuration page still
reported:
`warn: No Codex ACP credentials were detected. Hint: Set OPENAI_API_KEY
or run codex login before starting a Codex ACP agent.`
**Expected behavior**
The environment test should detect the same credentials that real agent
dispatch would use. When shared managed Codex auth is available, the
test should pass with an informational check instead of warning. When
credentials really are missing, the warning should explain that the
Paperclip server process is the one that needs them.
**Steps to reproduce**
1. Run the Paperclip server as an OS user whose shared Codex home
contains valid subscription `auth.json` (no `OPENAI_API_KEY` in the
adapter env or server env).
2. Configure an agent with the codex-local adapter using the ACP engine.
3. Click "Test" on the agent configuration page.
4. Observe the `codex_acp_credentials_missing` warning even though
dispatch would succeed.
**Agent adapter(s) involved**
Codex
**Additional context**
The confusion was amplified by the hint: users had run `/login` in a
Codex chat session and assumed the server was authenticated. That login
lives in a different process and home directory, so the server never saw
it.
## What Changed
- `testCodexAcpEnvironment`
(packages/adapters/codex-local/src/server/acp.ts) now calls the shared
`evaluateCodexCredentialReadiness` predicate from `codex-home.ts`
instead of a local ad-hoc `hasCodexNativeCredentials` probe, so the Test
button and real dispatch agree.
- An explicit empty `OPENAI_API_KEY` in the adapter config env no longer
falls through to the server environment key.
- An externally managed `CODEX_HOME` override is now reported as its own
informational check (`codex_acp_external_home_configured`).
- The `codex_acp_credentials_missing` warning now says the credentials
must be visible to the Paperclip server, and the hint explains that a
`/login` in a separate Codex or chat session does not authenticate the
server.
- Removed the now-unused `hasCodexNativeCredentials` helper.
- Added two regression tests: shared managed Codex auth is detected (no
false warning), and the missing-credentials warning carries the new
server-boundary wording.
## Verification
- `pnpm --filter @paperclip/adapter-codex-local test --
src/server/acp.test.ts` — the two new tests cover the shared-home
detection branch and the new warning wording; the existing ACP lane
tests cover the API-key and remote-target branches.
- Manual: with subscription auth in the server-visible shared Codex home
and no `OPENAI_API_KEY`, the agent configuration Test now reports
`codex_acp_native_auth_detected` (info) instead of
`codex_acp_credentials_missing` (warn).
## Risks
- Low risk. The change only affects the environment test path, not
dispatch. The readiness predicate is the same one dispatch already uses,
so drift between the two paths is now structurally prevented.
- Behavioral shift: an explicit empty adapter `OPENAI_API_KEY` no longer
silently falls back to the server env key in the test result. This
matches dispatch behavior and is intentional.
## Model Used
- Implementation authored by OpenAI Codex (gpt-5.5) running through the
Codex ACP lane with tool use.
- PR preparation, rebase onto master, and review fix-up by Anthropic
Claude (Claude Code CLI agent, extended thinking, tool use).
## Checklist
- [x] I have included a thinking path that traces from project context
to this change
- [x] I have specified the model used (with version and capability
details)
- [x] I have checked ROADMAP.md and confirmed this PR does not duplicate
planned core work
- [x] I have searched GitHub for duplicate or related PRs and linked
them above
- [x] I have either (a) linked existing issues with `Fixes: #` / `Closes
#` / `Refs #` OR (b) described the issue in-PR following the relevant
issue template
- [x] I have not referenced internal/instance-local Paperclip issues or
links (only public GitHub `#NNN` / `github.com/paperclipai/paperclip`
URLs)
- [x] My branch name describes the change (e.g. `docs/...`, `fix/...`)
and contains no internal Paperclip ticket id or instance-derived details
- [x] I have run tests locally and they pass
- [x] I have added or updated tests where applicable
- [x] I have updated relevant documentation to reflect my changes
- [x] I have considered and documented any risks above
- [x] All Paperclip CI gates are green
- [x] Greptile is 5/5 with no open P2s, recommendations, or follow-ups
- [x] I will address all Greptile and reviewer comments before
requesting merge
Co-authored-by: Cody <noreply@paperclip.ing>
## Thinking Path
> - Paperclip is the open source app people use to manage AI agents for
work
> - A Codex agent runs `codex exec`, and the adapter assembles its
argument vector from the agent's config plus execution-context options
> - For sandbox execution the adapter injects `--skip-git-repo-check`,
because a headless remote workspace has no git trust prompt to answer
> - The adapter also appends the operator's `extraArgs` verbatim, so an
agent that already lists `--skip-git-repo-check` in its config gets the
flag twice on a sandbox run
> - `codex exec` rejects a repeated `--skip-git-repo-check` and exits
with code 2, which the adapter surfaces as `adapter_failed` before any
work runs
> - This pull request skips the sandbox injection when the operator's
args already carry the flag
> - The benefit is that a common, harmless-looking config no longer
crashes every sandbox run
## Linked Issues or Issue Description
**What happened?**
A `codex_local` agent configured with `extraArgs:
["--skip-git-repo-check"]` fails on every sandbox run:
```
error: the argument '--skip-git-repo-check' cannot be used multiple times
Usage: codex exec [OPTIONS] [PROMPT]
```
The adapter reports `stopReason: "adapter_failed"` (Codex exited with
code 2). The flag appears twice in the argv: once injected by the
adapter for sandbox execution, once from the operator's `extraArgs`.
**Steps to reproduce**
1. Configure a `codex_local` agent with `extraArgs:
["--skip-git-repo-check"]` (or the legacy `args` field).
2. Point it at a sandbox environment.
3. Start a run — `codex exec` aborts immediately on the duplicate flag.
**Expected behavior**
The run launches with a single `--skip-git-repo-check`. An operator
listing the flag the adapter already injects should be a no-op, not a
hard failure.
**Paperclip version**
Current `master`.
**Deployment mode**
Any deployment running Codex agents in sandbox environments.
## What Changed
- `buildCodexExecArgs` no longer pushes the sandbox
`--skip-git-repo-check` when the resolved args (`extraArgs`, or the
legacy `args` fallback) already contain it. The operator's copy stands;
the argv carries the flag exactly once. Non-sandbox runs and configs
without the flag are unchanged.
## Verification
- `cd packages/adapters/codex-local && pnpm vitest run
src/server/codex-args.test.ts` — new cases: `extraArgs` already carrying
the flag (single occurrence), the legacy `args` field carrying it
(single occurrence), and the operator's flag preserved when the sandbox
injection is not requested. Existing "adds --skip-git-repo-check when
requested" case unchanged.
- `cd packages/adapters/codex-local && pnpm vitest run` — full package
suite (218 tests).
- `pnpm run typecheck` in the package.
## Risks
- Low. The change only suppresses a duplicate of a single, idempotent
flag; it never removes an operator-supplied argument and never adds one
that was not already going to be present.
## Model Used
Claude Fable 5 (`claude-fable-5`, Anthropic) via Claude Code — extended
thinking, agentic tool use (file edits, vitest/tsc runs). No other
models involved.
## Checklist
- [x] I have included a thinking path that traces from project context
to this change
- [x] I have specified the model used (with version and capability
details)
- [x] I have checked ROADMAP.md and confirmed this PR does not duplicate
planned core work
- [x] I have searched GitHub for duplicate or related PRs and linked
them above
- [x] I have either (a) linked existing issues with `Fixes: #` / `Closes
#` / `Refs #` OR (b) described the issue in-PR following the relevant
issue template
- [x] I have not referenced internal/instance-local Paperclip issues or
links (only public GitHub `#NNN` / `github.com/paperclipai/paperclip`
URLs)
- [x] My branch name describes the change (e.g. `docs/...`, `fix/...`)
and contains no internal Paperclip ticket id or instance-derived details
- [x] I have run tests locally and they pass
- [x] I have added or updated tests where applicable
- [x] I have updated relevant documentation to reflect my changes
- [x] I have considered and documented any risks above
- [ ] All Paperclip CI gates are green
- [ ] Greptile is 5/5 with no open P2s, recommendations, or follow-ups
- [x] I will address all Greptile and reviewer comments before
requesting merge
## Thinking Path
> - Paperclip is the open source app people use to manage AI agents for
work
> - Codex agents can run inside sandbox environments, and operators can
bake a Codex login into the sandbox image during interactive image setup
> - Two credential gates (the control plane's pre-dispatch
configuration-incomplete gate and the adapter's execute-time fail-fast)
required host-side Codex credentials — a usable `auth.json` in the
managed home or a configured `OPENAI_API_KEY` — regardless of where the
run executes
> - On managed cloud hosts a local Codex login never exists, so every
sandbox run of a Codex agent failed immediately with "configuration
incomplete: no Codex credentials available for managed home …", even
though the adapter's inbound auth merge already supports the image-login
case end to end
> - This pull request makes the execute-time gate probe the sandbox for
its own `~/.codex/auth.json` before failing, and exempts
sandbox-destined runs from the pre-dispatch host check
> - The benefit is that a sandbox image signed in to Codex is a
first-class credential source, matching what the auth-merge,
precedence-warning, and copy-back machinery were already built for
## Linked Issues or Issue Description
**What happened?**
Running a `codex_local` agent in a sandbox environment whose image
carries a Codex login failed instantly with `configuration incomplete:
no Codex credentials available for managed home "…/codex-home". Sign in
to Codex on the host with a ChatGPT subscription, or bind a per-agent
OPENAI_API_KEY secret for this agent.` The host has no Codex login and
never will on a managed cloud deployment; the sandbox's own login was
never consulted.
**Steps to reproduce**
1. Configure a sandbox environment and capture a custom image after
signing in to Codex inside the interactive image setup.
2. Create a `codex_local` agent that uses that environment, on a host
with no Codex login and no `OPENAI_API_KEY` bound.
3. Start a run: it fails pre-dispatch with the configuration-incomplete
blocker above.
**Expected behavior**
The run launches and Codex authenticates with the sandbox image's own
login, the same way the adapter's host↔sandbox auth merge already keeps
the sandbox credential when the host ships none. A run should only fail
fast when neither the host, a bound `OPENAI_API_KEY`, nor the sandbox
has credentials.
**Paperclip version**
Current `master` (cloud image deployments).
**Deployment mode**
Managed cloud stacks (any deployment where the server host has no local
Codex login).
## What Changed
- Extracted the adapter's execute-time gate into
`assertCodexCredentialsLaunchable`: when host readiness fails and the
target is a sandbox, it probes `~/.codex/auth.json` in the sandbox (same
command the auth-precedence warning uses) and proceeds with a log line
naming the credential source; when the sandbox has no login either, the
error now names all three remediation options (sandbox image sign-in,
per-agent `OPENAI_API_KEY`, host sign-in). Non-sandbox targets keep
today's strict behavior byte-for-byte.
- The control plane's pre-dispatch gate in
`resolveExecutionRunAdapterConfig` now takes the selected environment's
driver and skips the host-credential check for sandbox-destined runs —
only the adapter can probe the sandbox once it is up, so the
execute-time gate is the authority there. Non-sandbox runs keep the
early, well-attributed configuration-incomplete blocker.
- The codex Test flow needed no change: it already seeds host
credentials only when they exist and otherwise leaves the sandbox's
`CODEX_HOME` alone; this aligns the run path with it.
## Verification
- `cd packages/adapters/codex-local && pnpm vitest run` — 210 tests,
including new gate cases: sandbox login present (proceeds + logs
source), sandbox and host both credential-less (fails with the extended
message), non-sandbox target (strict host requirement kept, no sandbox
probe), per-agent API key (no probe at all).
- `cd server && pnpm vitest run
src/__tests__/heartbeat-project-env.test.ts
src/__tests__/codex-local-adapter-environment.test.ts` — includes the
new sandbox-exemption case next to the existing blocker tests.
- `pnpm run typecheck` in `server` and `packages/adapters/codex-local`.
## Risks
- Sandbox-destined misconfigurations (no credentials anywhere) now
surface at adapter execute time instead of pre-dispatch, so they read as
an adapter failure with a precise message rather than a
configuration-incomplete blocker. The trade-off is deliberate: the
sandbox must be up to know whether credentials exist, and the failure
message names the exact remediations.
- The sandbox probe adds one short (5s-capped) shell command to sandbox
runs whose host has no credentials; runs with host credentials or a
bound key are untouched.
- Self-hosted behavior is unchanged for local and SSH targets.
## Model Used
Claude Fable 5 (`claude-fable-5`, Anthropic) via Claude Code — extended
thinking, agentic tool use (file edits, vitest/tsc runs). No other
models involved.
## Checklist
- [x] I have included a thinking path that traces from project context
to this change
- [x] I have specified the model used (with version and capability
details)
- [x] I have checked ROADMAP.md and confirmed this PR does not duplicate
planned core work
- [x] I have searched GitHub for duplicate or related PRs and linked
them above
- [x] I have either (a) linked existing issues with `Fixes: #` / `Closes
#` / `Refs #` OR (b) described the issue in-PR following the relevant
issue template
- [x] I have not referenced internal/instance-local Paperclip issues or
links (only public GitHub `#NNN` / `github.com/paperclipai/paperclip`
URLs)
- [x] My branch name describes the change (e.g. `docs/...`, `fix/...`)
and contains no internal Paperclip ticket id or instance-derived details
- [x] I have run tests locally and they pass
- [x] I have added or updated tests where applicable
- [x] I have updated relevant documentation to reflect my changes
- [x] I have considered and documented any risks above
- [ ] All Paperclip CI gates are green
- [ ] Greptile is 5/5 with no open P2s, recommendations, or follow-ups
- [x] I will address all Greptile and reviewer comments before
requesting merge
## Thinking Path
> - Paperclip is the open source control plane people use to manage AI
agents for work
> - The `codex_local` adapter runs OpenAI's Codex CLI through direct CLI
and ACP execution lanes
> - The adapter defaulted to the bare `gpt-5.6` alias while the bundled
ACP Codex version lacked GPT-5.6-family metadata
> - Default and legacy-configured runs therefore emitted
fallback-metadata warnings and could use generic context limits
> - This pull request upgrades the bundled Codex ACP dependency, selects
the concrete `gpt-5.6-sol` model, and normalizes the legacy alias in
both execution lanes
> - The benefit is correct model metadata without hiding genuine stderr
or transcript warnings
## Linked Issues or Issue Description
Related public PRs: Refs #9342, Refs #9352, and Refs #9382. This PR is
narrower: it upgrades bundled Codex metadata and normalizes the legacy
bare alias in both execution lanes.
**Bug report**
### What happened
Default `codex_local` runs, and agents still configured with the bare
`gpt-5.6` model, print a model-metadata fallback warning and use generic
context-window limits.
Root cause: the ACP lane bundled a Codex release predating
GPT-5.6-family metadata, while Paperclip's default and advertised model
used the bare `gpt-5.6` alias for which Codex publishes no metadata.
### Expected behavior
A default Codex run resolves to a concrete model slug with published
metadata and does not emit a fallback-metadata warning.
### Deployment mode
Self-hosted/local `codex_local` adapter.
## What Changed
- Upgraded `@agentclientprotocol/codex-acp` from `^1.1.0` to `^1.1.4`
- Changed `DEFAULT_CODEX_LOCAL_MODEL` from `gpt-5.6` to `gpt-5.6-sol`
- Removed the bare alias from advertised models and listed concrete
GPT-5.6 Fast-mode variants
- Added `normalizeCodexModel()` and applied it in both CLI and ACP
execution lanes
- Updated adapter docs, Storybook fixtures, and regression tests
- Preserved warning visibility; no stderr, transcript, or log filtering
changed
## Verification
- `pnpm --filter @paperclipai/adapter-codex-local typecheck`
- `pnpm check:token-gates`
- `cd packages/adapters/codex-local && pnpm exec vitest run` — 205 tests
passed
- `cd server && pnpm exec vitest run
src/__tests__/adapter-models.test.ts` — 17 tests passed
- Confirmed the PR diff excludes `pnpm-lock.yaml` and
`.github/workflows/**` as required by repository policy
- Confirmed `.github/workflows/pr.yml` regenerates and uploads the PR
lockfile artifact before downstream `pnpm install --frozen-lockfile`
steps
## Risks
Low risk. The behavior change is scoped to `codex_local` model
selection. Existing concrete model IDs pass through unchanged; only the
legacy bare `gpt-5.6` alias is rewritten. Dependency resolution may
select a newer compatible `codex-acp` release within the declared range,
so CI remains the final compatibility gate.
> For core feature work, check [`ROADMAP.md`](ROADMAP.md) first and
discuss it in `#dev` before opening the PR. Feature PRs that overlap
with planned core work may need to be redirected — check the roadmap
first. See `CONTRIBUTING.md`.
## Model Used
- Original implementation: Anthropic Claude Opus 4.8 (`claude-opus-4-8`,
1M context, tool use and code execution)
- Conflict resolution and PR preparation: OpenAI GPT-5.5 (`gpt-5.5`,
Codex CLI coding agent, high-reasoning tool use and code execution;
host-managed context window)
## Checklist
- [x] I have included a thinking path that traces from project context
to this change
- [x] I have specified the model used (with version and capability
details)
- [x] I have checked ROADMAP.md and confirmed this PR does not duplicate
planned core work
- [x] I have searched GitHub for duplicate or related PRs and linked
them above
- [x] I have either (a) linked existing issues with `Fixes: #` / `Closes
#` / `Refs #` OR (b) described the issue in-PR following the relevant
issue template
- [x] I have not referenced internal/instance-local Paperclip issues or
links (only public GitHub `#NNN` / `github.com/paperclipai/paperclip`
URLs)
- [ ] My branch name describes the change (e.g. `docs/...`, `fix/...`)
and contains no internal Paperclip ticket id or instance-derived details
— branch name is fixed by the assigned execution workspace and cannot be
renamed in-place
- [x] I have run tests locally and they pass
- [x] I have added or updated tests where applicable
- [x] I have updated relevant documentation to reflect my changes
- [x] I have considered and documented any risks above
- [x] All Paperclip CI gates are green
- [x] Greptile is 5/5 with no open P2s, recommendations, or follow-ups
- [x] I will address all Greptile and reviewer comments before
requesting merge
---------
Co-authored-by: Paperclip <noreply@paperclip.ing>
## Thinking Path
> - Paperclip is the control plane people use to create and govern
AI-agent companies
> - Agent creation persists runtime configuration that controls which
model profiles future runs may select
> - Adapters can expose a `cheap` profile, and existing creation paths
implicitly left that profile available when operators made no choice
> - That made a newly created agent eligible for a lower-cost model
without an explicit operator opt-in
> - The UI also dropped an explicit opt-in when the operator selected
the adapter's default cheap model rather than a custom model ID
> - Codex additionally hardcoded `gpt-5.3-codex-spark` into its cheap
profile and static fallback model list, making Paperclip choose an
auth-dependent model rather than requiring an operator choice
> - This pull request makes new-agent creation disable an available
cheap profile by default while preserving explicit opt-in from the UI or
API
> - The Codex cheap profile now remains available for explicit
configuration but supplies no model default, so an unconfigured cheap
request stays on the primary model
> - The benefit is predictable model quality for new agents and an
intentional, auditable choice before lower-cost routing is enabled
## Linked Issues or Issue Description
**Problem**
New agents created with an adapter that exposes a `cheap` model profile
can inherit that profile without the operator explicitly enabling it. In
the UI, enabling the adapter-default cheap model is also omitted because
runtime configuration is only written when a custom model ID is present.
**Expected behavior**
- New agents default an available `cheap` model profile to `{ enabled:
false }` when the caller does not specify it.
- Explicit API configuration remains authoritative.
- UI opt-in persists even when the adapter default model is used.
- Codex does not advertise or automatically select
`gpt-5.3-codex-spark`; operators must explicitly configure any
lower-cost Codex model.
**Related public work**
- Refs #4881, which introduced cheap model profiles for local adapters.
- Supersedes the default-selection portions of #8032 and #10004 by
removing the Codex model default instead of replacing it with another
hardcoded model.
## What Changed
- Detect whether the selected adapter exposes a `cheap` model profile
during agent creation and hiring.
- Persist `runtimeConfig.modelProfiles.cheap.enabled = false` only when
the caller did not explicitly configure the profile.
- Preserve UI cheap-profile opt-in when using the adapter's default
model by writing an empty adapter config.
- Remove `gpt-5.3-codex-spark` from the Codex static model list.
- Keep the Codex `cheap` profile explicitly configurable while giving it
an empty adapter config, so Paperclip never chooses a cheap Codex model
automatically.
- Verify that a Codex cheap request without an explicit model leaves the
primary model unchanged.
- Extend server route and UI runtime-config tests for default-disable
and explicit-opt-in behavior.
## Verification
- `env -u PAPERCLIP_IN_WORKTREE -u PAPERCLIP_WORKTREE_NAME -u
PAPERCLIP_CONFIG -u PAPERCLIP_HOME -u PAPERCLIP_INSTANCE_ID -u
PAPERCLIP_CONTEXT pnpm exec vitest run
packages/adapters/codex-local/src/index.test.ts
packages/adapters/codex-local/src/server/codex-args.test.ts
server/src/__tests__/adapter-models.test.ts
server/src/__tests__/adapter-registry.test.ts
server/src/__tests__/heartbeat-model-profile.test.ts
server/src/__tests__/agent-permissions-routes.test.ts
ui/src/lib/new-agent-runtime-config.test.ts`
- Result: 7 test files passed, 105 tests passed.
- GitHub `Typecheck + Release Registry` check passed on the final head.
- `git diff --check public-gh/master...HEAD`
## Risks
- Low behavioral risk: only newly created or hired agents are
normalized; existing agents are unchanged.
- Explicit `cheap` profile settings remain untouched, including explicit
opt-in.
- Codex users who explicitly opt into the cheap lane must choose a
model; requests without a configured override intentionally continue on
the primary model.
- Adapter profile discovery is now awaited during creation, adding a
small amount of adapter metadata lookup work.
- The source branch name is automation-provided and retained as required
by the task, so it does not satisfy the preferred public branch naming
convention.
> For core feature work, check [`ROADMAP.md`](ROADMAP.md) first and
discuss it in `#dev` before opening the PR. Feature PRs that overlap
with planned core work may need to be redirected — check the roadmap
first. See `CONTRIBUTING.md`.
## Model Used
- OpenAI `gpt-5.4` via Codex CLI, with reasoning, repository editing,
terminal execution, and GitHub/Paperclip tool access. The runtime did
not expose a context-window size.
## Checklist
- [x] I have included a thinking path that traces from project context
to this change
- [x] I have specified the model used (with version and capability
details)
- [x] I have checked ROADMAP.md and confirmed this PR does not duplicate
planned core work
- [x] I have searched GitHub for duplicate or related PRs and linked
them above
- [x] I have either (a) linked existing issues with `Fixes: #` / `Closes
#` / `Refs #` OR (b) described the issue in-PR following the relevant
issue template
- [x] I have not referenced internal/instance-local Paperclip issues or
links (only public GitHub `#NNN` / `github.com/paperclipai/paperclip`
URLs)
- [ ] My branch name describes the change (e.g. `docs/...`, `fix/...`)
and contains no internal Paperclip ticket id or instance-derived details
- [x] I have run tests locally and they pass
- [x] I have added or updated tests where applicable
- [x] I have updated relevant documentation to reflect my changes
- [x] I have considered and documented any risks above
- [x] All Paperclip CI gates are green
- [x] Greptile is 5/5 with no open P2s, recommendations, or follow-ups
- [x] I will address all Greptile and reviewer comments before
requesting merge
---------
Co-authored-by: Paperclip <noreply@paperclip.ing>
## Thinking Path
> - Paperclip is the control plane people use to coordinate AI agents
and their execution environments.
> - Environment realization decides where an agent runs and which
filesystem and toolchain are authoritative.
> - Copy-based realization is unsafe for container-anchored tasks
because absolute paths such as `/app` can point outside the synchronized
tree and task-specific binaries may be absent.
> - That mismatch can let an agent successfully verify work in a phantom
writable path while sync-back silently discards the result.
> - Existing task environments already provide the authoritative
filesystem and toolchain, so they should be executed in place rather
than copied.
> - Copy mode still needs explicit confinement rules so aliases target
the synchronized workspace and unsynchronized writable paths fail
visibly.
> - This pull request adds typed realization metadata, propagates the
authoritative root through orchestration, and teaches Codex to honor it.
> - The benefit is that container-anchored tasks operate on
verifier-visible state with the intended tools, while copy mode remains
safe and backward compatible.
## Linked Issues or Issue Description
No public GitHub issue exists for this defect.
GitHub duplicate searches for in-place execution, workspace realization,
and authoritative workspace roots found no related pull request to link.
### What happened?
Environment-backed agent runs were always realized through a copied
workspace. Tasks anchored to absolute container paths could therefore
write outside the synchronized tree, and task-provided toolchains were
unavailable in the copy. A run could report success even though
sync-back discarded its output.
### Expected behavior
Existing task environments should run against their real authoritative
root and toolchain. Copy-mode runs should map declared absolute aliases
into the synchronized tree and reject writable paths that cannot be
restored.
### Steps to reproduce
1. Run a Codex task environment whose required files live under `/app`
or `/workspace` and whose required binary exists only in the task
container.
2. Observe that copy realization changes the effective
filesystem/toolchain or permits writes outside the synchronized root.
3. Complete and verify the task inside the agent sandbox.
4. Observe that the verifier cannot see out-of-tree artifacts or that
task-specific commands were unavailable.
### Reproduction context
- Paperclip commit: `3a16b91217483d2c233926de5b7f7bc3a1077924`
- Deployment: built from source in a task-container execution
environment
- Adapter: Codex local
- Database: not database-related
- Access context: agent execution
## What Changed
- Added typed `copy | in_place` workspace-realization metadata,
authoritative roots, confined aliases, and outbound restore paths to
shared execution-target contracts.
- Selected in-place realization for existing task environments and
skipped archive prepare/restore when the authoritative environment is
used directly.
- Propagated the authoritative root into adapter context so Codex uses
it for cwd and `PAPERCLIP_WORKSPACE_*` semantics, including ACP
execution.
- Bound copy-mode aliases such as `/app` to the synchronized workspace
and rejected writable out-of-tree paths without explicit restore
mappings.
- Added focused regression coverage while preserving existing copy-mode
archive restore behavior.
## Verification
- `pnpm --filter @paperclipai/shared typecheck` — passed.
- `pnpm exec vitest run
packages/adapter-utils/src/local-process-sandbox.test.ts
packages/adapters/codex-local/src/server/acp.test.ts
packages/adapters/codex-local/src/server/execute.remote.test.ts
server/src/__tests__/environment-run-orchestrator.test.ts` — 48 passed,
4 skipped.
- `pnpm -r typecheck` — passed.
- `env -u AWS_ACCESS_KEY_ID -u AWS_SECRET_ACCESS_KEY -u
AWS_SESSION_TOKEN pnpm test:run` — passed across all general and
serialized Vitest shards.
- `pnpm build` — passed.
- Codex `k=1` acceptance run completed July 24, 2026 at 23:54:30 UTC
with 4 completed, 0 exceptions, and mean reward 1.0: `build-cython-ext`,
`openssl-selfsigned-cert`, `prove-plus-comm`, and `sqlite-db-truncate`
each received terminal grade 1.0 against real task-environment paths and
toolchains.
## Risks
- In-place mode deliberately exposes the authoritative task root to the
adapter; incorrect environment metadata could point execution at the
wrong root. Typed metadata and focused orchestration tests cover
selection and propagation.
- Copy-mode writable-path validation is stricter and may reject
previously accepted unsafe configurations. The rejection is intentional
and produces a visible error instead of silently losing output.
- The acceptance run is focused on four Codex task-environment
workloads, not a broad cross-adapter benchmark. Existing copy-mode
archive tests and the full repository suite remain green.
> For core feature work, check [`ROADMAP.md`](ROADMAP.md) first and
discuss it in `#dev` before opening the PR. Feature PRs that overlap
with planned core work may need to be redirected — check the roadmap
first. See `CONTRIBUTING.md`.
## Model Used
- OpenAI Codex CLI coding agent; exact model ID and context-window size
were not exposed to this runtime. Capabilities used: extended reasoning,
repository editing, shell execution, test/build execution, Git, GitHub
CLI, and Paperclip API tool use.
## Checklist
- [x] I have included a thinking path that traces from project context
to this change
- [x] I have specified the model used (with version and capability
details)
- [x] I have checked ROADMAP.md and confirmed this PR does not duplicate
planned core work
- [x] I have searched GitHub for duplicate or related PRs and linked
them above
- [x] I have either (a) linked existing issues with `Fixes: #` / `Closes
#` / `Refs #` OR (b) described the issue in-PR following the relevant
issue template
- [x] I have not referenced internal/instance-local Paperclip issues or
links (only public GitHub `#NNN` / `github.com/paperclipai/paperclip`
URLs)
- [x] My branch name describes the change (e.g. `docs/...`, `fix/...`)
and contains no internal Paperclip ticket id or instance-derived details
- [x] I have run tests locally and they pass
- [x] I have added or updated tests where applicable
- [x] I have updated relevant documentation to reflect my changes
- [x] I have considered and documented any risks above
- [x] All Paperclip CI gates are green
- [x] Greptile is 5/5 with no open P2s, recommendations, or follow-ups
- [x] I will address all Greptile and reviewer comments before
requesting merge
---------
Co-authored-by: Paperclip <noreply@paperclip.ing>
## Thinking Path
> - Paperclip is the open source control plane people use to manage AI
agents and their work
> - Heartbeat execution relies on adapters distinguishing agent failures
from failures in the harness running beneath the agent
> - Codex MCP transport crashes can kill the CLI after the JSONL
protocol has started but before it emits a protocol-terminal event
> - Those interrupted streams were left unclassified, so the control
plane terminalized the heartbeat as `heartbeat_failed` / `agent_failure`
with no continuation
> - Agent-level failure is already expressible through the JSONL
protocol via an `error` event, `turn.failed`, or `turn.completed`, so an
interrupted nonzero exit can be classified structurally without
inspecting unstable error strings
> - This pull request reports that shape as `codex_harness_crash` in the
`transient_upstream` family and routes it through Paperclip's existing
bounded retry and recovery-continuation paths
> - The benefit is that transient Codex harness failures recover safely
without misclassifying quoted agent output or depending on
transport-specific wording
## Linked Issues or Issue Description
- **What happened:** Codex MCP transport failures, including rmcp worker
death, could terminate the CLI mid-turn after protocol output began but
before any terminal JSONL event. The run then became an unclassified
terminal heartbeat failure with `continuationCount: 0`; this occurred in
3 of 44 L3 Codex-lane trials during the associated benchmark
investigation.
- **Expected behavior:** a nonzero Codex exit after the protocol starts
but before an `error`, `turn.failed`, or `turn.completed` event should
be treated as a harness/infrastructure crash and enter the existing
bounded retry policy.
- **Why structural classification:** transport error strings vary, and
stdout may quote agent output that merely discusses network failures.
The protocol boundary identifies whether the agent itself produced a
terminal result without regex matching.
- **Recovery behavior:** `codex_harness_crash` maps to `errorFamily:
transient_upstream`, using the existing `same_session` →
`safer_invocation` → `fresh_session` ladder plus the
recovery-continuation transient-infrastructure path.
- Supersedes the regex-based approach in #10150, which is closed.
## What Changed
- Added protocol-state tracking that identifies a nonzero exit after
protocol start and before any protocol-terminal event as
`codex_harness_crash`.
- Propagated the structural classification as `transient_upstream`
through the Codex adapter.
- Added parse unit coverage, including a faithful crash-shaped stream,
without matching stderr transport strings.
- Added adapter execution coverage using a fake Codex process that emits
a protocol prefix and then dies with the observed rmcp stderr line.
- Added heartbeat bounded-retry coverage, including the `errorCode`-only
fallback, and recovery-continuation classification coverage.
## Verification
- `parse.test.ts` — 16 passed.
- `codex-local-execute.test.ts` — 16 passed.
- `heartbeat-retry-scheduling.test.ts` — 30 passed.
- `service.pause-durability.test.ts` — 6 passed.
- Server and Codex adapter TypeScript checks passed.
- The branch commit is unchanged from the tested and pushed `88f5464d40`
handoff.
## Risks
- Low risk: the classification requires a nonzero exit after protocol
start and before any protocol-terminal event, so normal agent-declared
failures and completed turns keep their existing behavior.
- The change intentionally broadens recovery for structurally
interrupted Codex runs; bounded retry limits still prevent indefinite
continuation loops.
- No schema, migration, public API, UI, lockfile, or workflow changes.
> For core feature work, check [`ROADMAP.md`](ROADMAP.md) first and
discuss it in `#dev` before opening the PR. Feature PRs that overlap
with planned core work may need to be redirected — check the roadmap
first. See `CONTRIBUTING.md`.
## Model Used
- OpenAI Codex coding agent. The exact runtime model ID and
context-window size were not exposed by the execution environment;
capabilities used for the implementation included repository analysis,
reasoning, code editing, and terminal-based test execution.
## Checklist
- [x] I have included a thinking path that traces from project context
to this change
- [x] I have specified the model used (with version and capability
details)
- [x] I have checked ROADMAP.md and confirmed this PR does not duplicate
planned core work
- [x] I have searched GitHub for duplicate or related PRs and linked
them above
- [x] I have either (a) linked existing issues with `Fixes: #` / `Closes
#` / `Refs #` OR (b) described the issue in-PR following the relevant
issue template
- [x] I have not referenced internal/instance-local Paperclip issues or
links (only public GitHub `#NNN` / `github.com/paperclipai/paperclip`
URLs)
- [ ] My branch name describes the change (e.g. `docs/...`, `fix/...`)
and contains no internal Paperclip ticket id or instance-derived details
— the pre-existing, already-pushed branch name was explicitly prescribed
for this replacement PR
- [x] I have run tests locally and they pass
- [x] I have added or updated tests where applicable
- [x] I have updated relevant documentation to reflect my changes
- [x] I have considered and documented any risks above
- [x] All Paperclip CI gates are green
- [x] Greptile is 5/5 with no open P2s, recommendations, or follow-ups
- [x] I will address all Greptile and reviewer comments before
requesting merge
Co-authored-by: Paperclip <noreply@paperclip.ing>
## Thinking Path
> - Paperclip is the open source control plane people use to coordinate
AI agents and their work.
> - Local adapters are responsible for observing agent subprocesses and
terminating runs that are genuinely stuck.
> - The Codex local adapter currently treats output bytes as its only
liveness signal for the 30-minute inactivity monitor.
> - Healthy compile and test loops can consume CPU and perform disk I/O
for longer than that without producing terminal output.
> - Terminating those runs loses valid work, while removing the monitor
entirely would allow truly wedged processes to run indefinitely.
> - This pull request adds a Linux process-group activity probe that
recognizes meaningful CPU, disk I/O, and child-process churn while
retaining the existing timeout for idle processes.
> - The benefit is that long silent builds can finish without weakening
the adapter's hung-run safety net.
## Linked Issues or Issue Description
No public GitHub issue was found in the duplicate/related search.
**What happened**
A healthy Codex local run executing a long compile/test loop could be
terminated at the default 30-minute output-inactivity threshold when the
child process emitted no stdout or stderr.
**Expected behavior**
The inactivity monitor should keep a silent run alive while its process
group is doing meaningful work, but should still terminate a process
group that is alive and idle.
**Steps to reproduce**
1. Run Codex local with the default `outputInactivityTimeoutMs`.
2. Have the agent start a compile or test command that consumes CPU or
disk I/O without terminal output for longer than the threshold.
3. Observe the adapter terminate the otherwise healthy process group as
output-inactive.
**Affected version / deployment mode**
Observed on a local-process Paperclip deployment using the Codex local
adapter with the 30-minute default inactivity monitor.
## What Changed
- Added a Linux `/proc` process-group sampler that tracks meaningful CPU
tick growth, disk I/O growth, and child-process membership changes.
- Reset the existing Codex inactivity timer when that sampler observes
real process work, while leaving remote and non-Linux behavior
unchanged.
- Added diagnostics for the number of process-activity resets and
documented the expanded liveness semantics.
- Added unit coverage for process-activity timer resets and subprocess
regressions for both a long silent CPU build and a genuinely wedged
child.
## Verification
- `pnpm --filter @paperclipai/adapter-codex-local typecheck`
- `pnpm exec vitest run
packages/adapters/codex-local/src/server/process-activity-monitor.test.ts
packages/adapters/codex-local/src/server/output-inactivity-monitor.test.ts
packages/adapters/codex-local/src/server/output-inactivity-monitor.integration.test.ts`
- Focused result: 23 tests passed, including a silent CPU-bound
subprocess that runs four times beyond the simulated inactivity window
and an idle subprocess that is still terminated.
## Risks
- Low risk and Linux-scoped: the new probe reads `/proc` every 15
seconds only while a monitored local Codex child is running.
- The CPU threshold requires sustained work rather than any single
scheduler tick, reducing the risk that a nearly idle event loop is
treated as productive.
- If `/proc` sampling is unavailable or fails, the adapter falls back to
the existing output-only behavior.
> For core feature work, check [`ROADMAP.md`](ROADMAP.md) first and
discuss it in `#dev` before opening the PR. Feature PRs that overlap
with planned core work may need to be redirected — check the roadmap
first. See `CONTRIBUTING.md`.
## Model Used
- OpenAI Codex with model `gpt-5.6-sol`, high reasoning effort,
terminal/tool execution, code editing, and test execution. The runtime
did not expose a context-window size.
## Checklist
- [x] I have included a thinking path that traces from project context
to this change
- [x] I have specified the model used (with version and capability
details)
- [x] I have checked ROADMAP.md and confirmed this PR does not duplicate
planned core work
- [x] I have searched GitHub for duplicate or related PRs and linked
them above
- [x] I have either (a) linked existing issues with `Fixes: #` / `Closes
#` / `Refs #` OR (b) described the issue in-PR following the relevant
issue template
- [x] I have not referenced internal/instance-local Paperclip issues or
links (only public GitHub `#NNN` / `github.com/paperclipai/paperclip`
URLs)
- [x] My branch name describes the change (e.g. `docs/...`, `fix/...`)
and contains no internal Paperclip ticket id or instance-derived details
- [x] I have run tests locally and they pass
- [x] I have added or updated tests where applicable
- [x] I have updated relevant documentation to reflect my changes
- [x] I have considered and documented any risks above
- [x] All Paperclip CI gates are green
- [x] Greptile is 5/5 with no open P2s, recommendations, or follow-ups
- [x] I will address all Greptile and reviewer comments before
requesting merge
---------
Co-authored-by: Paperclip <noreply@paperclip.ing>
## Thinking Path
> - Paperclip is the open source control plane people use to manage AI
agents for work
> - Local coding adapters can be confined with filesystem and network
sandbox policies
> - Codex confinement proxied only user-allowlisted hosts, so it denied
Paperclip's own run API and managed MCP endpoints
> - The same proxy returned untyped plaintext denials, which MCP clients
could treat as a fatal unexpected content type
> - Host-only `PAPERCLIPAI_CMD` values could also leak into agents even
when the referenced checkout module did not exist
> - This pull request gives the sandbox an explicit trusted-URL channel
for Paperclip endpoints, returns structured JSON errors, and removes the
inherited host CLI pointer
> - The benefit is confined Codex agents retain control-plane access
without broadening the operator's external network allowlist or crashing
on policy denials
## Linked Issues or Issue Description
Refs #3802
Related foundation: #9504
### What happened?
A `codex_local` agent configured with `networkScope: "allowlist"` and an
external-only allowlist could not reach its own Paperclip API or
Paperclip-managed MCP endpoints. The sandbox proxy returned a `403`
plaintext response without `Content-Type`, and an inherited
`PAPERCLIPAI_CMD` could point at a missing checkout-local CLI module.
### Expected behavior
Paperclip's run-scoped API and managed MCP endpoints remain reachable
regardless of the user external allowlist. Policy denials are valid
structured JSON responses with an explicit media type, and host-only CLI
pointers are not inherited by agent processes.
### Steps to reproduce
1. Configure a `codex_local` agent with `networkScope: "allowlist"` and
`networkAllowlist: ["api.openai.com"]`.
2. Run the agent and request its Paperclip issue API or a
Paperclip-managed MCP endpoint.
3. Observe the sandbox proxy deny the request with an untyped plaintext
`403` response.
### Environment
- Paperclip commit: `f49a3f99` originally exhibited the defect; fix is
based on current `master`.
- Deployment: local source build on Linux.
- Adapter: Codex.
- Database: not database-related.
- Access context: agent bearer/run-scoped credentials.
## What Changed
- Added internal trusted URL rules to the local network allowlist proxy
and supplied the Codex run API plus managed MCP endpoints.
- Returned JSON error envelopes with `Content-Type` and `Content-Length`
for HTTP and CONNECT policy denials.
- Removed inherited `PAPERCLIPAI_CMD` from child process environments
while preserving explicitly constructed runtime variables.
- Added focused proxy and environment sanitizer regression tests.
## Verification
- `pnpm exec vitest run
packages/adapter-utils/src/local-process-sandbox.test.ts
packages/adapter-utils/src/server-utils-env.test.ts --reporter=verbose`
- 2 test files passed; 8 tests passed; 4 platform-dependent tests
skipped.
- `git diff --check`
- Package typecheck was attempted; it reaches unrelated current-`master`
type drift in untouched files (`spawnCwd` in `adapter-utils`, and
staged-runtime ACP types in `codex-local`).
## Risks
- Low risk: trusted access is restricted to exact HTTP(S) hostname and
port pairs derived from Paperclip-provided URLs.
- Invalid or non-HTTP trusted URL values are ignored rather than
broadening access.
- Denial response bodies change from plaintext to structured JSON;
status codes remain unchanged.
> For core feature work, check [`ROADMAP.md`](ROADMAP.md) first and
discuss it in `#dev` before opening the PR. Feature PRs that overlap
with planned core work may need to be redirected — check the roadmap
first. See `CONTRIBUTING.md`.
## Model Used
- OpenAI Codex coding agent (exact model ID and context window are not
exposed to this runtime), reasoning and terminal/tool execution enabled.
## Checklist
- [x] I have included a thinking path that traces from project context
to this change
- [x] I have specified the model used (with version and capability
details)
- [x] I have checked ROADMAP.md and confirmed this PR does not duplicate
planned core work
- [x] I have searched GitHub for duplicate or related PRs and linked
them above
- [x] I have either (a) linked existing issues with `Fixes: #` / `Closes
#` / `Refs #` OR (b) described the issue in-PR following the relevant
issue template
- [x] I have not referenced internal/instance-local Paperclip issues or
links (only public GitHub `#NNN` / `github.com/paperclipai/paperclip`
URLs)
- [x] My branch name describes the change (e.g. `docs/...`, `fix/...`)
and contains no internal Paperclip ticket id or instance-derived details
- [x] I have run tests locally and they pass
- [x] I have added or updated tests where applicable
- [x] I have updated relevant documentation to reflect my changes
- [x] I have considered and documented any risks above
- [x] All Paperclip CI gates are green
- [x] Greptile is 5/5 with no open P2s, recommendations, or follow-ups
- [x] I will address all Greptile and reviewer comments before
requesting merge
---------
Co-authored-by: Paperclip <noreply@paperclip.ing>
## Thinking Path
> - Paperclip coordinates autonomous agent work across isolated
company-scoped sessions
> - The ACP remote lane stages workspaces and managed home state inside
the sandbox so sessions can resume safely
> - If a compatible resume re-staged everything every time, it would
waste work and risk inconsistent session reuse behavior
> - If an incompatible resume reused the wrong staged runtime, it could
cross session boundaries or leak credentials
> - This pull request keeps the session fingerprint as the scoping key
and adds a staged-runtime cache keyed to that fingerprint
> - Compatible resumes now reuse the already staged runtime while
incompatible fingerprints stage fresh
> - The benefit is faster safe resumes without weakening session
isolation or credential separation
## Linked Issues or Issue Description
### Problem or motivation
The ACP remote lane currently needs to preserve safe resume behavior
without repeatedly re-staging work that is already valid for the same
session. The failure mode to avoid is letting one session reuse another
session's staged workspace or credentials.
### Proposed solution
Keep the session fingerprint as the scoping key and add a staged-runtime
cache keyed to that fingerprint. When the fingerprint matches, reuse the
already staged workspace and managed home. When the fingerprint changes,
stage fresh.
### Alternatives considered
- Always restage on resume: safest mechanically, but wastes work and
breaks the compatible-resume optimization.
- Reuse without fingerprint scoping: too risky because it could cross
session boundaries.
### Roadmap alignment
This is a narrow implementation change for the ACP remote lane and does
not duplicate any broader roadmap item I could find in `ROADMAP.md`.
### Additional context
The change preserves the existing session fingerprint contents and codex
auth copy-back cadence while adding tests for compatible reuse,
incompatible fresh staging, no cross-session credential reuse, and
failed-turn eviction.
## What Changed
- Added a staged-runtime cache in the ACP remote execution path keyed by
the session fingerprint.
- Reused the existing staged workspace and managed home for compatible
resumes.
- Kept incompatible fingerprints on the fresh staging path.
- Preserved the existing session fingerprint contents and codex auth
copy-back cadence.
- Added tests for compatible reuse, incompatible fresh staging, no
cross-session credential reuse, and failed-turn eviction.
## Verification
- `pnpm exec vitest run
packages/adapter-utils/src/acpx-engine/execute.test.ts` (74/74 pass,
including the active-turn lease regression)
- `pnpm exec tsc -p packages/adapter-utils/tsconfig.json --noEmit`
- Verified the PR touches only
`packages/adapter-utils/src/acpx-engine/execute.ts` and
`packages/adapter-utils/src/acpx-engine/execute.test.ts`
## Risks
- A cache eviction bug could cause an unavailable or partially staged
runtime to be reused.
- If the fingerprint scoping regressed, a session could incorrectly
reuse another session's state.
- The change is isolated to the ACP remote lane, but it still affects
resume behavior for that path.
## Model Used
OpenAI Codex, GPT-5, reasoning-capable coding agent, tool-enabled
session.
## Checklist
- [x] I have included a thinking path that traces from project context
to this change
- [x] I have specified the model used (with version and capability
details)
- [x] I have checked ROADMAP.md and confirmed this PR does not duplicate
planned core work
- [x] I have searched GitHub for duplicate or related PRs and linked
them above
- [x] I have either (a) linked existing issues with `Fixes: #` / `Closes
#` / `Refs #` OR (b) described the issue in-PR following the relevant
issue template
- [x] I have not referenced internal/instance-local Paperclip issues or
links (only public GitHub `#NNN` / `github.com/paperclipai/paperclip`
URLs)
- [x] My branch name describes the change (e.g. `docs/...`, `fix/...`)
and contains no internal Paperclip ticket id or instance-derived details
- [x] I have run tests locally and they pass
- [x] I have added or updated tests where applicable
- [x] I have updated relevant documentation to reflect my changes
- [ ] All Paperclip CI gates are green
- [x] Greptile is 5/5 with no open P2s, recommendations, or follow-ups
- [x] I will address all Greptile and reviewer comments before
requesting merge
---------
Co-authored-by: Paperclip <noreply@paperclip.ing>
## Thinking Path
> - Paperclip is a control plane that orchestrates AI agents and adapter
execution for human operators.
> - Agents run across local and remote execution contexts, and
reliability in remote sessions depends on consistent adapter
bootstrapping.
> - The ACP path must prepare per-adapter runtime homes so managed
credentials and config are available in sandboxed runner environments.
> - Before this change, the new remote ACP lane did not yet consistently
stage managed-home paths for all affected adapters or restore Codex auth
state on teardown.
> - We added a shared per-adapter seam in the ACP engine, then wired
Codex, Claude, and Gemini remote lanes to seed managed homes and remap
to in-sandbox locations.
> - Codex additionally reuses the existing atomic auth restore flow to
copy auth back on teardown, matching CLI behavior.
> - This improves remote runner parity with existing CLI behavior and
avoids credential drift in shared-code-path executions.
## Linked Issues or Issue Description
- This change continues the ACP remote managed-home work by completing
per-adapter remote bootstrapping and Codex auth restore behavior for the
remote ACP lane.
- It specifically covers: `acpx-engine`, `codex-local`, `claude-local`,
and `gemini-local`.
- Related prior work in this repo: PR #10070.
## What Changed
- Add a per-adapter remote managed-home seam
(`prepareRemoteManagedHome`) in `acpx-engine` and thread it through ACP
execution options.
- Implement Codex ACP preparation to:
- stage `CODEX_HOME` (auth/config/skills) into a sandboxed remote home,
- repoint `CODEX_HOME` to the in-sandbox path,
- and wire teardown copy-back through the existing managed-auth restore
path.
- Implement Claude ACP preparation to seed a sanitized `config-seed`
into `CLAUDE_CONFIG_DIR` and remap that directory to the sandbox root.
- Implement Gemini ACP preparation to seed `~/.gemini/skills`, set
`HOME` to managed runtime root, and preselect API-key auth in
`settings.json`.
- Keep local and runner-less ACP→CLI behavior unchanged by only invoking
the remote managed-home seam when running in remote ACP mode.
- Preserve existing authorization and activity boundaries in the shared
engine and adapter layers.
## Verification
- `git log --oneline
origin/master..origin/feat/acp-remote-managed-home-seed` confirms only
the expected 4 commits.
- `tsc --noEmit` is clean in `adapter-utils`, `codex-local`,
`claude-local`, and `gemini-local`.
- Vitest selection used during validation passed (120 tests across the
ACP-related suites).
## Risks
- If remote sandbox teardown occurs after token rotation but before
restore timing, Codex credentials can become stale and require re-auth
on next startup.
- Partial provisioning of managed-home assets would cause adapter
bootstrap failures in runner-backed ACP sessions.
- This change is scoped to execution-path behavior; it should not affect
CLI behavior.
## Model Used
None — human-authored.
## Checklist
- [x] I have included a thinking path that traces from project context
to this change
- [x] I have specified the model used (with version and capability
details)
- [x] I have checked ROADMAP.md and confirmed this PR does not duplicate
planned core work
- [x] I have searched GitHub for duplicate or related PRs and linked
them above
- [x] I have either (a) linked existing issues with `Fixes: #` / `Closes
#` / `Refs #` OR (b) described the issue in-PR following the relevant
issue template
- [x] I have not referenced internal/instance-local Paperclip issues or
links (only public GitHub `#NNN` / `github.com/paperclipai/paperclip`
URLs)
- [x] My branch name describes the change (e.g. `docs/...`, `fix/...`)
and contains no internal Paperclip ticket id or instance-derived details
- [x] I have run tests locally and they pass
- [x] I have added or updated tests where applicable
- [x] I have updated relevant documentation to reflect my changes
- [x] I have considered and documented any risks above
- [x] All Paperclip CI gates are green
- [x] Greptile is 5/5 with no open P2s, recommendations, or follow-ups
- [x] I will address all Greptile and reviewer comments before
requesting merge
---------
Co-authored-by: Harold Kim <harold@paperclip.ing>
Co-authored-by: Paperclip <noreply@paperclip.ing>
## Thinking Path
> - Paperclip is the open source app people use to manage AI agents for
work
> - The shared ACP engine
(`packages/adapter-utils/src/acpx-engine/execute.ts`) is responsible for
launching local and remote agent processes via the ACP protocol
> - On runner-backed remote sandbox (Daytona) targets, `buildRuntime`
never crossed the CLI's staging seam: it never called
`prepareAdapterExecutionTargetRuntime`, left `runtimeRootDir: null` in
both the paperclip and process-session bridges, and handed the agent the
**HOST filesystem path** as the `session/new` cwd — meaning
Claude/Gemini silently operated on a path that does not exist inside the
sandbox (Codex additionally crashes on its HOST home path, addressed in
a follow-up PR)
> - The fix must cross the staging seam for remote sandboxes, thread the
real `runtimeRootDir` through both bridges, and bind the in-sandbox
workspace path as the session cwd — without touching local ACP runs or
the runner-less ACP→CLI fallback
> - This pull request introduces a `stageAcpRemoteRuntime` helper that
calls `prepareAdapterExecutionTargetRuntime` for runner-backed remote
runs, captures `{ workspaceRemoteDir, runtimeRootDir, assetDirs,
restoreWorkspace }`, and reuses the in-sandbox `workspaceRemoteDir` as
the single `sessionCwd` across `session/new`, the fingerprint,
compatibility check, persistence, `ensureSession`, the process-session
bridge cwd, and the error path
> - The benefit is that remote ACP runs now operate in the correct
in-sandbox cwd and receive a non-null `runtimeRootDir` in both bridges —
fixing silent wrong-cwd degradation for Claude/Gemini on Daytona
targets; this is PR 1 of 3 and seeds no credential material
## Linked Issues or Issue Description
No public GitHub issue exists for this change. Inline description
follows the feature request template:
### Subsystem affected
packages/adapters — agent adapter implementations
### Problem or motivation
The shared ACP engine
(`packages/adapter-utils/src/acpx-engine/execute.ts`) never crossed the
CLI's staging seam on runner-backed remote sandbox targets. It never
called `prepareAdapterExecutionTargetRuntime`, always passed
`runtimeRootDir: null` to both the paperclip and process-session
bridges, and handed the agent the **HOST filesystem path** as the ACP
`session/new` cwd. As a result, Claude and Gemini silently operated on a
cwd that does not exist inside the sandbox; Codex crashed with a fatal
error on the HOST `CODEX_HOME` path (that crash is in a follow-up PR).
### Proposed solution
Gate on `usesRunnerBackedSandbox` (`kind === "remote" && transport ===
"sandbox" && runner`). For runs that pass the gate, call
`prepareAdapterExecutionTargetRuntime` via a new `stageAcpRemoteRuntime`
helper that ships the workspace into the sandbox and captures `{
workspaceRemoteDir, runtimeRootDir, assetDirs, restoreWorkspace }`.
Thread the real `runtimeRootDir` into both bridges. Bind a single
`sessionCwd` (the in-sandbox `workspaceRemoteDir`) and use it at every
cwd-keyed session site (`session/new`, fingerprint, compat, persist,
`ensureSession`, process-session bridge, error path) so a warm/resumable
session is reused rather than invalidated. For local runs and the
runner-less ACP→CLI fallback, `sessionCwd` resolves to the HOST cwd —
byte-identical to the previous behavior.
### Alternatives considered
Patching each per-adapter bridge individually — rejected because the bug
is in the shared engine layer and the fix belongs there so all three
adapters (Codex, Claude, Gemini) benefit without per-adapter
duplication.
### Roadmap alignment
Internal correctness fix enabling remote ACP to work as designed; no new
user-facing features. This is PR 1 of 3 in a sequential chain: PR 1
(this PR) stages the workspace and routes the cwd; PR 2 adds per-adapter
home seeding and copy-back; PR 3 wires session-lifecycle restore.
## What Changed
- **`packages/adapter-utils/src/acpx-engine/execute.ts`** — added
`stageAcpRemoteRuntime` helper that calls
`prepareAdapterExecutionTargetRuntime` for runner-backed remote
sandboxes and returns `{ sessionCwd, runtimeRootDir, stagedRuntime }`;
`buildRuntime` now uses this helper to derive `sessionCwd` (in-sandbox
`workspaceRemoteDir` for remote; HOST cwd unchanged for
local/runner-less) and threads the real `runtimeRootDir` to both the
paperclip bridge and process-session bridge; `stagedRuntime` is stashed
for the follow-up credential PR
- **`packages/adapter-utils/src/acpx-engine/execute.test.ts`** — new
engine-level unit tests: staging seam crossed with no credential asset,
non-null `runtimeRootDir` in both bridges, in-sandbox `session/new` cwd,
warm-handle reuse after the cwd change, local-unchanged; 60 tests green
- **`packages/adapters/codex-local/src/server/acp.test.ts`** — new
per-adapter test: runner-backed remote asserts `ensureSession` cwd ==
`workspaceRemoteDir`; runner-less sandbox falls back to CLI
- **`packages/adapters/claude-local/src/server/acp.test.ts`** — same
per-adapter coverage for Claude
- **`packages/adapters/gemini-local/src/server/acp.test.ts`** — same
per-adapter coverage for Gemini
## Verification
- `adapter-utils` typecheck clean; `codex/claude/gemini-local` typecheck
clean
- Engine units (`acpx-engine/execute.test.ts`): 60/60 green — staging
seam crossed with no credential asset, non-null `runtimeRootDir` to both
bridges, in-sandbox `session/new` cwd, warm-handle reuse after cwd
change, local unchanged
- Per-adapter ACP test suites (`codex/claude/gemini-local`
`acp.test.ts`): 103 tests green — runner-backed remote asserts
`ensureSession` cwd == `workspaceRemoteDir`; runner-less sandbox falls
back to CLI
- CI green on PR (in progress)
## Risks
This is PR 1 of 3 in a strictly sequential chain; it seeds **no
credential material** (no `assets`, no `installCommand`). The
per-adapter home seeding is deferred to PR 2, which consumes the
`stagedRuntime` object stashed here. The `restoreWorkspace` callback is
carried on `stagedRuntime` for PR 3's session-lifecycle wiring (see the
`stageAcpRemoteRuntime` function comment). Local ACP runs and the
runner-less ACP→CLI fallback are untouched — `sessionCwd` resolves to
the HOST cwd for those paths, preserving existing behavior. The
`stageAcpRemoteRuntime` helper is gated on `usesRunnerBackedSandbox`, so
there is no regression risk for local or CLI-lane runs.
## Model Used
Claude Sonnet 4.6 (`claude-sonnet-4-6`) via Paperclip ACPX engine —
extended context, tool use enabled, co-authored with Paperclip agent
orchestration.
## Checklist
- [x] I have included a thinking path that traces from project context
to this change
- [x] I have specified the model used (with version and capability
details)
- [x] I have checked ROADMAP.md and confirmed this PR does not duplicate
planned core work
- [x] I have searched GitHub for duplicate or related PRs and linked
them above
- [x] I have either (a) linked existing issues with `Fixes: #` / `Closes
#` / `Refs #` OR (b) described the issue in-PR following the relevant
issue template
- [x] I have not referenced internal/instance-local Paperclip issues or
links (only public GitHub `#NNN` / `github.com/paperclipai/paperclip`
URLs)
- [x] My branch name describes the change (e.g. `docs/...`, `fix/...`,
`feat/...`) and contains no internal Paperclip ticket id or
instance-derived details
- [x] I have run tests locally and they pass
- [x] I have added or updated tests where applicable
- [ ] I have updated relevant documentation to reflect my changes
- [x] I have considered and documented any risks above
- [ ] All Paperclip CI gates are green
- [ ] Greptile is 5/5 with no open P2s, recommendations, or follow-ups
- [x] I will address all Greptile and reviewer comments before
requesting merge
---------
Co-authored-by: Harold Kim <harold@paperclip.ing>
Co-authored-by: Paperclip <noreply@paperclip.ing>
## Thinking Path
> - Paperclip is the open source app people use to manage AI agents for
work
> - Agent runs get their environment from user/adapter/project/routine
env bindings resolved by the server heartbeat, plus `PAPERCLIP_*`
runtime vars (identity, wake, workspace, API access) injected by the
harness
> - The heartbeat stripped **every** `PAPERCLIP_`-prefixed binding
before resolution, so legitimately user-named keys (e.g. cloud provider
token bindings like `PAPERCLIP_CLOUD_PROD_PROVIDER_RAILWAY_*`) were
silently dropped and never reached the run env
> - At the same time, several adapters honored an explicitly configured
`PAPERCLIP_API_KEY` over the harness-minted run token, which is exactly
the one key config must never control
> - This pull request replaces the blanket prefix strip with a precise
three-rule policy: never accept `PAPERCLIP_API_KEY` from config, always
let harness-assigned runtime vars win, and let every other
`PAPERCLIP_*`-named user binding flow through
> - The benefit is that user secrets with a `PAPERCLIP_`-style name work
like any other binding, while runtime identity and API credentials stay
fully harness-controlled
## Linked Issues or Issue Description
**Bug description** (no public issue exists):
- **What happened:** Env bindings whose key starts with `PAPERCLIP_`
(e.g. a cloud provider token a user deliberately named
`PAPERCLIP_CLOUD_PROD_PROVIDER_RAILWAY_TOKEN`) were silently stripped by
the server before secret resolution, so the spawned agent never received
them. No error, no access event — the variable just never appeared.
- **Expected behavior:** A user-named `PAPERCLIP_*` binding should reach
the run env unless the harness itself uses that key. Only
`PAPERCLIP_API_KEY` should be categorically rejected, and
harness-assigned runtime vars (`PAPERCLIP_RUN_ID`, `PAPERCLIP_AGENT_ID`,
wake/workspace vars, …) should always win over config.
- **Steps to reproduce:** Configure an agent/project env binding named
`PAPERCLIP_<ANYTHING>` (plain or secret_ref), run a heartbeat, and
inspect the spawned process env — the key is absent.
- **Deployment mode:** local server, any local adapter.
Related prior PRs (different, save-time/API-layer blanket-ban approach;
this PR supersedes that direction with a runtime allow-except-reserved
policy): Refs #8239, Refs #8439.
## What Changed
- `server/src/services/heartbeat.ts`: the pre-resolution strip now
removes only `PAPERCLIP_API_KEY` (hard denylist) instead of every
`PAPERCLIP_`-prefixed binding; other `PAPERCLIP_*` keys flow into
binding resolution. Low-trust inline-sensitive-env checks now also cover
those keys.
- `packages/adapter-utils/src/server-utils.ts`: new
`isForbiddenConfigEnvKey()` helper; the shared
`refreshPaperclipWorkspaceEnvForExecution` merge drops
`PAPERCLIP_API_KEY` from config and keeps harness-assigned `PAPERCLIP_*`
keys authoritative.
- `packages/adapter-utils/src/acpx-engine/execute.ts`: removed the
explicit-`PAPERCLIP_API_KEY`-from-config allowance; the run token
(`authToken`) is now always applied; config `PAPERCLIP_API_KEY` is
ignored.
- All local adapters (`claude-local`, `codex-local`, `cursor-local`,
`gemini-local`, `grok-local`, `opencode-local`, `pi-local`) plus
`cursor-cloud`, `hermes`, and the server `process` adapter: removed
`hasExplicitApiKey`-style allowances so the harness token always wins,
and guarded the remaining unguarded env-merge loops (claude-local inline
loop, process adapter) with the same policy.
- Tests updated/added: heartbeat binding-strip test now asserts the
three-rule policy; adapter-utils merge tests assert the
`PAPERCLIP_API_KEY` ban and `PAPERCLIP_*` pass-through; acpx engine
tests moved credential fixtures to `authToken` and assert config
`PAPERCLIP_API_KEY` is ignored while other `PAPERCLIP_*` config keys
forward and still bust the session fingerprint on rotation.
## Verification
- `pnpm vitest run packages/adapter-utils/src/server-utils.test.ts
packages/adapter-utils/src/acpx-engine/execute.test.ts` — 127 passed
- `pnpm vitest run server/src/__tests__/heartbeat-project-env.test.ts
server/src/__tests__/heartbeat-local-environment.test.ts
server/src/__tests__/claude-local-execute.test.ts
server/src/__tests__/codex-local-execute.test.ts
server/src/__tests__/cursor-local-execute.test.ts
server/src/__tests__/gemini-local-execute.test.ts` — 68 passed
- Adapter package execute suites and the server tests touching API-key
fixtures (`heartbeat-run-log`, `redaction`,
`effective-run-config-fingerprints`, `agent-permissions-routes`) —
green. Three pre-existing sandbox/SSH fixture failures reproduce
identically on clean `master` on this host and are unrelated.
- `pnpm --filter <pkg> typecheck` for server, adapter-utils, and all
nine touched adapter packages — all pass.
## Risks
- Behavioral change: a deployment that relied on configuring a static
`PAPERCLIP_API_KEY` in adapter config env loses that override — by
design; the harness-minted run token is now the only source. When no run
token exists, no API key is injected at all.
- `PAPERCLIP_*`-named user bindings now reach binding resolution and run
envs; a key that collides with a harness runtime var is still discarded
at merge time, so runtime identity/wake/workspace vars cannot be
spoofed.
- Low risk otherwise: no migrations, no API surface changes.
## Model Used
- Claude Fable 5 (`claude-fable-5`, Anthropic Claude 5 family,
Mythos-class tier), extended thinking enabled, agentic tool use (file
edits, shell, test runner) via Claude Agent SDK.
## Checklist
- [x] I have included a thinking path that traces from project context
to this change
- [x] I have specified the model used (with version and capability
details)
- [x] I have checked ROADMAP.md and confirmed this PR does not duplicate
planned core work
- [x] I have searched GitHub for duplicate or related PRs and linked
them above
- [x] I have either (a) linked existing issues with `Fixes: #` / `Closes
#` / `Refs #` OR (b) described the issue in-PR following the relevant
issue template
- [x] I have not referenced internal/instance-local Paperclip issues or
links (only public GitHub `#NNN` / `github.com/paperclipai/paperclip`
URLs)
- [x] My branch name describes the change (e.g. `docs/...`, `fix/...`)
and contains no internal Paperclip ticket id or instance-derived details
- [x] I have run tests locally and they pass
- [x] I have added or updated tests where applicable
- [x] I have updated relevant documentation to reflect my changes
- [x] I have considered and documented any risks above
- [ ] All Paperclip CI gates are green
- [x] Greptile is 5/5 with no open P2s, recommendations, or follow-ups
- [x] I will address all Greptile and reviewer comments before
requesting merge
🤖 Generated with [Claude Code](https://claude.com/claude-code)
---------
Co-authored-by: Paperclip <noreply@paperclip.ing>
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
## Thinking Path
> - Paperclip manages AI agents for work; agents can run in remote
sandboxes via the codex-local adapter
> - When a remote sandbox run starts, the codex-local adapter syncs the
host CODEX_HOME directory into the sandbox so Codex can authenticate and
load its configuration
> - The old approach staged the entire CODEX_HOME and relied on a 4-name
denylist (`tmp`, `.tmp`, `sessions`, `shell_snapshots`) to exclude
unnecessary state
> - The denylist was too narrow: large host-local runtime files
(`logs_2.sqlite`, `plugins/`, etc.) still uploaded on every remote run —
state the sandbox never uses
> - Additionally, `config.toml` (which embeds the managed MCP
`Authorization: Bearer …` header) was staged at 0644, relying solely on
the staging directory's 0700 permission for protection — a
credential-bearing file without per-file least-privilege
> - This pull request replaces the denylist with an explicit allowlist:
only `{auth.json, config.toml, config.json, instructions.md, skills/}`
are staged into a private `0700` temp dir; all staged regular files are
written `0600` (fail-closed on any staging error)
> - The fix is adapter-local to `packages/adapters/codex-local` with
zero changes to the shared `packages/adapter-utils` seam; inbound
auth-merge and outbound copy-back are preserved unchanged
## Linked Issues or Issue Description
No public GitHub issue for this specific bug. Describing the problem
directly:
**Bug:** The codex-local remote sandbox sync uploads more CODEX_HOME
content than the sandbox uses, and stages a credential-bearing file with
an over-permissive mode.
- **What happened:** The codex-local adapter synced the entire
CODEX_HOME directory minus four directory names (`tmp`, `.tmp`,
`sessions`, `shell_snapshots`). Large host-local runtime artifacts — a
SQLite write-ahead log that can reach 420 MB, plugin directories, crash
logs — uploaded on every remote run. `config.toml` (which embeds the
managed MCP `Authorization: Bearer …` header) was staged at `0644`; only
the staging temp dir's `0700` mode prevented broader exposure.
- **Expected behavior:** Only the files Codex needs inside the sandbox
should sync: `auth.json`, `config.toml`, `config.json`,
`instructions.md`, and `skills/`. All staged regular files should use
mode `0600` (single-layer credential protection is insufficient for
secret-bearing files).
- **Steps to reproduce:** Run a remote codex sandbox from a host with a
mature CODEX_HOME. Observe the sync payload includes `sessions/`,
`*.sqlite`, `plugins/`, etc.
Related work: Refs #9621 (outbound auth sync-back in the same
codex-local adapter area; this PR covers the inbound staging direction)
## What Changed
- New `stageCodexHomeForSync(codexHomeDir)` helper in
`packages/adapters/codex-local/src/server/codex-home.ts` stages only
allowlisted files into a fresh `fs.mkdtemp`-created `0700` temp dir,
then registers that dir as the `home` asset (dropping the old `exclude`
field entirely)
- All staged regular files written with mode `0600` — auth.json holds
single-use OAuth tokens; config.toml embeds MCP bearer headers; least
privilege applies throughout
- Symlinks dereferenced to resolved file bytes (`auth.json` single-use
token, each `skills/` entry land as real files)
- Missing-but-optional allowlist entries skipped silently (keyring-only
mode with no `auth.json`; absent `config.json`)
- Staged temp dir removed on teardown AND on error; staging is
fail-closed — a staging error aborts the run rather than shipping a
partial CODEX_HOME
- `execute.ts` updated to call `stageCodexHomeForSync` instead of the
old whole-dir + denylist path; the `exclude` field is removed from the
`home` asset
- Adapter-local fix: zero diff to `packages/adapter-utils` (shared
managed-runtime seam untouched); inbound auth-merge (`provision`) and
outbound copy-back (`restore`) preserved unchanged
## Verification
```sh
# Full codex-local test suite
npx vitest run packages/adapters/codex-local
# 174/174 pass
# Shared seam regression — must be zero diff to packages/adapter-utils
npx vitest run packages/adapter-utils
# 14/14 pass; zero changes to packages/adapter-utils source
# TypeScript (both packages)
npx tsc --noEmit -p packages/adapters/codex-local/tsconfig.json
npx tsc --noEmit -p packages/adapter-utils/tsconfig.json
# Both clean
```
Key new test coverage
(`packages/adapters/codex-local/src/server/codex-home.test.ts`):
- Stager copies exactly the allowlisted entries and nothing else
- Staged `auth.json` dereferenced from symlink and written mode `0600`
- Staged temp dir created with mode `0700`
- ALL staged regular files mode `0600`, including `config.toml` carrying
an MCP bearer token (regression test fails against the pre-fix `0644`
path)
- Fail-closed: I/O error during staging removes the temp dir and
re-throws
- Missing-but-optional entries (`config.json`) skipped without error
- Dangling symlink entries skipped without error
Execute-path integration test (`execute.remote.test.ts`): home asset
uses staged dir; staged config preserves provider block; skills survive;
staged temp dir removed after run teardown.
## Risks
- **Allowlist narrower than denylist:** Any CODEX_HOME file not in
`{auth.json, config.toml, config.json, instructions.md, skills/}` is now
excluded. If a codex-local configuration stores additional files in
CODEX_HOME that the sandbox requires, those runs will miss that state.
This tradeoff is intentional — only known-necessary files sync.
- **Permission change:** `config.toml` previously staged at `0644`, now
staged at `0600`. Strictly more restrictive; no known behavioral impact
(the sandbox reads as the file owner). This was the security gap that
motivated the fix.
- **Fail-closed staging:** A staging error (disk full, temp dir creation
failed) now aborts the run rather than falling back to the raw
CODEX_HOME. This is the safer behavior, but it changes the failure mode
from "run with potentially wrong state" to "run aborted cleanly."
## Model Used
Claude Opus 4.8 (claude-opus-4-8), Anthropic. Reasoning/agentic tool-use
mode (extended thinking, multi-step tool calls).
## Checklist
- [x] I have included a thinking path that traces from project context
to this change
- [x] I have specified the model used (with version and capability
details)
- [x] I have checked ROADMAP.md and confirmed this PR does not duplicate
planned core work
- [x] I have searched GitHub for duplicate or related PRs and linked
them above
- [x] I have either (a) linked existing issues with `Fixes: #` / `Closes
#` / `Refs #` OR (b) described the issue in-PR following the relevant
issue template
- [x] I have not referenced internal/instance-local Paperclip issues or
links (only public GitHub `#NNN` / `github.com/paperclipai/paperclip`
URLs)
- [x] My branch name describes the change (e.g. `docs/...`, `fix/...`)
and contains no internal Paperclip ticket id or instance-derived details
- [x] I have run tests locally and they pass
- [x] I have added or updated tests where applicable
- [x] I have updated relevant documentation to reflect my changes
- [x] I have considered and documented any risks above
- [ ] All Paperclip CI gates are green
- [ ] Greptile is 5/5 with no open P2s, recommendations, or follow-ups
- [ ] I will address all Greptile and reviewer comments before
requesting merge
---------
Co-authored-by: Harold Kim <harold@paperclip.ing>
Co-authored-by: Paperclip <noreply@paperclip.ing>
Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
## Thinking Path
> - Paperclip runs Codex as a sandboxed agent with its own CODEX_HOME;
at the start of each run Paperclip merges the sandbox's auth.json with
the host's so the agent can authenticate to the Codex API
> - The merge uses `codex-auth-merge-decision.cjs`, a small CJS script
that reads two `auth.json` snapshots and exits `10` (use source copy) or
`20` (keep destination copy)
> - Until now the script only handled the **inbound** direction (host →
sandbox) with hard-coded sandbox/host semantics
> - Sandbox-to-host copy-back (**outbound**) requires the same freshness
guard but with swapped roles — and exposing both directions via a flag
requires hard-coding knowledge about which role is "sandbox" and which
is "host" into the predicate itself
> - A cleaner design: the predicate is **direction-agnostic** — it
accepts a `source` path and a `destination` path; the caller decides
direction by which argument it passes first. The script answers one
question: "should destination be replaced by source?" Exit `10` = yes
(use source); exit `20` = no (keep destination)
> - This PR refactors `codex-auth-merge-decision.cjs` to this
source/destination model, removes the `--direction` flag, and extends
coverage with a 15-row source/destination predicate matrix plus updated
extract integration tests
## Linked Issues or Issue Description
No public GitHub issue exists. The underlying problem:
**Context** — Paperclip merges `auth.json` files between sandbox and
host to keep Codex credentials in sync across agent runs. The existing
decision predicate (`codex-auth-merge-decision.cjs`) handled the inbound
direction (host→sandbox) with hard-coded sandbox/host naming. A
copy-back (outbound) direction was needed, but rather than add a
`--direction` flag that hard-codes role knowledge into the predicate,
the design was simplified: the predicate is now **direction-agnostic**,
taking `source` and `destination` paths and answering "replace
destination with source?" The freshness guard is unified: exit `10` only
when source and destination share the same `account_id`, same
subscription-kind, both carry a parseable `last_refresh`, and
`source.last_refresh > destination.last_refresh` (strict). Ties and null
freshness keep the destination copy so a spent single-use refresh token
(openai/codex#15502) is never written back over a valid one.
Related PRs:
- Refs #9785 — Phase 2: relocated the decision predicate from
`adapter-utils` into `codex-local`; this PR builds on that merged change
- Refs #9621 — community PR targeting Codex sandbox auth sync-back via a
different approach; operates on different files (runtime layer vs.
decision script), non-conflicting
## What Changed
-
`packages/adapters/codex-local/src/server/codex-auth-merge-decision.cjs`
— refactored from direction-parameterized (sandbox/host with
`--direction`) to **direction-agnostic source/destination** semantics.
Args: first = source path, second = destination path. Exit `10` (use
source) only when: same `account_id`, same subscription-kind, both carry
a parseable `last_refresh`, and `source.last_refresh >
destination.last_refresh` (strict). Every other case exits `20` (keep
destination). Unknown/extra flags are still rejected with exit `1`. No
token bytes are emitted to stdout/stderr.
- `packages/adapters/codex-local/src/server/codex-auth-merge.test.ts` —
rewritten to source/destination API: 15-row predicate matrix
(strictly-newer→`10`;
tie/older/identity-mismatch/null-refresh/apikey/kind-mismatch/unusable→`20`)
plus token-bytes non-emission guard. The inbound extract integration
suite updated to match.
- Runtime callers (`codex-auth-merge-extract.sh`) already invoke `node
decision <sandbox> <host>` = `(source, destination)` and branch on exit
`10` → keep sandbox, so **no caller changes are needed**.
**Behavioral note:** unifying the two directions under a single "use
source only when strictly fresher" rule means a tie (exact equal
`last_refresh`) now keeps the **destination** for both the inbound and
outbound frames. For inbound restore, this means an exact timestamp tie
keeps the host copy. On a tie, both credentials are the same generation
— preferring the canonical destination (host) copy is more conservative.
This is an intentional consequence of the unified guard.
## Verification
```sh
# From the codex-local package root:
pnpm vitest run src/server/codex-auth-merge.test.ts
# Expected: all predicate rows green (15-row source/destination matrix + token-leak guard)
# Plus: codex-auth-merge-extract integration rows green
```
The test suite drives the `.cjs` as a child process over temporary
`auth.json` fixtures:
- Source/destination predicate matrix (15 rows): strictly-newer→`10`;
tie→`20`; older→`20`; identity-mismatch→`20`; source last_refresh
unparseable→`20`; destination last_refresh unparseable→`20`; apikey
either side→`20`; kind-mismatch→`20`; unusable either side→`20`
- Token-bytes non-emission guard: asserts no auth token bytes appear in
`stdout`/`stderr`
- Inbound extract integration: `codex-auth-merge-extract.sh` end-to-end
rows including tie/identity-mismatch/sandbox-newer cases
## Risks
- **Tie behavior change**: exact `last_refresh` ties now always keep the
destination copy (was: keep-sandbox for inbound). On a tie both
credentials are the same generation; keeping the canonical destination
(host) copy is the more conservative choice. Covered by test assertions.
- **Token non-emission**: `stdout`/`stderr` of the `.cjs` must not echo
credential bytes; covered by the token-bytes non-emission guard test.
- **Unknown/extra flags**: handled with `process.exit(1)` rather than a
silent default, so misconfigured callers surface errors immediately.
- **Caller compatibility**: the only runtime caller,
`codex-auth-merge-extract.sh`, already calls `node decision <sandbox>
<host>` with no flags — compatible with the new source/destination API
with no changes required.
- **Low risk overall**: the script is exercised end-to-end via the
extract integration tests; all paths covered.
## Model Used
Claude claude-sonnet-4-6 (Anthropic, `claude-sonnet-4-6`), 200 k context
window, tool use and agentic code execution enabled. Automated Paperclip
multi-agent system.
## Checklist
- [x] I have included a thinking path that traces from project context
to this change
- [x] I have specified the model used (with version and capability
details)
- [x] I have checked ROADMAP.md and confirmed this PR does not duplicate
planned core work
- [x] I have searched GitHub for duplicate or related PRs and linked
them above
- [x] I have either (a) linked existing issues with `Fixes: #` / `Closes
#` / `Refs #` OR (b) described the issue in-PR following the relevant
issue template
- [x] I have not referenced internal/instance-local Paperclip issues or
links (only public GitHub `#NNN` / `github.com/paperclipai/paperclip`
URLs)
- [x] My branch name describes the change (e.g. `docs/...`, `fix/...`)
and contains no internal Paperclip ticket id or instance-derived details
- [x] I have run tests locally and they pass
- [x] I have added or updated tests where applicable
- [x] I have updated relevant documentation to reflect my changes
- [x] I have considered and documented any risks above
- [x] All Paperclip CI gates are green
- [x] Greptile is 5/5 with no open P2s, recommendations, or follow-ups
- [x] I will address all Greptile and reviewer comments before
requesting merge
---------
Co-authored-by: Harold Kim <harold@paperclip.ing>
Co-authored-by: Paperclip <noreply@paperclip.ing>
## Thinking Path
> - Paperclip is an open source platform for managing AI agent teams,
with adapter packages providing concrete runtime environments (e.g.
`codex-local` runs a local OpenAI Codex sandbox)
> - `adapter-utils` is the shared utilities package — it should hold
only generic, adapter-agnostic primitives (sandbox lifecycle helpers,
merge logic, type definitions) usable by every adapter
> - Three files lived in `adapter-utils/src/` that are entirely
Codex-specific: `codex-auth-merge-extract.sh` (shell script that
extracts auth tokens), `codex-auth-merge-decision.cjs` (CJS decision
helper), and `codex-auth-merge-scripts.ts` (TypeScript factory that
wires them into the inbound provision seam added in PR #9778)
> - Their presence in a "generic" package violates the adapter isolation
principle and requires a cross-package build step to copy `.sh`/`.cjs`
files into `codex-local/dist/server/` at build time
> - This PR completes Phase 2 of the inbound-seam refactor: move all
three files to `packages/adapters/codex-local/src/server/`, rebase the
TypeScript imports, update `execute.ts` to import locally, move the
Codex-auth tests into a new `codex-local` test file, and fix packaging
so `codex-local` copies its own scripts to `dist/server/`
> - Script bytes are identical after the move; no change to inbound
auth-merge behavior or to which bytes cross the sandbox boundary
> - The benefit is a clean ownership boundary: `adapter-utils` retains
only generic runtime code, and the structural "Codex-free core" test in
the adapter-utils suite validates this invariant going forward
## Linked Issues or Issue Description
No pre-existing public GitHub issue. Describing the underlying problem
inline:
**Problem:** `packages/adapter-utils/src/` contains three files
(`codex-auth-merge-extract.sh`, `codex-auth-merge-decision.cjs`,
`codex-auth-merge-scripts.ts`) consumed exclusively by the `codex-local`
adapter. Their presence in a generic utilities package violates adapter
isolation and requires a cross-package build step (copy `.sh`/`.cjs`
into `codex-local/dist/server/`). The inbound provision seam landed in
PR #9778 routed these through `adapter-utils`; this PR finishes the
relocation.
**Related:** Refs #9778 (Phase 1 — generic asset-lifecycle-seam, now
merged).
## What Changed
- Moved `codex-auth-merge-extract.sh`, `codex-auth-merge-decision.cjs`,
and `codex-auth-merge-scripts.ts` from `packages/adapter-utils/src/` →
`packages/adapters/codex-local/src/server/` (script bytes are unchanged;
`codex-auth-merge-scripts.ts` imports rebased to `adapter-utils`
subpaths for `shellQuote` and `SandboxManagedRuntimeAssetProvision`)
- `packages/adapters/codex-local/src/server/execute.ts`: updated import
of `buildCodexAuthInboundProvision` from `adapter-utils` → local
`./codex-auth-merge-scripts`
- New
`packages/adapters/codex-local/src/server/codex-auth-merge.test.ts` — 5
Codex-auth test rows extracted from
`adapter-utils/src/workspace-restore-merge.test.ts`; the generic test
file stays and no longer references Codex
- `packages/adapters/codex-local/package.json`: build now copies
`.sh`/`.cjs` from `src/server/` into `dist/server/` directly
- `packages/adapter-utils/package.json`: removed the now-obsolete
cross-package copy step for those files
## Verification
```sh
# Type-check both affected packages
pnpm --filter @paperclipai/adapter-utils --filter @paperclipai/adapter-codex-local typecheck
# Codex auth-merge suite (moved tests — 5/5 pass)
pnpm --filter @paperclipai/adapter-codex-local test -- --reporter=verbose codex-auth-merge
# Full adapter-utils suite including the structural "core free of Codex literals" test (243 passed, 4 pre-existing skips)
pnpm --filter @paperclipai/adapter-utils test
# Confirm build copies scripts to dist/server
pnpm --filter @paperclipai/adapter-codex-local build
ls packages/adapters/codex-local/dist/server/*.sh packages/adapters/codex-local/dist/server/*.cjs
```
All of the above were run locally and passed before this PR was opened.
## Risks
Low risk — pure file relocation:
- No change to `.sh` or `.cjs` script bytes; the same content reaches
the sandbox boundary as before
- No behavioral change to inbound auth-merge logic from the sandbox's
perspective
- `adapter-utils`' structural "core free of Codex literals" test now
validates that the relocation is complete and will catch any regression
- Only `codex-local` consumed these files from `adapter-utils`; no other
package in the monorepo imported them from there
## Model Used
Claude Sonnet 4.6 (`claude-sonnet-4-6`) — Anthropic, 200k context
window, tool use, agentic reasoning. Used to author the refactor.
`Co-authored-by: Paperclip <noreply@paperclip.ing>` trailer present on
all commits.
## Checklist
- [x] I have included a thinking path that traces from project context
to this change
- [x] I have specified the model used (with version and capability
details)
- [x] I have checked ROADMAP.md and confirmed this PR does not duplicate
planned core work
- [x] I have searched GitHub for duplicate or related PRs and linked
them above
- [x] I have either (a) linked existing issues with `Fixes: #` / `Closes
#` / `Refs #` OR (b) described the issue in-PR following the relevant
issue template
- [x] I have not referenced internal/instance-local Paperclip issues or
links (only public GitHub `#NNN` / `github.com/paperclipai/paperclip`
URLs)
- [x] My branch name describes the change (e.g. `docs/...`, `fix/...`)
and contains no internal Paperclip ticket id or instance-derived details
- [x] I have run tests locally and they pass
- [x] I have added or updated tests where applicable
- [x] I have updated relevant documentation to reflect my changes
- [x] I have considered and documented any risks above
- [ ] All Paperclip CI gates are green
- [ ] Greptile is 5/5 with no open P2s, recommendations, or follow-ups
- [x] I will address all Greptile and reviewer comments before
requesting merge
Co-authored-by: Harold Kim <harold@paperclip.ing>
Co-authored-by: Paperclip <noreply@paperclip.ing>
## Thinking Path
> - Paperclip's sandbox managed runtime is responsible for provisioning
the agent's execution environment — it extracts a home directory asset
into the sandbox before the adapter runs.
> - The sandbox runtime core was directly branching on the adapter key
(`codex`) to decide which merge scripts to stage and which merge-extract
command to run, coupling generic infrastructure to a specific adapter's
credential-merge protocol.
> - This makes it harder to add, remove, or modify per-adapter asset
provisioning without touching the runtime core; it also prevents other
adapters from contributing staged files or a custom extract command at
all.
> - The fix is to move the adapter-specific knowledge into the adapter
itself: the asset descriptor gains optional `provision` (stageFiles +
extractCommand) and `restore` contribution fields that any adapter can
populate, and the runtime core consumes them generically.
> - This pull request introduces those contribution fields, wires the
Codex adapter's inbound credential-merge as a `provision` contribution,
and removes the adapter-specific branching from the runtime core.
> - The benefit is a clean seam: the runtime core is now
adapter-agnostic for asset provisioning, the inbound behavior is
unchanged (same merge matrix, same scripts), and other adapters can
attach custom staged files or extract commands without modifying shared
infrastructure.
## Linked Issues or Issue Description
No pre-existing public GitHub issue. Describing the problem inline per
the feature template:
**Problem or motivation**
The sandbox managed-runtime asset provisioning in
`sandbox-managed-runtime.ts` branched directly on the adapter key
(`codex`) to decide which merge scripts to stage and which shell command
to use during asset extraction. This tight coupling prevents other
adapters from customizing their provisioning without modifying the
runtime core, and it means the runtime core must import and know about
adapter-specific merge scripts.
**Proposed solution**
Add an optional `provision` contribution (array of `stageFiles` entries
+ an `extractCommand` string) and an optional `restore` contribution to
the asset descriptor returned by adapters. The runtime core now consumes
these generically — if a `provision` contribution is present, it stages
those files and uses the supplied command; otherwise it falls back to
the default `tar -xf` extraction. The Codex adapter populates the
`provision` contribution where it previously depended on core branching.
**Alternatives considered**
Keeping the adapter-specific logic in the core as a documented
exception; rejected because it makes the seam inextensible.
**Roadmap alignment**
Decoupling — removes a latent coupling between the runtime core and a
specific adapter.
## What Changed
- Added `provision` contribution field (`stageFiles: Array<{src, dest}>`
+ `extractCommand: string`) to the `SandboxManagedRuntimeAsset`
descriptor type in `adapter-utils`.
- Added `restore` contribution field (hook for post-restore logic,
populated in a later phase) to the descriptor.
- Removed adapter-key branching (`if adapterKey === 'codex'`) from the
runtime core in `sandbox-managed-runtime.ts`; the core now reads
`provision.stageFiles` and `provision.extractCommand` generically.
- Extracted Codex-specific merge-script paths and the merge-extract
command into `codex-auth-merge-scripts.ts` in `adapter-utils`; the Codex
adapter's `execute.ts` now attaches them as a `provision` contribution
when it builds its managed-home asset descriptor.
- Updated `execution-target.ts` to pass the extended asset type through
to the adapter call site so the new fields are load-bearing end-to-end.
- Added seam-proving unit tests in `sandbox-managed-runtime.test.ts`:
contribution-less asset uses the default path; a non-adapter asset
round-trips the generic provision+restore seam; a structural assertion
verifies the runtime core carries no Codex-specific string literals.
- Added one test in `workspace-restore-merge.test.ts` confirming the
inbound merge matrix is unaffected.
## Verification
```bash
# Unit tests (20 pass):
npx vitest run packages/adapter-utils/src/sandbox-managed-runtime.test.ts packages/adapter-utils/src/workspace-restore-merge.test.ts
# Type-check both affected packages:
cd packages/adapter-utils && npx tsc --noEmit
cd packages/adapters/codex-local && npx tsc --noEmit
# Structural: runtime core carries no adapter string literals
grep -n 'codex\|auth\.json' packages/adapter-utils/src/sandbox-managed-runtime.ts
# Expected: zero matches
```
## Risks
**Low risk.** This is a behavior-preserving refactor: the inbound
provisioning output (which files get staged, which command runs) is
identical to before, now driven by the adapter-supplied contribution
instead of core branching. The existing inbound merge matrix tests are
the regression guard. No change to which bytes cross the sandbox
boundary. The SSH transport is untouched.
## Model Used
- **Provider:** Anthropic
- **Model:** Claude Sonnet 4.6 (`claude-sonnet-4-6`)
- **Context window:** 200 K tokens
- **Capabilities used:** tool use (file read/edit, bash execution,
Paperclip API), extended reasoning over multi-file TypeScript refactor
- **Mode:** agentic (Paperclip ACPX platform)
## Checklist
- [x] I have included a thinking path that traces from project context
to this change
- [x] I have specified the model used (with version and capability
details)
- [x] I have checked ROADMAP.md and confirmed this PR does not duplicate
planned core work
- [x] I have searched GitHub for duplicate or related PRs and linked
them above
- [x] I have either (a) linked existing issues with `Fixes: #` / `Closes
#` / `Refs #` OR (b) described the issue in-PR following the relevant
issue template
- [x] I have not referenced internal/instance-local Paperclip issues or
links (only public GitHub `#NNN` / `github.com/paperclipai/paperclip`
URLs)
- [x] My branch name describes the change (e.g. `docs/...`, `fix/...`)
and contains no internal Paperclip ticket id or instance-derived details
- [x] I have run tests locally and they pass
- [x] I have added or updated tests where applicable
- [x] I have updated relevant documentation to reflect my changes
- [x] I have considered and documented any risks above
- [x] All Paperclip CI gates are green
- [x] Greptile is 5/5 with no open P2s, recommendations, or follow-ups
- [x] I will address all Greptile and reviewer comments before
requesting merge
---------
Co-authored-by: Paperclip <noreply@paperclip.ing>
Co-authored-by: Harold Kim <harold@paperclip.ing>
## Thinking Path
> - Paperclip is the open source control plane for managing AI-agent
companies.
> - Agent adapters are responsible for launching, observing, and
terminating provider processes safely.
> - The local Codex adapter uses an output-inactivity monitor to stop
genuinely hung child processes.
> - The existing seven-minute default also stopped legitimate
long-running tasks that emitted no output while tests or remote checks
were still running.
> - Explicit per-agent timeout overrides already provide configuration
flexibility, so the smallest safe correction is to increase only the
default window.
> - This pull request raises the default to 30 minutes, retains the
existing termination behavior, and adds a regression assertion for the
new value.
> - The benefit is fewer unnecessary Codex restarts while still bounding
genuinely silent processes well below the platform safety limit.
## Linked Issues or Issue Description
**Pre-submission checklist**
- Searched open and closed GitHub issues and pull requests; no duplicate
fix exists. The original inactivity monitor was introduced in #5017.
- Reproduces on the current `master` implementation.
- The termination originates in Paperclip's Codex adapter inactivity
monitor rather than the model provider or local configuration.
**What happened?**
Long-running `codex_local` tasks were terminated after seven minutes
without stdout or stderr, even when the child process was still
performing legitimate work such as a quiet test suite or waiting for
remote checks.
**Expected behavior**
The default inactivity window should tolerate common long-running quiet
tasks while continuing to terminate processes that remain silent for an
extended period.
**Steps to reproduce**
1. Start a `codex_local` run with the default
`outputInactivityTimeoutMs` configuration.
2. Have the child process perform legitimate work without emitting
stdout or stderr for more than seven minutes.
3. Observe the adapter terminate the process at the old default
threshold.
**Version / deployment**
Current `master`, built from source in local/self-hosted deployments,
using the Codex adapter. This behavior is not database- or
access-context-specific.
## What Changed
- Raise `DEFAULT_CODEX_OUTPUT_INACTIVITY_TIMEOUT_MS` from seven minutes
to 30 minutes.
- Update the adapter configuration documentation to state the new
default.
- Add a focused regression assertion that pins the default to 30
minutes.
## Verification
- `pnpm vitest run
packages/adapters/codex-local/src/server/output-inactivity-monitor.test.ts`
— 15 tests passed.
- `pnpm --filter @paperclipai/adapter-codex-local typecheck` — passed.
## Risks
- Low risk: only the fallback default changes; explicit positive timeout
values and `null` disablement retain their existing behavior.
- A genuinely silent Codex process now remains alive up to 23 minutes
longer before the same SIGTERM/SIGKILL cleanup path runs.
- No schema, API, migration, UI, or workflow changes are included.
> This is a focused bug fix and does not overlap with planned core
feature work in `ROADMAP.md`.
## Model Used
- Anthropic Claude Fable 5 via the `claude_local` adapter produced the
initial investigation and implementation with repository/tool access.
- OpenAI Codex via the Codex CLI prepared the PR, added the focused
regression assertion, and ran verification; the runtime did not expose a
more specific model ID or context-window value.
## Checklist
- [x] I have included a thinking path that traces from project context
to this change
- [x] I have specified the model used (with version and capability
details)
- [x] I have checked ROADMAP.md and confirmed this PR does not duplicate
planned core work
- [x] I have searched GitHub for duplicate or related PRs and linked
them above
- [x] I have either (a) linked existing issues with `Fixes: #` / `Closes
#` / `Refs #` OR (b) described the issue in-PR following the relevant
issue template
- [x] I have not referenced internal/instance-local Paperclip issues or
links (only public GitHub `#NNN` / `github.com/paperclipai/paperclip`
URLs)
- [ ] I have used a public-friendly branch name without internal tracker
identifiers (execution-workspace exception: this branch is
runtime-managed and cannot be renamed)
- [x] I have run tests locally and they pass
- [x] I have added or updated tests where applicable
- [x] I have updated relevant documentation to reflect my changes
- [x] I have considered and documented any risks above
- [x] All Paperclip CI gates are green
- [x] Greptile is 5/5 with no open P2s, recommendations, or follow-ups
- [x] I will address all Greptile and reviewer comments before
requesting merge
The source branch is execution-workspace managed and cannot be renamed
during this run; the PR title and body intentionally contain no internal
tracker references.
Co-authored-by: Paperclip <noreply@paperclip.ing>
## Thinking Path
> - Paperclip is the open source control plane teams use to manage AI
agents and their work
> - The Codex local adapter supervises CLI child processes and
terminates genuinely silent runs
> - The existing inactivity timer only recognized parsed JSONL stdout
events as activity
> - Long verification commands can emit ordinary stdout or stderr while
producing no JSONL, so healthy children could be killed
> - This pull request makes the watchdog observe raw child-process
output before filtering or parsing
> - The benefit is that long typecheck, build, and test phases survive
while truly silent children remain bounded
## Linked Issues or Issue Description
- **Bug:** The Codex output-inactivity watchdog could terminate healthy
runs during long verification phases because non-JSON stdout and stderr
did not reset its timer.
- **Expected:** Any bytes emitted by the child process count as output
activity; only a child with no stdout or stderr for the configured
interval is terminated.
- **Reproduction:** Configure a short `outputInactivityTimeoutMs`, run a
Codex child that periodically emits plain-text verification progress
without JSONL events, and observe the old monitor firing despite
continued output.
- **Deployment mode:** Local `codex_local` adapter execution.
- Related implementation: #5017
- Related recovery behavior: #8680
## What Changed
- Reset the Codex inactivity monitor on every non-empty stdout or stderr
chunk before stderr noise filtering.
- Track raw output chunk and byte counts in monitor diagnostics while
retaining parsed JSONL event counts.
- Add regression coverage for more than 21 simulated minutes of non-JSON
verification output and retain silent-child termination coverage.
- Document that `outputInactivityTimeoutMs` observes raw child output
and that `null` still disables the monitor.
## Verification
- `/srv/paperclip/home/paperclipai/paperclip/node_modules/.bin/vitest
run
packages/adapters/codex-local/src/server/output-inactivity-monitor.test.ts
packages/adapters/codex-local/src/server/output-inactivity-monitor.integration.test.ts`
- `/srv/paperclip/home/paperclipai/paperclip/node_modules/.bin/tsc -p
packages/adapters/codex-local/tsconfig.json --noEmit`
## Risks
- Low risk: the monitor becomes more conservative and may allow a
noisy-but-stuck child to run longer, but the configured hard timeout and
platform silent-run safety net remain unchanged.
- No schema, API, migration, or UI changes.
> For core feature work, check [`ROADMAP.md`](ROADMAP.md) first and
discuss it in `#dev` before opening the PR. Feature PRs that overlap
with planned core work may need to be redirected — check the roadmap
first. See `CONTRIBUTING.md`.
## Model Used
- OpenAI Codex CLI coding agent; runtime model ID and context-window
size were not exposed to the agent. Used repository/tool access, code
execution, and focused test verification.
## Checklist
- [x] I have included a thinking path that traces from project context
to this change
- [x] I have specified the model used (with version and capability
details)
- [x] I have checked ROADMAP.md and confirmed this PR does not duplicate
planned core work
- [x] I have searched GitHub for duplicate or related PRs and linked
them above
- [x] I have either (a) linked existing issues with `Fixes: #` / `Closes
#` / `Refs #` OR (b) described the issue in-PR following the relevant
issue template
- [x] I have not referenced internal/instance-local Paperclip issues or
links (only public GitHub `#NNN` / `github.com/paperclipai/paperclip`
URLs)
- [x] My branch name describes the change (e.g. `docs/...`, `fix/...`)
and contains no internal Paperclip ticket id or instance-derived details
- [x] I have run tests locally and they pass
- [x] I have added or updated tests where applicable
- [x] I have updated relevant documentation to reflect my changes
- [x] I have considered and documented any risks above
- [x] All Paperclip CI gates are green
- [x] Greptile is 5/5 with no open P2s, recommendations, or follow-ups
- [x] I will address all Greptile and reviewer comments before
requesting merge
---------
Co-authored-by: Paperclip <noreply@paperclip.ing>
## Thinking Path
> - Paperclip is the open source control plane people use to coordinate
AI-agent companies and their work.
> - Its recovery subsystem detects stranded issue execution and decides
whether to retry, escalate, or request operator intervention.
> - The existing recovery path used a mostly generic owner ladder and
generic execution contract, so transient failures could wake a manager
who then performed the deliverable instead of repairing and returning
the task.
> - Provider quota failures also entered the same takeover path even
when the correct action was to wait for capacity and retry the original
assignee.
> - Recovery actions already retain the source owner and evidence needed
to choose a cause-specific route, render a scoped contract, and measure
whether work was handed back.
> - This pull request adds a cause-keyed recovery playbook, propagates
its contract through every built-in adapter, and makes resolved recovery
actions return work to the original owner by default.
> - The benefit is bounded self-recovery that preserves task ownership,
avoids needless management takeover, and makes recovery outcomes
observable.
## Linked Issues or Issue Description
No matching public GitHub issue was found.
Related recovery work was reviewed but is not duplicated here: #9630
restores bounded recovery continuations, #8807 changes one
assignee-ranking case, and #9404 records runtime-failure transition
evidence. This change instead introduces cause-specific routing and
recovery contracts across the recovery lifecycle.
### What happened?
When an issue became stranded, recovery generally selected an owner
through the same fallback ladder and rendered the normal execution
contract. That made the recovery wake look like ordinary deliverable
work, even when the correct action was to retry the original agent,
repair its runtime, or wait for a provider quota reset.
### Expected behavior
Recovery should select a response by failure cause, tell the recipient
to recover rather than complete the deliverable, suppress takeover wakes
for provider quota waits, and return repaired work to its original
assignee unless the recovery owner explicitly completes it.
### Actual behavior
Recovery could escalate transient failures to management, omit the
cause-specific next action from the wake, and leave the recovery owner
assigned after the runtime problem was resolved.
### Impact
The generic path creates avoidable management work, ownership churn, and
budget consumption while obscuring whether recovery successfully
returned work to the responsible agent.
## What Changed
- Added cause-keyed routing for process loss, missing disposition,
provider quota limits, Codex output inactivity, workspace validation
failures, and fallback recovery causes.
- Added recovery-scoped wake rendering that replaces the generic
execution contract with the failure summary, original assignee, attempt
count, next action, and cause-specific playbook instruction.
- Propagated the structured recovery contract through all built-in
adapter execution paths, including Hermes local and gateway adapters.
- Added provider-quota wait monitoring so capacity failures schedule the
original assignee instead of enqueueing a takeover wake.
- Added hand-back behavior and `handed_back` / `owner_completed` outcome
accounting when recovery actions are resolved.
- Added focused routing, renderer, quota-monitor, and hand-back
regression coverage plus implementation-spec documentation.
## Verification
- `pnpm exec vitest run packages/adapter-utils/src/server-utils.test.ts
server/src/__tests__/heartbeat-process-recovery.test.ts
server/src/__tests__/heartbeat-workspace-branch-containment.test.ts
server/src/__tests__/issue-recovery-actions.test.ts`
- 4 test files passed; 194 tests passed.
- Targeted `pnpm --filter ... typecheck` across
`@paperclipai/adapter-utils`, `@paperclipai/shared`,
`@paperclipai/server`, `@paperclipai/ui`, and all nine changed adapter
packages.
- 13 affected workspace packages passed typecheck.
- `pnpm check:token-gates`
- All UI token gates passed.
## Risks
- Recovery routing behavior changes for stranded work, so an incorrectly
classified cause could select a different recipient than before;
fallback causes retain the existing management ladder.
- Provider quota detection depends on structured failure evidence and
conservative text matching; unmatched failures continue through fallback
recovery.
- Adapter prompt plumbing changes across built-ins, covered by shared
renderer tests and compile-time call signatures.
> For core feature work, check [`ROADMAP.md`](ROADMAP.md) first and
discuss it in `#dev` before opening the PR. Feature PRs that overlap
with planned core work may need to be redirected — check the roadmap
first. See `CONTRIBUTING.md`.
## Model Used
- OpenAI Codex with exact model ID `gpt-5.6-sol`, using reasoning, tool
use, and code execution. The runtime does not expose its configured
context-window size.
## Checklist
- [x] I have included a thinking path that traces from project context
to this change
- [x] I have specified the model used (with version and capability
details)
- [x] I have checked ROADMAP.md and confirmed this PR does not duplicate
planned core work
- [x] I have searched GitHub for duplicate or related PRs and linked
them above
- [x] I have either (a) linked existing issues with `Fixes: #` / `Closes
#` / `Refs #` OR (b) described the issue in-PR following the relevant
issue template
- [x] I have not referenced internal/instance-local Paperclip issues or
links (only public GitHub `#NNN` / `github.com/paperclipai/paperclip`
URLs)
- [x] My branch name describes the change (e.g. `docs/...`, `fix/...`)
and contains no internal Paperclip ticket id or instance-derived details
- [x] I have run tests locally and they pass
- [x] I have added or updated tests where applicable
- [x] I have updated relevant documentation to reflect my changes
- [x] I have considered and documented any risks above
- [x] All Paperclip CI gates are green
- [x] Greptile is 5/5 with no open P2s, recommendations, or follow-ups
- [x] I will address all Greptile and reviewer comments before
requesting merge
---------
Co-authored-by: Paperclip <noreply@paperclip.ing>
## Thinking Path
> - Paperclip is the open source app people use to manage AI agents for
work
> - Paperclip supports the Codex local adapter, which runs OpenAI Codex
CLI sessions on behalf of agents
> - Codex uses OAuth refresh tokens to maintain long-running
authenticated sessions
> - When a refresh fails, the failure has distinct root causes: a
refresh token was already reused in a parallel request, the token
expired by TTL, or the token was invalidated/revoked by the provider
> - Without classifying these failure modes, all refresh auth errors
surface identically — operators cannot distinguish retryable transient
collisions from permanent invalidations, and run logs carry no
actionable diagnosis
> - This pull request adds structured classification
(`refresh_token_reused`, `refresh_token_expired`,
`refresh_token_invalidated`) of Codex refresh-token auth failures across
the CLI quota-probe, ACP auth path, and execute path
> - The benefit is that these distinct failure modes can be surfaced in
run logs and acted on appropriately — transient reuse can be retried;
true invalidations require re-auth
## Linked Issues or Issue Description
<!-- Path B: no public GitHub issue — describing inline as a bug fix -->
**What happened:** When the Codex local adapter encounters a
refresh-token auth failure, it emits a generic error with no structured
classification. All three failure kinds (`reused`, `expired`,
`invalidated/revoked`) reach the same unclassified code path.
**Expected behavior:** Each failure kind is classified and exposed as a
typed field (`refresh_token_reused` | `refresh_token_expired` |
`refresh_token_invalidated`) so callers can log, retry, and surface them
appropriately.
**Steps to reproduce:**
1. Run a Codex agent session with a reused or expired OAuth refresh
token.
2. Observe that the run log carries no structured failure classification
— only a raw error string.
**Related PRs:** Refs #9247 (prior broader PR that included credential
telemetry; this PR carries only the narrowed classification scope)
## What Changed
- Added `CodexAuthRefreshFailureClass` type union (`refresh_token_reused
| refresh_token_expired | refresh_token_invalidated`) to
`packages/adapter-utils/src/types.ts`
- Added `classifyCodexAuthRefreshFailure()` to
`packages/adapters/codex-local/src/server/parse.ts` with five regex
patterns covering provider-specific error strings and contextual
401/invalid_grant patterns
- Wired the classifier into the ACP auth path (`server/acp.ts`), execute
path (`server/execute.ts`), and CLI quota-probe (`cli/quota-probe.ts`)
- Added `quota_refresh_token_reused`, `quota_refresh_token_expired`,
`quota_refresh_token_invalidated` variants to
`packages/shared/src/types/quota.ts`
- Added classification unit tests (`parse.test.ts`,
`quota-spawn-error.test.ts`, `acp.test.ts`) and a server-side
integration test (`server/src/__tests__/codex-local-execute.test.ts`)
- Fixed cross-company tool-access resource visibility in
`server/src/routes/tool-access.ts`
- Stabilized `heartbeat-retry-scheduling.test.ts` (CASCADE cleanup),
`heartbeat-run-log.test.ts`, and `quota-windows.test.ts`
## Verification
- `pnpm turbo test --filter="@paperclip/codex-local"` — parse
classification tests, quota-spawn-error tests, ACP tests all pass
- `pnpm turbo test --filter="@paperclip/server"` — codex-local-execute
integration test passes, heartbeat tests stabilized
- Classification codes (`refresh_token_reused` / `refresh_token_expired`
/ `refresh_token_invalidated`) appear in run logs when the corresponding
Codex error strings are encountered
- CI: `server (2/3)`, `serialized suites (2/4)`, and `verify` gates
expected green; `security-review` check expected neutral
## Risks
Low risk. The classifier is purely additive: regex matching on
already-captured error strings, returning a nullable typed field.
Callers that do not inspect the classification field are unaffected. No
execution paths, retry logic, or existing error surfaces changed.
## Model Used
- **Provider:** Anthropic
- **Model ID:** `claude-sonnet-4-6`
- **Context window:** 200K tokens
- **Mode:** standard tool use (no extended thinking)
## Checklist
- [x] I have included a thinking path that traces from project context
to this change
- [x] I have specified the model used (with version and capability
details)
- [x] I have checked ROADMAP.md and confirmed this PR does not duplicate
planned core work
- [x] I have searched GitHub for duplicate or related PRs and linked
them above
- [x] I have either (a) linked existing issues with `Fixes: #` / `Closes
#` / `Refs #` OR (b) described the issue in-PR following the relevant
issue template
- [x] I have not referenced internal/instance-local Paperclip issues or
links (only public GitHub `#NNN` / `github.com/paperclipai/paperclip`
URLs)
- [x] My branch name describes the change (e.g. `docs/...`, `fix/...`)
and contains no internal Paperclip ticket id or instance-derived details
- [x] I have run tests locally and they pass
- [x] I have added or updated tests where applicable
- [x] I have updated relevant documentation to reflect my changes
- [x] I have considered and documented any risks above
- [x] All Paperclip CI gates are green
- [x] Greptile is 5/5 with no open P2s, recommendations, or follow-ups
- [x] I will address all Greptile and reviewer comments before
requesting merge
---------
Co-authored-by: Paperclip <noreply@paperclip.ing>
## Thinking Path
> - Paperclip is the open source platform people use to manage AI agents
for work
> - The Codex adapter runs agent tasks in isolated sandbox environments
on the user's machine
> - When a Codex sandbox is reused across agent runs, its home directory
(including `~/.codex/auth.json`) is restored from a prior snapshot
> - Both the host machine and the sandbox independently maintain
`auth.json` credentials; on sandbox reuse, these can diverge
> - The previous merge code had fail-open edge cases: if host auth was
in an unusable state, if the auth JSON object shapes differed between
host and sandbox, or if the subscription account identities didn't
match, the merge would proceed silently with whatever data was available
> - This PR adds fail-closed behavior: if host Codex auth is unusable,
if auth parser shapes differ, or if subscription account identities
don't match, the merge fails explicitly rather than silently continuing
with stale or incorrect credentials
> - The benefit is that Codex agents on reused sandboxes now fail fast
and loudly when auth is in a broken state, instead of silently running
with wrong credentials and producing confusing downstream failures
## Linked Issues or Issue Description
No pre-existing public GitHub issue. This is a targeted security
hardening fix for the Codex reused-sandbox auth merge path.
**Problem:** When a Codex sandbox is reused, the merge logic that
reconciles host and sandbox `auth.json` credentials failed open in
several cases:
- Host `auth.json` present but in an unusable state (missing required
keys, empty token material, malformed JSON) → merge would proceed with
whatever the sandbox had
- Host and sandbox auth payloads had different shapes (e.g., one uses
`OPENAI_API_KEY`, the other uses a `tokens` object) →
parser-differential case not detected
- Subscription account identities (`tokens.account_id`) differed between
host and sandbox → stale sandbox identity would be used silently
**Fix:** All three cases now fail closed. The merge returns an explicit
error rather than proceeding with potentially stale or mismatched
credentials.
Related PRs:
- Refs #9262 — sandbox Codex auth shadow warning (adjacent auth area)
- Refs #9259 — auth precedence exports (adjacent auth area)
## What Changed
- `packages/adapters/codex-local/src/server/codex-home.ts` — New file
with `hasUsableAuthPayload()`, `codexHomeHasUsableAuth()`, and full
Codex home setup/teardown. Includes fail-closed auth merge guards:
rejects unusable host auth, detects parser shape differentials, and
checks subscription account identity match before merging
- `packages/adapter-utils/src/workspace-restore-merge.ts` — New file
with directory snapshot diffing and restore-merge logic; the merge
operation fails closed when auth validation fails
- `packages/adapters/codex-local/src/server/codex-home.test.ts` — Unit
tests covering auth usability checks, symlink management, and
fail-closed merge paths
- `packages/adapter-utils/src/workspace-restore-merge.test.ts` — Unit
tests for snapshot/restore-merge behavior including fail-closed cases
- `packages/adapter-utils/src/sandbox-managed-runtime.ts` — Updated to
invoke the fail-closed auth merge during sandbox restore
## Verification
Tests run and passing:
```sh
corepack pnpm exec vitest run packages/adapter-utils/src/workspace-restore-merge.test.ts packages/adapters/codex-local/src/server/codex-home.test.ts
corepack pnpm exec vitest run packages/adapter-utils/src/sandbox-managed-runtime.test.ts
corepack pnpm --filter @paperclipai/adapter-utils typecheck
corepack pnpm --filter @paperclipai/adapter-codex-local typecheck
git diff --check origin/master HEAD
```
All passed locally before push.
## Risks
- **Intentional behavioral change (breaking for previously-silent
failures):** Reused sandboxes that previously completed auth merge with
unusable host auth, parser-differential auth shapes, or mismatched
account identities will now fail with an explicit error. This is the
correct behavior — the prior silent-proceed path was the bug. Users
affected will see a clear error message rather than a confusing
downstream auth failure.
- **Auth.json symlink migration:** `ensureSymlink()` detects stale
copied `auth.json` files (written by older Paperclip versions) and
replaces them with symlinks on first run. This is safe: the target is
always under the Paperclip-managed company home, never the user's real
`~/.codex`. Directories at the symlink path are left untouched (EISDIR
is not silently swallowed).
- **Low risk for non-reuse paths:** The fail-closed logic only activates
during sandbox restore/reuse. Fresh sandbox allocations are unaffected.
## Model Used
- **Provider:** Anthropic
- **Model ID:** claude-sonnet-4-6 (Claude Sonnet 4.6)
- **Context window:** 200k tokens
- **Mode:** Agentic coding with tool use; extended thinking not used
- **Role:** Code author (Priya Raman, BackendEngineer) with Harold Kim
(Git Expert) handling push and PR operations
## Checklist
- [x] I have included a thinking path that traces from project context
to this change
- [x] I have specified the model used (with version and capability
details)
- [x] I have checked ROADMAP.md and confirmed this PR does not duplicate
planned core work
- [x] I have searched GitHub for duplicate or related PRs and linked
them above
- [x] I have either (a) linked existing issues with `Fixes: #` / `Closes
#` / `Refs #` OR (b) described the issue in-PR following the relevant
issue template
- [x] I have not referenced internal/instance-local Paperclip issues or
links (only public GitHub `#NNN` / `github.com/paperclipai/paperclip`
URLs)
- [x] My branch name describes the change (e.g. `docs/...`, `fix/...`)
and contains no internal Paperclip ticket id or instance-derived details
- [x] I have run tests locally and they pass
- [x] I have added or updated tests where applicable
- [x] I have updated relevant documentation to reflect my changes
- [x] I have considered and documented any risks above
- [x] All Paperclip CI gates are green
- [x] Greptile is 5/5 with no open P2s, recommendations, or follow-ups
- [x] I will address all Greptile and reviewer comments before
requesting merge
---------
Co-authored-by: Priya Raman <priya.raman@paperclip.local>
Co-authored-by: Paperclip <noreply@paperclip.ing>
Co-authored-by: Harold Kim <harold@paperclip.ing>
## Thinking Path
> - Paperclip is the open source app people use to manage AI agents for
work
> - Governed MCP access spans contracts, runtime enforcement, adapters,
UI surfaces, and operator verification
> - The parity reference PR #9534 is too large for effective automated
or human review
> - The feature therefore needs a linear stack whose individual diffs
stay below the 100-file review limit
> - This pull request is split 4/8 and focuses on gateway runtime, Smoke
Lab, plugins, and server wiring
> - The benefit is a standalone, testable review boundary while
preserving byte-for-byte parity at the top of the stack
## Linked Issues or Issue Description
- Related parity reference: #9534
- Problem: The policy core needs runtime execution, endpoint guards,
route registration, heartbeat integration, and adapter MCP injection to
become operational.
- Proposed solution: Adds the remaining server routes/wiring/consumers,
runtime tests, adapter-utils MCP contracts, and Claude/Codex injection
implementations required by the server layer.
- Alternatives considered: keeping #9534 as one 403-file review, or
rewriting the feature to manufacture seams; both were rejected in favor
of path extraction plus compile-driven boundary moves.
- Roadmap alignment: this advances the existing governed MCP/tool-access
work already represented by #9534; it does not introduce a separate
roadmap initiative.
- Stack position: base branch is `pap10341-split/03-server-tool-access`.
- Merge policy: merge bottom-up, in order, only after the complete
eight-PR stack has been reviewed and the top-of-stack parity gate
remains empty.
- Requested review: SecurityEngineer for gateway, endpoint guard, token
issuance, and runtime wiring; Greptile on every PR.
## What Changed
- Adds the remaining server routes/wiring/consumers, runtime tests,
adapter-utils MCP contracts, and Claude/Codex injection implementations
required by the server layer.
- Keeps this PR below 100 changed files and independently typecheckable.
- Preserves the final tree from #9534 when combined with the other seven
stack levels.
## Verification
- `pnpm typecheck`
- Changed server test set — 26 files, 382 tests passed
- Affected server adapter tests — 38 tests passed after concrete adapter
boundary move
- Adapter-utils and Codex focused tests — 76 tests passed
## Risks
- Remote endpoint validation, token handling, and runtime supervision
are security-sensitive and can fail closed or deny legitimate access if
misconfigured.
- Stack risk: merging out of order can expose incomplete layers;
mitigate by following the documented bottom-up merge policy.
- Parity risk: later edits to an intermediate branch can drift from
#9534; mitigate by re-running the empty top-of-stack diff before merge.
> For core feature work, check [`ROADMAP.md`](ROADMAP.md) first and
discuss it in `#dev` before opening the PR. Feature PRs that overlap
with planned core work may need to be redirected — check the roadmap
first. See `CONTRIBUTING.md`.
## Model Used
- OpenAI Codex, exact model ID `gpt-5.4`; runtime-managed context
window; medium reasoning with repository, shell, Git, GitHub CLI, and
code-execution tools enabled.
## Checklist
- [x] I have included a thinking path that traces from project context
to this change
- [x] I have specified the model used (with version and capability
details)
- [x] I have checked ROADMAP.md and confirmed this PR does not duplicate
planned core work
- [x] I have searched GitHub for duplicate or related PRs and linked
them above
- [x] I have either (a) linked existing issues with `Fixes: #` / `Closes
#` / `Refs #` OR (b) described the issue in-PR following the relevant
issue template
- [x] Internal references are omitted except the execution-plan link
explicitly required for this coordinated split stack
- [x] My branch name describes the change and contains no internal
Paperclip ticket id or instance-derived details
- [x] I have run tests locally and they pass
- [x] I have added or updated tests where applicable
- [x] I have updated relevant documentation to reflect my changes
- [x] I have considered and documented any risks above
- [ ] All Paperclip CI gates are green
- [ ] Greptile is 5/5 with no open P2s, recommendations, or follow-ups
- [x] I will address all Greptile and reviewer comments before
requesting merge
## Stack Coordination
- Internal execution plan:
[PAP-13874](/PAP/issues/PAP-13874#document-plan)
- Parity reference: #9534
- Stack: #9556 → #9557 → #9558 → #9559 → #9560 → #9561 → #9562 → #9563
- Merge bottom-up only after full-stack review and an empty parity diff
at #9563.
---------
Co-authored-by: Paperclip <noreply@paperclip.ing>
## Thinking Path
> - Paperclip is the open source control plane people use to manage AI
agents for work.
> - Local coding adapters currently spawn their CLI processes directly
on the Paperclip host.
> - CLI-native approval and sandbox flags do not provide a reliable host
filesystem or network boundary.
> - An agent can therefore inspect unrelated host files or fetch
external material when an operator needs stronger isolation.
> - The confinement must stay opt-in so existing local adapter behavior
does not change unexpectedly.
> - This pull request adds a shared Linux Bubblewrap spawn layer for
workspace filesystem and deny/allowlist network scopes.
> - The benefit is enforceable defense in depth around Codex and Claude
local runs while preserving explicit provider connectivity.
## Linked Issues or Issue Description
### What happened?
`codex_local` and `claude_local` processes could read arbitrary host
paths and make unrestricted outbound network requests because Paperclip
did not impose a spawn-level boundary.
### Expected behavior
Operators can opt into a workspace-only filesystem view and either deny
network egress or allow exact provider/API hosts, independently of CLI
approval flags.
### Steps to reproduce
1. Run current `master` on Linux and configure a Codex or Claude local
adapter.
2. Ask the agent to read a canary file outside its active workspace.
3. Ask the agent to `curl` a public host.
4. Observe that both operations succeed without a Paperclip-level
confinement option.
### Environment
- Paperclip commit: `c36f1a4af` / current `master` base.
- Deployment mode: Linux local dev or self-hosted server.
- Installation: built from source.
- Adapters: Codex and Claude Code.
- Database: not related.
## What Changed
- Added a shared Bubblewrap process wrapper with opt-in
`filesystemScope: "workspace"`, managed/extra path mounts, private
`/tmp`, and Linux-only validation.
- Added `networkScope: "deny" | "allowlist"`; both use a private network
namespace, while allowlist mode exposes an exact-host HTTP(S) proxy over
a Unix-socket bridge.
- Wired Codex and Claude local CLI execution through the wrapper and
forced scoped auto runs onto the CLI lane because ACP processes are not
covered.
- Added unit and gated Bubblewrap canaries for outside-file denial,
workspace writes, direct network denial, allowlisted forwarding, and
rejected destinations.
- Documented both scopes, provider allowlist examples, Bubblewrap
requirements, and default-off behavior.
## Verification
- `pnpm exec vitest run
packages/adapter-utils/src/local-process-sandbox.test.ts
packages/adapters/codex-local/src/server/acp.test.ts
packages/adapters/claude-local/src/server/acp.test.ts` — 34 passed, 4
gated Bubblewrap tests skipped by default.
- `pnpm --filter @paperclipai/adapter-utils typecheck`
- `pnpm --filter @paperclipai/adapter-codex-local typecheck`
- `pnpm --filter @paperclipai/adapter-claude-local typecheck`
- `pnpm --filter @paperclipai/adapter-utils build`
- `pnpm --filter @paperclipai/adapter-codex-local build`
- `pnpm --filter @paperclipai/adapter-claude-local build`
- Attempted the gated tests with a vendored Bubblewrap binary; this
container blocks unprivileged namespace setup (`setting up uid map:
Permission denied` / loopback `RTM_NEWADDR: Operation not permitted`),
so kernel-level execution remains for CI or a namespace-enabled Linux
host.
## Risks
- Bubblewrap must be installed and unprivileged user/mount/network
namespaces must be enabled on the host; scoped runs fail clearly if the
prerequisite is missing.
- Allowlist mode depends on the coding CLI honoring standard
`HTTP_PROXY` / `HTTPS_PROXY` variables; custom providers must list every
required exact hostname and port.
- Exact-host allowlists intentionally reject wildcards, which is safer
but may require operators to enumerate multi-host provider setups.
- No behavior changes unless an operator enables `filesystemScope` or
`networkScope`.
> This aligns with the ROADMAP direction toward safer remote and
sandboxed agent environments and does not duplicate an open PR or issue
found in the repository search.
## Model Used
- OpenAI GPT-5.5 (`gpt-5.5`) via Codex CLI, with reasoning, repository
tool use, shell execution, code editing, and test execution. The serving
context-window size is not exposed to the agent.
## Checklist
- [x] I have included a thinking path that traces from project context
to this change
- [x] I have specified the model used (with version and capability
details)
- [x] I have checked ROADMAP.md and confirmed this PR does not duplicate
planned core work
- [x] I have searched GitHub for duplicate or related PRs and linked
them above
- [x] I have either (a) linked existing issues with `Fixes: #` / `Closes
#` / `Refs #` OR (b) described the issue in-PR following the relevant
issue template
- [x] I have not referenced internal/instance-local Paperclip issues or
links (only public GitHub `#NNN` / `github.com/paperclipai/paperclip`
URLs)
- [x] My branch name describes the change (e.g. `docs/...`, `fix/...`)
and contains no internal Paperclip ticket id or instance-derived details
- [x] I have run tests locally and they pass
- [x] I have added or updated tests where applicable
- [x] I have updated relevant documentation to reflect my changes
- [x] I have considered and documented any risks above
- [ ] All Paperclip CI gates are green
- [ ] Greptile is 5/5 with no open P2s, recommendations, or follow-ups
- [x] I will address all Greptile and reviewer comments before
requesting merge
---------
Co-authored-by: Paperclip <noreply@paperclip.ing>
## Thinking Path
> - Paperclip is the open source control plane people use to manage
AI-agent companies
> - Budgets and spend telemetry are control-plane safety features, not
just reporting
> - Local Codex and Claude adapters can execute through either ACP or
their native CLI engines
> - The ACP lane records usage and reported cost, but CLI JSON output
often reports tokens without a price
> - The CLI lane was either losing per-run usage semantics or coercing
missing cost to zero, making real usage indistinguishable from a
genuinely free run
> - This pull request preserves CLI usage as per-run totals and records
token-bearing runs without a reported price as explicitly unpriced
ledger events
> - The benefit is accurate usage accounting and a visible pricing gap
instead of silently misleading zero-cost telemetry
## Linked Issues or Issue Description
Refs #9471
Refs #9230
**Bug description**
A `codex_local` run using the CLI engine can emit a final
`turn.completed` event with millions of input tokens and tens of
thousands of output tokens while the agent's spend ledger remains
indistinguishable from a true zero-usage, zero-cost run. Claude CLI
output has the same missing-price edge case.
**Expected behavior**
Token-bearing CLI runs should persist their usage. If the adapter
reports a price, the ledger should record it as reported; if the CLI
reports usage but no price, the ledger should explicitly mark the event
as unpriced rather than silently treating missing price data as a
reported `$0` cost.
**Reproduction shape**
1. Configure `codex_local` with `engine: cli`.
2. Run a task that produces a `turn.completed` usage payload.
3. Observe token usage in the run stream.
4. Before this change, missing price data is represented as ordinary
zero-cost spend and the CLI usage basis is not consistently propagated.
## What Changed
- Mark Codex and Claude native CLI usage totals as `per_run` and
propagate that basis through success and failure results.
- Stop coercing missing Claude CLI cost to `0`.
- Add `cost_status` to cost events with `reported` and `unpriced`
values, including an idempotent migration and shared validation/types.
- Persist token-bearing runs without a reported price as `unpriced`
ledger events while retaining zero cents until an authoritative price
exists.
- Add parser, execute-path, heartbeat-accounting, and cost-service
regression coverage for both local CLI adapters.
- Document the cost-status invariant and CLI accounting behavior.
## Verification
- `pnpm exec vitest run
packages/adapters/codex-local/src/server/parse.test.ts
packages/adapters/claude-local/src/server/parse.test.ts
server/src/__tests__/codex-local-execute.test.ts
server/src/__tests__/claude-local-execute.test.ts
server/src/__tests__/heartbeat-cost-accounting.test.ts
server/src/__tests__/costs-service.test.ts` — 6 files / 102 tests
passed.
- `pnpm --filter @paperclipai/shared typecheck`
- `pnpm --filter @paperclipai/db typecheck` — includes migration
numbering and safety checks.
- `pnpm --filter @paperclipai/adapter-codex-local typecheck`
- `pnpm --filter @paperclipai/adapter-claude-local typecheck`
- `pnpm --filter @paperclipai/server typecheck`
## Risks
- Existing cost rows default to `reported`, preserving current
interpretation; only new token-bearing events with absent cost are
marked `unpriced`.
- This change does not invent model pricing. Budget hard stops still
cannot charge an unknown amount, but operators and evals can now
distinguish missing pricing from a genuinely reported zero cost.
- Consumers that enumerate cost-event fields should tolerate the
additive `costStatus` field.
> For core feature work, check [`ROADMAP.md`](ROADMAP.md) first and
discuss it in `#dev` before opening the PR. Feature PRs that overlap
with planned core work may need to be redirected — check the roadmap
first. See `CONTRIBUTING.md`.
## Model Used
- OpenAI Codex, exact model `gpt-5.3-codex`, with repository tool use
and code execution; default reasoning mode.
## Checklist
- [x] I have included a thinking path that traces from project context
to this change
- [x] I have specified the model used (with version and capability
details)
- [x] I have checked ROADMAP.md and confirmed this PR does not duplicate
planned core work
- [x] I have searched GitHub for duplicate or related PRs and linked
them above
- [x] I have either (a) linked existing issues with `Fixes: #` / `Closes
#` / `Refs #` OR (b) described the issue in-PR following the relevant
issue template
- [x] I have not referenced internal/instance-local Paperclip issues or
links (only public GitHub `#NNN` / `github.com/paperclipai/paperclip`
URLs)
- [x] My branch name describes the change (e.g. `docs/...`, `fix/...`)
and contains no internal Paperclip ticket id or instance-derived details
- [x] I have run tests locally and they pass
- [x] I have added or updated tests where applicable
- [x] I have updated relevant documentation to reflect my changes
- [x] I have considered and documented any risks above
- [x] All Paperclip CI gates are green
- [x] Greptile is 5/5 with no open P2s, recommendations, or follow-ups
- [x] I will address all Greptile and reviewer comments before
requesting merge
Co-authored-by: Paperclip <noreply@paperclip.ing>
## Thinking Path
> - Paperclip is the open source control plane for running and governing
AI-agent companies.
> - Adapter executions feed token usage, billing identity, and run cost
into the control plane's spend telemetry.
> - The default ACP execution lane for local Claude and Codex adapters
did not propagate per-turn usage or cost, so paid runs could be recorded
with zero spend and no tokens.
> - Claude CLI result events could also undercount output tokens by
reading only the main-loop usage block instead of the complete per-model
ledger.
> - The shared executor needs to distinguish per-run usage from
session-cumulative usage so the server does not apply the wrong delta
heuristic.
> - This pull request captures ACP usage and cumulative-cost deltas,
resolves adapter billing identity, uses Claude's complete model-usage
ledger, and preserves per-run usage in server normalization.
> - The benefit is accurate token and cost accounting across the default
paid Claude and Codex execution paths.
## Linked Issues or Issue Description
### What happened?
Paid `claude_local` and `codex_local` runs using the default ACP engine
can complete successfully while the control plane records zero or null
cost and missing token usage. Claude CLI result parsing can additionally
undercount output tokens when subagent or sidechain usage is present.
### Steps to reproduce
1. Run a paid Claude or Codex local adapter through the ACP engine.
2. Complete a turn that reports usage and cumulative cost through ACP
status/events.
3. Inspect the execution result and normalized run telemetry.
### Expected behavior
The execution result contains per-turn token usage, a per-run USD cost
delta, and the correct billing identity. Server normalization records
those per-run values without applying a session-cumulative delta a
second time.
### Actual behavior before this change
ACP execution results returned no usage and `costUsd: null` with unknown
billing. The server therefore recorded zero spend and no tokens for paid
runs. Claude CLI parsing could use an incomplete usage block.
## What Changed
- Capture ACP usage from runtime status and `usage_update` events,
reporting it as `usageBasis: per_run`.
- Convert agent-reported cumulative ACP cost into a per-turn delta,
including counter-reset and no-report safeguards.
- Add a shared billing-identity resolver and map Claude and Codex
authentication/provider modes to control-plane billing types.
- Prefer Claude result-event `modelUsage` totals so subagent and
sidechain tokens are included.
- Skip the server's session-cumulative usage delta when an adapter
explicitly reports per-run usage.
- Add regression coverage for usage capture, event fallback, cost
resets, stale reports, billing identities, model-usage totals, and
server spend normalization.
## Verification
- `pnpm exec vitest run
packages/adapter-utils/src/acpx-engine/execute.test.ts
packages/adapters/claude-local/src/server/parse.test.ts
packages/adapters/claude-local/src/server/acp.test.ts
packages/adapters/codex-local/src/server/acp.test.ts
server/src/__tests__/costs-service.test.ts
server/src/__tests__/monthly-spend-service.test.ts` — 6 files, 126 tests
passed.
- `pnpm --filter @paperclipai/adapter-utils typecheck` — passed.
- `pnpm --filter @paperclipai/adapter-claude-local typecheck` — passed.
- `pnpm --filter @paperclipai/adapter-codex-local typecheck` — passed.
- `pnpm --filter @paperclipai/server typecheck` — passed.
- A broader Claude-local suite has a pre-existing rate-limit
classification failure in `test.probe.test.ts`; it also fails on clean
`master` and is unrelated to this change.
## Risks
- Cost reporting depends on the agent's cumulative counter semantics;
reset handling falls back to the post-turn amount and is covered by
regression tests.
- Incorrect billing-mode inference could misclassify spend;
provider/auth mappings mirror each adapter's existing CLI behavior and
have focused tests.
- The new `usageBasis` contract changes server normalization only when
adapters explicitly opt into `per_run`; existing adapters retain prior
behavior.
- No database migration, workflow, lockfile, or UI changes are included.
> For core feature work, check [`ROADMAP.md`](ROADMAP.md) first and
discuss it in `#dev` before opening the PR. Feature PRs that overlap
with planned core work may need to be redirected — check the roadmap
first. See `CONTRIBUTING.md`.
## Model Used
- Implementation commit: Anthropic Claude Fable 5, tool-enabled coding
workflow (exact context window and runtime configuration were not
recorded in the commit metadata).
- PR preparation and verification: OpenAI Codex, tool-enabled coding
agent (runtime model ID and context window are not exposed to this
session).
## Checklist
- [x] I have included a thinking path that traces from project context
to this change
- [x] I have specified the model used (with version and capability
details)
- [x] I have checked ROADMAP.md and confirmed this PR does not duplicate
planned core work
- [x] I have searched GitHub for duplicate or related PRs and linked
them above
- [x] I have either (a) linked existing issues with `Fixes: #` / `Closes
#` / `Refs #` OR (b) described the issue in-PR following the relevant
issue template
- [x] I have not referenced internal/instance-local Paperclip issues or
links (only public GitHub `#NNN` / `github.com/paperclipai/paperclip`
URLs)
- [x] My branch name describes the change (e.g. `docs/...`, `fix/...`)
and contains no internal Paperclip ticket id or instance-derived details
- [x] I have run tests locally and they pass
- [x] I have added or updated tests where applicable
- [x] I have updated relevant documentation to reflect my changes
- [x] I have considered and documented any risks above
- [x] All Paperclip CI gates are green
- [x] Greptile is 5/5 with no open P2s, recommendations, or follow-ups
- [x] I will address all Greptile and reviewer comments before
requesting merge
---------
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
Co-authored-by: Paperclip <noreply@paperclip.ing>
## Thinking Path
> - Paperclip is the open source app people use to manage AI agents for
work
> - Agent adapters (Claude, Codex, Gemini) default to the ACP engine
lane, which needs a live bidirectional stdio session with the agent
process
> - Sandbox execution targets only exposed one-shot command execution,
so every ACP-capable adapter refused remote targets and fell back to the
CLI lane with a "supports only the local Paperclip host" warning
> - Running agents in sandboxes is a core deployment mode, and losing
ACP there means losing streaming updates, structured events, and
default-lane parity with local runs
> - This pull request adds a provider-agnostic process-session bridge
that relays the ACP stdio session into the sandbox over the existing
sandbox runner contract, and updates the adapters to use it
> - The benefit is that the default ACP lane now behaves the same on the
local host and in any sandbox provider, with CLI fallback reserved for
targets that genuinely cannot host a bidirectional session
## Linked Issues or Issue Description
No existing public issue covers this; inline description following the
feature request template:
**Problem or motivation**
Configuring an ACP-capable adapter (e.g. Claude) with a sandbox
environment made every run fall back to the CLI lane with the warning
"Claude ACP currently supports only the local Paperclip host, but this
run targets a remote environment." The ACP engine only knew how to spawn
a local subprocess, while sandbox providers only expose one-shot command
execution — so there was no way to hold the bidirectional stdio session
ACP requires.
**Proposed solution**
Add a process-session bridge in `adapter-utils`: a local ACPX-spawnable
proxy script connects to a token-authenticated loopback TCP server,
which relays JSON-framed stdin/stdout/stderr events to and from a small
relay script executed inside the sandbox via the provider's ordinary
runner. Claude/Codex/Gemini adapters now treat sandbox targets with a
runner as ACP-capable, resolve agent commands against the remote target,
and fall back to CLI only when the sandbox exposes no bidirectional
path. The sandbox callback bridge injects a run-scoped API endpoint and
bridge token so the agent inside the sandbox can reach Paperclip
(including work-product handoffs) without ever receiving the host run
JWT.
**Alternatives considered**
A provider-specific lane was prototyped first: Daytona minting SSH
access metadata at lease time, converted into an SSH execution target.
It was dropped because it only worked for providers able to advertise
SSH, added per-provider surface area, and left every other sandbox
provider on the CLI fallback. The merged design rides the one-shot
runner contract all providers already implement; a regression test pins
that sandbox targets stay on the bridge lane even when lease metadata
advertises SSH access.
**Roadmap alignment**
Directly advances the "Cloud / Sandbox agents" roadmap item — agents
running in remote and sandboxed environments keep the same control-plane
behavior as local ones. No overlap with other planned core work.
## What Changed
- `packages/adapter-utils/src/execution-target.ts`: new
`startAdapterExecutionTargetProcessSessionBridge()` plus helpers —
writes a token-authenticated local proxy script (spawnable by ACPX) and
a remote relay script synced into the sandbox, with a loopback TCP
server streaming JSON-framed stdio between them; events emitted before
the ACP client attaches are buffered so none are lost.
- `packages/adapter-utils/src/acpx-engine/execute.ts`: the ACP engine
can execute against remote sandbox targets through the bridge instead of
requiring a local subprocess, including remote cwd/env shaping.
- `packages/adapter-utils/src/sandbox-callback-bridge.ts`:
sandbox-scoped API bridging extended to allow work-product handoffs; the
sandbox payload env carries a bridge token, never the host run JWT.
- `packages/adapters/claude-local`, `codex-local`, `gemini-local`
(`src/server/acp.ts`): default-lane selection no longer rejects all
remote targets; command resolution is remote-aware
(`ensureAdapterExecutionTargetCommandResolvable`,
`resolveAdapterExecutionTargetCwd`); the fallback reason is now scoped
to sandboxes that expose only one-shot execution.
- `server/src/__tests__/environment-execution-target.test.ts`: pins that
sandbox targets resolve to the bridge lane, including when lease
metadata advertises SSH access.
- Non-sandbox remote targets (e.g. SSH) keep the CLI lane: the ACP
engine's remote transport is sandbox-only, so default-lane selection
falls back for those targets across all three adapters, and tests
covering CLI-specific remote behavior pin `engine: "cli"` explicitly.
- The bridge authenticates loopback connections before they can own the
session or receive buffered output (token required, idle unauthenticated
peers dropped), and remote event writes are serialized so the exit event
always lands after stdout/stderr have drained.
- Daytona plugin: formatting-only residue from the earlier iteration; no
functional change.
## Verification
- `vitest run` over the touched suites —
`packages/adapter-utils/src/acpx-engine/execute.test.ts`,
`packages/adapter-utils/src/execution-target-sandbox.test.ts`,
`packages/adapter-utils/src/sandbox-callback-bridge.test.ts`, the three
adapter `acp.test.ts` files, and
`server/src/__tests__/environment-execution-target.test.ts` — 102 tests
pass.
- End to end: with a Claude agent configured on a Daytona sandbox
environment, the primary-model test now selects the default ACP lane (no
fallback warning), and the full round trip (wake → sandbox execution →
API bridge → comment post) was exercised twice from inside a live
sandbox.
## Risks
- Behavioral shift: adapters that previously always fell back to CLI on
sandbox targets now default to ACP there; `engine=cli` still pins the
CLI lane explicitly.
- The bridge relays stdio as JSON lines over loopback TCP guarded by a
per-session random token; the remote relay runs inside the sandbox under
the provider's runner. Providers with slow one-shot execution will see
higher session startup latency — the CLI fallback remains for genuinely
incapable targets.
- No schema or migration changes.
## Model Used
- Claude Fable 5 (`claude-fable-5`, Anthropic) — extended thinking
enabled, agentic tool use via the Claude Agent SDK harness;
implementation iterated with local Vitest verification.
## Checklist
- [x] I have included a thinking path that traces from project context
to this change
- [x] I have specified the model used (with version and capability
details)
- [x] I have checked ROADMAP.md and confirmed this PR does not duplicate
planned core work
- [x] I have searched GitHub for duplicate or related PRs and linked
them above
- [x] I have either (a) linked existing issues with `Fixes: #` / `Closes
#` / `Refs #` OR (b) described the issue in-PR following the relevant
issue template
- [x] I have not referenced internal/instance-local Paperclip issues or
links (only public GitHub `#NNN` / `github.com/paperclipai/paperclip`
URLs)
- [x] My branch name describes the change (e.g. `docs/...`, `fix/...`)
and contains no internal Paperclip ticket id or instance-derived details
- [x] I have run tests locally and they pass
- [x] I have added or updated tests where applicable
- [x] I have updated relevant documentation to reflect my changes (no
shipped docs describe the old local-only ACP limitation)
- [x] I have considered and documented any risks above
- [x] All Paperclip CI gates are green
- [x] Greptile is 5/5 with no open P2s, recommendations, or follow-ups
- [x] I will address all Greptile and reviewer comments before
requesting merge
---------
Co-authored-by: Cody <noreply@paperclip.ing>
Co-authored-by: Cody <cody@paperclip.local>
## Thinking Path
> - Paperclip is the open source app people use to manage AI agents for
work.
> - Codex local is the adapter subsystem that exposes OpenAI Codex CLI
model choices to agents and issue overrides.
> - OpenAI has GPT-5.6 Codex-capable models that should appear in
Paperclip's built-in Codex model list and refresh behavior.
> - Paperclip's server model listing falls back to the adapter metadata
and merges OpenAI refresh results with known Codex defaults.
> - This pull request updates the Codex default model metadata to
include GPT-5.6 options and adds regression coverage for fallback and
refresh paths.
> - The benefit is that operators can select the new Codex models
without relying on manual model IDs, and refresh behavior keeps known
GPT-5.6 options visible.
## Linked Issues or Issue Description
Refs #9322.
Refs #9342.
Refs #9346.
### Agent or provider
Codex CLI (OpenAI).
### Why this adapter is useful
OpenAI's GPT-5.6 Codex-capable models should be available in Paperclip's
Codex adapter defaults and model refresh path.
### How the agent is invoked
`codex`
## What Changed
- Changed the `codex_local` default model metadata from `gpt-5.5` to
`gpt-5.6`.
- Added `gpt-5.6-sol`, `gpt-5.6-terra`, and `gpt-5.6-luna` to the
built-in Codex adapter model list.
- Updated adapter and server model-listing tests to cover GPT-5.6
fallback and refresh behavior.
- Aligned Codex Fast mode support and helper text with the new `gpt-5.6`
default, while preserving GPT-5.5, GPT-5.4, and manual model ID support.
## Verification
- `git diff --check origin/master...HEAD`
- `pnpm exec vitest run packages/adapters/codex-local/src/index.test.ts
packages/adapters/codex-local/src/server/codex-args.test.ts
server/src/__tests__/adapter-models.test.ts
server/src/__tests__/adapter-model-refresh-routes.test.ts`
- `pnpm check:token-gates`
- `pnpm --filter @paperclipai/adapter-codex-local typecheck`
- `pnpm --filter @paperclipai/server typecheck`
- `pnpm --filter @paperclipai/ui typecheck`
## Risks
Medium risk because changing `DEFAULT_CODEX_LOCAL_MODEL` from `gpt-5.5`
to `gpt-5.6` changes the adapter's default model selection for new blank
configurations. The model-list additions are otherwise low risk and
covered by adapter/server metadata tests. This PR intentionally overlaps
related PRs #9342 and #9346, so reviewers may prefer to close or fold it
into one of those branches.
> For core feature work, check [`ROADMAP.md`](ROADMAP.md) first and
discuss it in `#dev` before opening the PR. Feature PRs that overlap
with planned core work may need to be redirected — check the roadmap
first. See `CONTRIBUTING.md`.
## Model Used
OpenAI Codex coding agent based on GPT-5, with shell, git, GitHub CLI,
and repository editing tool use. Exact served model ID and context
window were not exposed by the runtime.
## Checklist
- [x] I have included a thinking path that traces from project context
to this change
- [x] I have specified the model used (with version and capability
details)
- [x] I have checked ROADMAP.md and confirmed this PR does not duplicate
planned core work
- [x] I have searched GitHub for duplicate or related PRs and linked
them above
- [x] I have either (a) linked existing issues with `Fixes: #` / `Closes
#` / `Refs #` OR (b) described the issue in-PR following the relevant
issue template
- [x] I have not referenced internal/instance-local Paperclip issues or
links (only public GitHub `#NNN` / `github.com/paperclipai/paperclip`
URLs)
- [x] My branch name describes the change (e.g. `docs/...`, `fix/...`)
and contains no internal Paperclip ticket id or instance-derived details
- [x] I have run tests locally and they pass
- [x] I have added or updated tests where applicable
- [x] I have updated relevant documentation to reflect my changes
- [x] I have considered and documented any risks above
- [x] All Paperclip CI gates are green
- [x] Greptile is 5/5 with no open P2s, recommendations, or follow-ups
- [x] I will address all Greptile and reviewer comments before
requesting merge
---------
Co-authored-by: Paperclip <noreply@paperclip.ing>
## Thinking Path
> - Paperclip is the open source app people use to manage AI agents for
work
> - Agent runtimes are selected through adapters, and each adapter
exposes the model IDs an operator can pick in the UI
> - The `codex_local` and `opencode_local` adapters hardcode those
lists, so a newly released model stays unreachable until it is added
> - `gpt-5.4-mini` is available via both the Codex CLI and the OpenCode
CLI, but neither adapter lists it; `openai/gpt-5.5` is likewise missing
from `opencode_local`
> - This pull request adds those entries, and deliberately keeps
`gpt-5.4-mini` out of the Codex Fast mode allowlist because the model
does not support Fast mode
> - The benefit is that operators can select these models from the UI
instead of falling back to a manual model ID, and Fast mode fails closed
rather than sending unsupported overrides to the CLI
## Linked Issues or Issue Description
No existing issue covers this. Describing it inline, following the
feature request template:
### Problem or motivation
`gpt-5.4-mini` is absent from the `models` list of both the
`codex_local` and `opencode_local` adapters, and `openai/gpt-5.5` is
absent from `opencode_local`. (`codex_local` already ships `gpt-5.5` —
it is the default model on master.) Operators who want these models must
type a manual model ID.
For `codex_local` that has a real side effect.
`isCodexLocalFastModeSupported` treats any *unknown* model as
Fast-mode-capable and passes `service_tier="fast"` and
`features.fast_mode=true` through to the CLI. Because `gpt-5.4-mini`
does not support Codex Fast mode, an agent configured with a manual
`gpt-5.4-mini` model ID and `fastMode` enabled silently sends overrides
the CLI cannot honor.
### Proposed solution
Add the three missing entries to the two `models` lists, and leave
`CODEX_LOCAL_FAST_MODE_SUPPORTED_MODELS` untouched. Listing
`gpt-5.4-mini` in `models` is precisely what makes it a *known* model,
so `isCodexLocalFastModeSupported` returns `false`, `buildCodexExecArgs`
omits the Fast mode overrides, and `fastModeIgnoredReason` is surfaced
to the operator.
### Alternatives considered
Adding `gpt-5.4-mini` to `CODEX_LOCAL_FAST_MODE_SUPPORTED_MODELS` as
well — rejected, because the model does not support Fast mode and the
overrides would be rejected at run time. Leaving the models unlisted so
operators keep using manual IDs — rejected, because that is the path
that silently enables Fast mode for a model that cannot use it.
### Roadmap alignment
Not core roadmap work. `ROADMAP.md` does not plan adapter model-list
maintenance; this is routine upkeep as upstream CLIs ship new models.
## What Changed
- Add `gpt-5.4-mini` to the `codex_local` adapter's `models` list,
positioned after `gpt-5.4` (newest-first ordering).
- Add `openai/gpt-5.5` and `openai/gpt-5.4-mini` to the `opencode_local`
adapter's `models` list.
- Add a `buildCodexExecArgs` test asserting Fast mode is ignored for
`gpt-5.4-mini`.
`CODEX_LOCAL_FAST_MODE_SUPPORTED_MODELS` is intentionally unchanged. No
behavior changes to existing models or adapter logic.
## Verification
```
pnpm --filter @paperclipai/adapter-codex-local --filter @paperclipai/adapter-opencode-local typecheck
npx vitest run packages/adapters/codex-local packages/adapters/opencode-local
```
Both pass: typecheck clean on both packages, and 22 test files / 133
tests green, including the new `ignores fast mode for gpt-5.4-mini`
case.
## Risks
Low risk. The change is additive: three entries appended to two
model-selection lists, plus one test. No default model changes, no
adapter logic changes, no migrations.
One behavioral shift is intended. An operator who had `gpt-5.4-mini`
configured as a *manual* model ID with `fastMode` enabled was getting
Fast mode overrides passed through to the Codex CLI. After this change
`gpt-5.4-mini` is a known model, so those overrides are dropped and
`fastModeIgnoredReason` explains why.
## Model Used
- OpenAI Codex CLI with GPT-5 / GPT-5.5-assisted code editing (the
original commits on this branch).
- Anthropic Claude Opus 4.8 (`claude-opus-4-8`, 1M context, extended
thinking, tool use) for the master merge, conflict resolution, and the
scope reduction in the latest commit.
## Checklist
- [x] I have included a thinking path that traces from project context
to this change
- [x] I have specified the model used (with version and capability
details)
- [x] I have checked ROADMAP.md and confirmed this PR does not duplicate
planned core work
- [x] I have searched the GitHub PR list for similar or duplicate PRs
and confirmed this one is not a duplicate
- [x] I have either (a) linked existing issues with `Fixes: #` / `Closes
#` / `Refs #` OR (b) described the issue in-PR following the relevant
issue template
- [x] I have not referenced internal/instance-local Paperclip issues or
links (only public GitHub `#NNN` / `github.com/paperclipai/paperclip`
URLs)
- [x] My branch name describes the change (e.g. `docs/...`, `fix/...`)
and contains no internal Paperclip ticket id or instance-derived details
- [x] I have run tests locally and they pass
- [x] I have added or updated tests where applicable
- [x] I have updated relevant documentation to reflect my changes (N/A;
the adapter docs describe Fast mode support, which is unchanged)
- [x] I have considered and documented any risks above
- [ ] All Paperclip CI gates are green
- [ ] Greptile is 5/5 with no open P2s, recommendations, or follow-ups
- [x] I will address all Greptile and reviewer comments before
requesting merge
---------
Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
Co-authored-by: Andrew Aymeloglu <aaymeloglu@gmail.com>
## Thinking Path
> - Paperclip is the open source app people use to manage AI agents for
work.
> - Adapter packages are the bridge between the control plane and local
agent harnesses such as Claude Code, Codex, and Gemini CLI.
> - ACP support was concentrated in a separate `acpx_local` adapter,
which made ACP feel like a separate agent choice instead of an execution
capability of the harness adapters.
> - Claude, Codex, and Gemini now have ACP-capable harnesses, so the
native adapter should own ACP selection, fallback, config, transcript
parsing, and environment diagnostics.
> - The standalone ACPX adapter still needs a compatibility path for
existing rows, but it should not be offered as an active adapter for new
agents.
> - This pull request moves the shared ACP runtime into
`@paperclipai/acpx-engine`, wires Claude/Codex/Gemini local adapters to
prefer ACP when prerequisites are available, and retires `acpx_local` to
a tombstone.
> - The benefit is one adapter per harness, richer ACP transcripts by
default where possible, and a migration path for existing Claude/Codex
ACPX agents.
## Linked Issues or Issue Description
Closes#5932 — the broken default `acpx_local` Claude path is replaced
by native `claude_local` ACP support, existing Claude/Codex ACPX rows
migrate to native adapters, and new agents no longer choose the
standalone ACPX adapter.
Refs #4893 — original merged ACPX local adapter runtime that this PR
replaces with native per-harness ACP engines.
Refs #6590 — prior ACPX-Claude seamlessness work folded into the new
native Claude ACP path.
Refs #197 — related open generic ACP/Kiro adapter work; this PR does not
close it because Kiro/custom generic ACP remains a separate adapter
decision.
Refs #7018 — related Kimi-specific `acpx_local` shell failure; this PR
retires the built-in standalone adapter but does not add a native Kimi
adapter.
Refs #8864 — related ACPX prompt/API guidance PR; this PR moves runtime
guidance into the shared/native ACP engine path instead of the old
standalone adapter.
Refs #8881 — related `acpx_local` POSIX shell failure from the old
`acpx` pin; this PR updates ACP dependencies but does not claim
custom/OMP ACP support as a first-class native adapter.
Refs #8964 — related open `acpx_local` stderr cleanup PR; this PR makes
the old runtime path obsolete for new agents but keeps it as a
non-closing reference.
Problem description:
- The standalone `acpx_local` adapter duplicates Claude/Codex agent
choices that already have first-class local adapters.
- ACP should be an execution engine capability of each harness adapter
when the underlying harness supports ACP.
- Existing `acpx_local` agents should either migrate to native harness
adapters or fail with an explicit retirement message instead of silently
falling back to the process adapter.
## What Changed
- Added `@paperclipai/acpx-engine` as the shared ACP execution,
session-codec, CLI formatter, and UI parser package.
- Wired `claude_local`, `codex_local`, and `gemini_local` to auto-select
ACP by default when prerequisites pass, with `engine=cli` opt-out and
`engine=acp` strict mode.
- Added ACP config schema/UI fields, environment checks, session-codec
preservation, transcript parsing, and adapter capability metadata for
the native adapters.
- Retired `acpx_local` to a server tombstone, removed its
UI/package/runtime image surface, and added a migration for existing
Claude/Codex ACPX agents.
- Updated package manifests, lockfile, release tooling, docs, Kubernetes
sandbox defaults, and tests.
## Verification
- `corepack pnpm --filter @paperclipai/acpx-engine typecheck`
- `corepack pnpm --filter @paperclipai/adapter-claude-local typecheck`
- `corepack pnpm --filter @paperclipai/adapter-codex-local typecheck`
- `corepack pnpm --filter @paperclipai/adapter-gemini-local typecheck`
- `corepack pnpm --filter @paperclipai/acpx-engine exec vitest run`
- `corepack pnpm --filter @paperclipai/adapter-claude-local exec vitest
run src/server/acp.test.ts src/server/execute.acp-fallback.test.ts
src/ui/build-config.test.ts`
- `corepack pnpm --filter @paperclipai/adapter-codex-local exec vitest
run src/server/acp.test.ts src/ui/build-config.test.ts`
- `corepack pnpm --filter @paperclipai/adapter-gemini-local exec vitest
run src/server/acp.test.ts src/ui/build-config.test.ts
src/ui/parse-stdout.test.ts`
- `corepack pnpm --filter @paperclipai/plugin-sdk ensure-build-deps &&
corepack pnpm --filter @paperclipai/server exec tsc --noEmit`
- `corepack pnpm --filter @paperclipai/server exec vitest run
src/__tests__/adapter-routes.test.ts
src/__tests__/adapter-session-codecs.test.ts
src/__tests__/adapter-models.test.ts`
- `corepack pnpm --filter @paperclipai/ui typecheck`
- `corepack pnpm --filter @paperclipai/ui exec vitest run
src/adapters/metadata.test.ts
src/adapters/adapter-display-registry.test.ts
src/components/AgentConfigForm.test.ts
src/components/AgentConfigForm.render.test.tsx
src/components/transcript/RunTranscriptView.test.tsx`
- `node --test scripts/bootstrap-npm-package.test.mjs
scripts/release-package-map.test.mjs
scripts/verify-release-registry-state.test.mjs`
Note: the server typecheck script calls `pnpm` internally; this dev
shell exposes pnpm through Corepack only, so I ran the two script steps
manually with `corepack pnpm`.
## Risks
- Migration changes existing `acpx_local` Claude/Codex agents to native
adapter types and clears old ACPX task sessions/runtime state.
- Custom ACP commands remain on the retired tombstone and will need a
separate future adapter/plugin path.
- ACP auto-selection depends on local Node and ACP server command
prerequisites; remote and unsupported environments fall back to CLI
unless `engine=acp` is explicit.
- `@paperclipai/acpx-engine` is a new public package and needs npm
trusted-publishing bootstrap before release automation can publish it.
> For core feature work, check [`ROADMAP.md`](ROADMAP.md) first and
discuss it in `#dev` before opening the PR. Feature PRs that overlap
with planned core work may need to be redirected — check the roadmap
first. See `CONTRIBUTING.md`.
## Model Used
OpenAI GPT-5 via Codex coding agent. Exact hosted model build and
context-window size are not exposed in this runtime. Tool use included
shell execution, repository editing, GitHub CLI operations, and local
test/typecheck execution.
## Checklist
- [x] I have included a thinking path that traces from project context
to this change
- [x] I have specified the model used (with version and capability
details)
- [x] I have checked ROADMAP.md and confirmed this PR does not duplicate
planned core work
- [x] I have searched GitHub for duplicate or related PRs and linked
them above
- [x] I have either (a) linked existing issues with `Fixes: #` / `Closes
#` / `Refs #` OR (b) described the issue in-PR following the relevant
issue template
- [x] I have not referenced internal/instance-local Paperclip issues or
links (only public GitHub `#NNN` / `github.com/paperclipai/paperclip`
URLs)
- [x] My branch name describes the change (e.g. `docs/...`, `fix/...`)
and contains no internal Paperclip ticket id or instance-derived details
- [x] I have run tests locally and they pass
- [x] I have added or updated tests where applicable
- [x] I have updated relevant documentation to reflect my changes
- [x] I have considered and documented any risks above
- [x] All Paperclip CI gates are green
- [x] Greptile is 5/5 with no open P2s, recommendations, or follow-ups
- [x] I will address all Greptile and reviewer comments before
requesting merge
---------
Co-authored-by: Paperclip <noreply@paperclip.ing>
## Thinking Path
> - Paperclip is the open source app people use to manage AI agents for
work
> - Agents run in heartbeat runs orchestrated by the server; run
lifecycle, retry scheduling, and the dashboard's run-activity metrics
are the subsystems involved
> - A spike in "failed" tasks traced to three causes: server restarts
killing in-flight runs and mislabeling them as failures, deterministic
workspace-validation loops when a worktree's branch diverged, and
provider quota/usage-limit errors being classified as generic transient
failures (putting agents into error state and polluting metrics)
> - Killed-then-recovered runs and quota waits are not product failures,
so both the runtime behavior and the reporting needed to distinguish
them
> - This pull request drains runs gracefully on shutdown with idempotent
restart retries, self-heals workspace branch mismatches, adds a
quota-aware failure class with reset-time retry, separates recovered
restart kills from true failures on the dashboard, and documents restart
hygiene for operators
> - The benefit is fewer spurious failures, automatic recovery instead
of manual repair, and dashboard metrics that reflect real failure rates
## Linked Issues or Issue Description
No public GitHub issue exists; describing the bug inline per the
bug-report template:
**What happened?**
In-flight heartbeat runs are marked `failed` when the server restarts,
even though a retry later succeeds. Worktrees whose checked-out branch
diverges from the issue branch fail workspace validation on every
subsequent run with no recovery path. Provider quota/usage-limit
responses are treated as generic transient upstream errors, putting
agents into an error state and retrying before the quota window resets.
The dashboard counts all of these as true failures, inflating failure
metrics.
**Expected behavior**
Graceful shutdown should interrupt (not fail) running runs and chain
exactly one recovery retry. Workspace validation should repair
recoverable branch mismatches automatically. Quota errors should get
their own error class with the retry scheduled at the provider reset
time and the agent left idle. The dashboard should report recovered
restart kills separately from true failures.
**Steps to reproduce**
1. Start a heartbeat run, then restart the server (SIGTERM) while it is
in flight — the run lands as `failed` with a process-loss error code
even when its retry succeeds
2. Check out an issue whose worktree branch has diverged (e.g. after a
force-moved branch) — every subsequent run fails
`workspace_validation_failed` deterministically
3. Drive an agent into a provider usage-limit window — the run fails as
a generic transient upstream error and the agent enters an error state
instead of idling until the reset time
**Paperclip version or commit**
master (base c07e650cd)
**Deployment mode**
Self-hosted dev plane (Linux, node server + embedded Postgres)
## What Changed
- Graceful shutdown (SIGTERM hook) now marks in-flight runs
`interrupted` instead of `failed` and enqueues an idempotent
process-loss retry (pre-insert existence check on `retryOfRunId`
prevents duplicates; bursts chain exactly one retry per interrupted run)
- Run-liveness classification routes `interrupted` to `needs_followup`
rather than `failed`
- Workspace validation self-heals branch mismatch / missing-branch
states instead of failing deterministically on every run
- New `provider_quota` error class: session/usage-limit responses
schedule the retry at the provider reset time and leave the agent idle
(not errored); fixes a case where a quota-terminated run with subtype
`success` was misclassified as failed; HTTP 529 remains transient
- Dashboard run-activity query separates recovered restart kills from
true failures via a recursive CTE over `retry_of_run_id` (ancestors of a
succeeded retry count as recovered), adds a per-day failed-by-error-code
breakdown, and binds the window start as a timestamptz string
- Activity charts UI: amber "Recovered" segment with legend and per-day
error-code tooltip; success-rate chart counts recovered runs as
successes
- New ops runbook: `docs/deploy/dev-plane-restart-hygiene.md`
## Verification
- Greptile follow-up fixes on `992705edd`: `pnpm exec vitest run
server/src/__tests__/server-startup-feedback-export.test.ts
server/src/__tests__/heartbeat-workspace-branch-containment.test.ts
server/src/__tests__/heartbeat-workspace-session.test.ts
server/src/__tests__/heartbeat-process-recovery.test.ts
server/src/__tests__/heartbeat-workspace-finalize-branch.test.ts` (209
tests); `pnpm --filter @paperclipai/server typecheck`; `git diff
--check`
- Post-rebase CI fixes: `pnpm exec vitest run
server/src/__tests__/heartbeat-retry-scheduling.test.ts`; `pnpm exec
vitest run server/src/__tests__/heartbeat-retry-scheduling.test.ts
server/src/__tests__/workspace-runtime.test.ts
server/src/__tests__/heartbeat-workspace-branch-containment.test.ts
server/src/__tests__/heartbeat-workspace-finalize-branch.test.ts
server/src/__tests__/heartbeat-workspace-session.test.ts` (227 tests);
`pnpm --filter @paperclipai/server typecheck`
- `npm test` server suites covering the changes:
`heartbeat-process-recovery`, `heartbeat-stop-metadata`,
`heartbeat-retry-scheduling` (95 tests), quota parse +
execute/retry-scheduling suites (100 tests), workspace self-heal suites
(200 tests), dashboard run-activity tests (3 tests) — all green,
typecheck exit 0
- Dashboard CTE cross-checked against a real development database: two
restart-burst days moved from 17 to 8 and 19 to 6 true failures once
recovered kills were separated, matching manual retry-chain inspection
- Screenshot verification of the real ActivityCharts component
(recovered segment + tooltip) during QA
## Risks
- Behavioral shift: runs killed by a restart no longer surface as
`failed`; anyone consuming raw run statuses will see `interrupted` (new
status value) — dashboards/queries in this repo were updated accordingly
- Retry chaining on repeated restarts is bounded (one chained retry per
interruption) but a pathological restart loop still delays work rather
than failing it; the runbook covers operator hygiene for that case
- Dashboard query adds a recursive CTE; cost is bounded by the
day-window row count and was verified against production-sized data
- No schema migrations; low migration risk
## Model Used
- Claude (Anthropic) — claude-fable-5 via Claude Code / Paperclip agent
harness, extended thinking with tool use; implementation commits also
produced with Codex CLI (GPT-5 class) agents under the same harness
## Checklist
- [x] I have included a thinking path that traces from project context
to this change
- [x] I have specified the model used (with version and capability
details)
- [x] I have checked ROADMAP.md and confirmed this PR does not duplicate
planned core work
- [x] I have searched GitHub for duplicate or related PRs and linked
them above
- [x] I have either (a) linked existing issues with `Fixes: #` / `Closes
#` / `Refs #` OR (b) described the issue in-PR following the relevant
issue template
- [x] I have not referenced internal/instance-local Paperclip issues or
links (only public GitHub `#NNN` / `github.com/paperclipai/paperclip`
URLs)
- [ ] My branch name describes the change (e.g. `docs/...`, `fix/...`)
and contains no internal Paperclip ticket id or instance-derived details
- [x] I have run tests locally and they pass
- [x] I have added or updated tests where applicable
- [x] I have updated relevant documentation to reflect my changes
- [x] I have considered and documented any risks above
- [x] All Paperclip CI gates are green
- [x] Greptile is 5/5 with no open P2s, recommendations, or follow-ups
- [x] I will address all Greptile and reviewer comments before
requesting merge
---------
Co-authored-by: Paperclip <noreply@paperclip.ing>
## Thinking Path
> - Paperclip is the open source app people use to manage AI agents for
work.
> - Agent runs often need provider credentials, API tokens, and other
environment-bound secrets.
> - Company-level secrets work for shared credentials, but they do not
model values that should differ by human operator.
> - Without a user-scoped model, a run can dispatch without knowing
whether the responsible human has supplied the needed value.
> - Paperclip also needs run attribution to make those user-scoped
runtime checks deterministic and auditable.
> - This pull request adds user-specific secret definitions, per-user
values, environment bindings, responsible-user attribution, and runtime
resolution gates.
> - The benefit is that teams can define the secret once, let each user
provide their own value, and block runs before dispatch when required
user secrets or active definitions are unavailable.
## Linked Issues or Issue Description
Refs #224
Refs #6057
This PR implements user-specific secret support as a core
secret-management capability rather than a one-off adapter setting. It
is related to existing public work on company secrets UI and runtime
secret refs, but is distinct because the value is owned by the
responsible user and resolved at run dispatch time.
Related PR search before opening found existing secrets work such as
#1550, #8256, #8614, #8634, and #8647; none of those add the full
user-secret definition/value/runtime gate covered here.
## What Changed
- Added user-secret definitions and per-user "My secrets" values,
keeping stored values out of access metadata.
- Added `user_secret_ref` environment bindings and UI affordances to
pick them alongside existing secret refs.
- Added responsible-user runtime resolution so user-secret refs resolve
against the human responsible for the run.
- Added pre-dispatch missing-secret gates so runs fail before adapter
dispatch when required user values are absent or definitions are
inactive.
- Added low-trust allowlist hardening for user-secret runtime access.
- Added issue, routine, run, and agent API key responsible-user
attribution and fail-closed dispatch behavior when attribution cannot be
resolved.
- Added denial-copy mapping so responsible-user authorization failures
surface as actionable run outcomes instead of opaque setup failures.
- Added OpenAPI documentation for the user-secret routes.
- Rebases cleanly on current `master`; migrations were renumbered
incrementally as `0128_user_specific_secrets`,
`0129_agent_api_key_responsible_user`, and
`0130_run_responsible_user_invariant` after upstream `0126`/`0127`
migrations.
- Removed previously committed local design screenshots so the PR
contains code/docs/tests only.
## Verification
- PASS: PR head `2527febd106bcf3ca264ca0da7fca491084192d6` is based on
`paperclipai/paperclip:master`.
- PASS: `git diff --check`
- PASS: `git diff --name-only public/master...HEAD | rg
'^(pnpm-lock\\.yaml|\\.github/workflows/|screenshots/)' || true`
produced no files.
- PASS: migration journal audit confirmed unique indexes through `130`
with tail entries `0126_issue_comment_derived_attribution`,
`0127_environment_custom_images_instance_scoped`,
`0128_user_specific_secrets`, `0129_agent_api_key_responsible_user`, and
`0130_run_responsible_user_invariant`.
- PASS: `pnpm --filter @paperclipai/ui typecheck`
- PASS: `pnpm --filter @paperclipai/server typecheck`
- PASS: `pnpm --filter @paperclipai/server exec vitest run
src/__tests__/heartbeat-responsible-user-invariant.test.ts`
- PASS: `pnpm --filter @paperclipai/server exec vitest run
src/__tests__/heartbeat-active-run-output-watchdog.test.ts
src/__tests__/heartbeat-stale-queue-invalidation.test.ts
src/__tests__/heartbeat-workspace-finalize-branch.test.ts
src/__tests__/issue-monitor-scheduler.test.ts`
- PASS: `pnpm --filter @paperclipai/server exec vitest run
src/__tests__/heartbeat-comment-wake-batching.test.ts
src/__tests__/heartbeat-retry-scheduling.test.ts
src/__tests__/heartbeat-accepted-plan-workspace-refresh.test.ts
src/__tests__/heartbeat-plugin-environment.test.ts`
- PASS: `pnpm --filter @paperclipai/server exec vitest run
src/__tests__/low-trust-red-team-routes.test.ts`
- PASS: `pnpm --filter @paperclipai/server exec vitest run
src/__tests__/secrets-service.test.ts` (55 tests)
- PASS: `pnpm vitest run server/src/__tests__/secrets-routes.test.ts
server/src/__tests__/secrets-service.test.ts` (89 tests after final
Greptile cleanup fixes)
- PASS: `pnpm --filter @paperclipai/server exec vitest run
src/__tests__/heartbeat-issue-liveness-escalation.test.ts` (17 tests
after the final rebase CI fix)
- PASS: focused server Vitest batches covering heartbeat recovery,
project env, plugin env, routines, low-trust, pipelines, monitors,
watchdog, and stale queue paths.
- PASS: GitHub checks are green on
`2527febd106bcf3ca264ca0da7fca491084192d6`, including Typecheck +
Release Registry, Build, General tests, serialized server suites, e2e,
Canary Dry Run, verify, security checks, and Greptile Review.
- PASS: Greptile Review completed successfully on
`2527febd106bcf3ca264ca0da7fca491084192d6` with Confidence Score 5/5,
and GraphQL review-thread audit returned zero unresolved non-outdated
threads.
## Risks
- Runtime behavior now depends on a run having a correct responsible
user; missing or incorrect responsibility assignment can block runs
before adapter dispatch.
- `user_secret_ref` bindings intentionally expose metadata without
values, but UI/API callers may need to handle the new binding kind
explicitly.
- External secret providers and IAM policies are not automatically
provisioned by this PR; operators still need to configure provider-side
access for non-local vaults.
- The PR is broad across db/shared/server/UI/runtime paths, so release
validation should include both API and UI secret workflows before merge.
- The migration renumbering is intentionally incremental after upstream
migrations; the branch migrations use guarded
column/table/index/constraint creation so users who tested the older
draft numbering should not hit duplicate DDL for the existing objects.
> For core feature work, check [`ROADMAP.md`](ROADMAP.md) first and
discuss it in `#dev` before opening the PR. Feature PRs that overlap
with planned core work may need to be redirected — check the roadmap
first. See `CONTRIBUTING.md`.
## Model Used
OpenAI Codex, GPT-5-based coding agent (`gpt-5`), Codex local adapter
with shell/tool use and code execution. Context window and internal
reasoning mode are not exposed by the runtime.
## Checklist
- [x] I have included a thinking path that traces from project context
to this change
- [x] I have specified the model used (with version and capability
details)
- [x] I have checked ROADMAP.md and confirmed this PR does not duplicate
planned core work
- [x] I have searched GitHub for duplicate or related PRs and linked
them above
- [x] I have either (a) linked existing issues with `Fixes: #` / `Closes
#` / `Refs #` OR (b) described the issue in-PR following the relevant
issue template
- [x] I have not referenced internal/instance-local Paperclip issues or
links (only public GitHub `#NNN` / `github.com/paperclipai/paperclip`
URLs)
- [x] My branch name describes the change (e.g. `docs/...`, `fix/...`)
and contains no internal Paperclip ticket id or instance-derived details
- [x] I have run tests locally and they pass
- [x] I have added or updated tests where applicable
- [x] I have updated relevant documentation to reflect my changes
- [x] I have considered and documented any risks above
- [x] All Paperclip CI gates are green
- [x] Greptile is 5/5 with no open P2s, recommendations, or follow-ups
- [x] I will address all Greptile and reviewer comments before
requesting merge
---------
Co-authored-by: Paperclip <noreply@paperclip.ing>
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
## Thinking Path
> - Paperclip is the open source app people use to manage AI agents for
work
> - A core part of that experience is watching active agent runs without
dropping into raw logs first
> - Local and sandbox-backed adapters already record useful run output,
progress, and tool activity
> - But active issue threads could sit visually stale while the agent
was syncing workspaces, tailing sandbox output, or emitting incremental
tool-call updates
> - Operators need timely, human-readable progress while preserving the
raw transcript underneath
> - This pull request streams sandbox run-log progress into runtime
status, keeps visible issue threads refreshed, and folds repeated ACPX
tool updates into stable transcript cards
> - The benefit is that long-running agent work becomes easier to
supervise without changing the task/comment control-plane model
## Linked Issues or Issue Description
No public GitHub issue exists for this exact change.
Problem/motivation:
- During long-running sandboxed agent work, the issue UI can appear idle
even though the agent is actively syncing, running tools, or producing
incremental output.
- Operators need realtime feedback at the issue-thread layer, not only
after opening raw logs or waiting for the final heartbeat result.
- Related public context: #1808 previously added live-run status dots to
Projects; #4362 touches heartbeat wakeup behavior but is not a duplicate
of this runtime/UI feedback change.
## What Changed
- Added sandbox run-log streaming support and defaulted sandbox-capable
local adapters into the richer live-feedback path.
- Surfaced environment/sandbox sync progress through heartbeat runtime
status with bounded, redacted snippets.
- Added live issue-thread cache patching so visible active runs update
as progress events arrive.
- Folded repeated ACPX `tool_call` updates into one transcript card
instead of stacking duplicate cards.
- Updated adapter docs and added focused regression coverage for sandbox
log streaming, runtime status, ACPX parsing, live updates, transcript
rendering, and issue chat messages.
## Verification
- `pnpm install --frozen-lockfile`
- `pnpm exec vitest run ui/src/context/LiveUpdatesProvider.test.ts`
- `pnpm exec vitest run
server/src/services/heartbeat-run-runtime-status.test.ts
server/src/__tests__/heartbeat-runtime-state.test.ts
ui/src/context/LiveUpdatesProvider.test.ts`
- `pnpm exec vitest run
packages/adapter-utils/src/execution-target-sandbox.test.ts
packages/adapter-utils/src/sandbox-managed-runtime.test.ts
server/src/services/heartbeat-run-runtime-status.test.ts
server/src/__tests__/agent-live-run-routes.test.ts
server/src/__tests__/heartbeat-runtime-state.test.ts
packages/adapters/acpx-local/src/ui/parse-stdout.test.ts
ui/src/context/LiveUpdatesProvider.test.ts
ui/src/components/transcript/RunTranscriptView.test.tsx
ui/src/lib/issue-chat-messages.test.ts
ui/src/components/IssueChatThread.test.tsx`
- GitHub PR workflow on head `8397953e7b41ccd42e5d9457ee7e4dfb996e4ec5`:
`verify`, build, typecheck/release-registry, e2e, general shards,
serialized server shards, and canary dry run passed.
- Greptile Review on head `8397953e7b41ccd42e5d9457ee7e4dfb996e4ec5`:
Confidence Score 5/5, no unresolved review threads.
## Risks
- Live issue-thread cache patching could miss an edge case for a route
shape not covered by tests.
- Surfacing active-run snippets needs continued care around redaction;
this PR keeps snippets bounded and adds redaction-focused coverage.
- More frequent active-run UI refreshes could expose performance issues
on very large issue threads, though updates are scoped to visible
run/query caches.
## Model Used
OpenAI GPT-5 via Codex, operating as a tool-enabled coding agent with
shell, git, and repository-editing capabilities. Context window size is
not exposed in this runtime.
## Checklist
- [x] I have included a thinking path that traces from project context
to this change
- [x] I have specified the model used (with version and capability
details)
- [x] I have checked ROADMAP.md and confirmed this PR does not duplicate
planned core work
- [x] I have searched GitHub for duplicate or related PRs and linked
them above
- [x] I have either (a) linked existing issues with `Fixes: #` / `Closes
#` / `Refs #` OR (b) described the issue in-PR following the relevant
issue template
- [x] I have not referenced internal/instance-local Paperclip issues or
links (only public GitHub `#NNN` / `github.com/paperclipai/paperclip`
URLs)
- [x] My branch name describes the change (e.g. `docs/...`, `fix/...`)
and contains no internal Paperclip ticket id or instance-derived details
- [x] I have run tests locally and they pass
- [x] I have added or updated tests where applicable
- [x] I have updated relevant documentation to reflect my changes
- [x] I have considered and documented any risks above
- [x] All Paperclip CI gates are green
- [x] Greptile is 5/5 with no open P2s, recommendations, or follow-ups
- [x] I will address all Greptile and reviewer comments before
requesting merge
Co-authored-by: Paperclip <noreply@paperclip.ing>
## Thinking Path
> - Paperclip is the open source app people use to manage AI agents for
work.
> - Adapter names are part of the board-facing agent setup and
management experience.
> - The product now treats adapters as harnesses, while execution
environments are modeled separately.
> - Several built-in adapter labels still carried legacy local wording
from the older harness-by-environment model.
> - That wording makes the UI noisier and implies a distinction users no
longer need to reason about.
> - This pull request normalizes adapter display labels while keeping
persisted adapter type identifiers unchanged.
> - The benefit is clearer adapter selection and management copy without
a database migration.
## Linked Issues or Issue Description
No public GitHub issue was found for this exact cleanup.
Related public PRs:
- Supersedes #8910, an earlier branch for the same cleanup that did not
include the later docs/gateway/Cursor alignment.
- Refs #8819, which is related display-registry work for external
multi-segment adapter labels, but not a duplicate of this built-in label
cleanup.
Feature request details:
- Subsystem affected: Cross-cutting (`ui/`, `packages/adapters`, and
docs).
- Problem or motivation: user-facing adapter names include legacy local
qualifiers even though adapters map to harnesses and environments are
first-class elsewhere.
- Proposed solution: remove the legacy local wording from built-in
display labels, keep machine-readable adapter type ids unchanged, and
keep gateway disambiguation where it is useful.
- Alternatives considered: changing persisted adapter type ids was ruled
out because it would create migration and compatibility risk; one-off UI
replacements were ruled out because the display registry is already the
correct central label boundary.
- Roadmap alignment: this is small adapter UX polish, not a new
roadmap-level core feature.
## What Changed
- Updated the adapter display registry so known adapter labels are final
and no built-in local adapter renders a legacy local suffix.
- Preserved clean derived labels for unknown plugin local types while
keeping gateway disambiguation for unknown gateway types.
- Updated `AdapterManager` to prefer registry labels when the server
reports raw adapter type ids for built-ins.
- Removed legacy local wording from built-in adapter metadata labels in
UI and adapter packages.
- Aligned Cursor adapter metadata with the central display registry
label.
- Updated adapter docs and Storybook fixtures to match the new display
names.
- Added focused registry coverage for built-in labels and unknown plugin
suffix behavior.
## Verification
- `pnpm check:tokens`
- `git diff --check origin/master...fix/adapter-display-labels`
- Patch-addition scan for added secrets, private paths, and internal
links: no matches.
- GitHub duplicate search for open adapter-label/local-suffix issues and
PRs; #8910 was identified as the older superseded public PR.
- `pnpm exec vitest run
ui/src/adapters/adapter-display-registry.test.ts`
- `pnpm --filter @paperclipai/ui typecheck`
- Stale-label scan found no remaining user-facing display-label
suffixes; remaining local wording is operational/test terminology such
as adapter ids, docs about running locally, and test descriptions.
## Risks
Low risk. The change is display-label and documentation focused, and
adapter type ids remain unchanged. The main risk is ambiguous gateway
naming, mitigated by keeping explicit gateway labels where variants need
disambiguation.
## Model Used
OpenAI GPT-5 via Codex, tool-enabled coding agent in a local repository
workspace. Context window size is not exposed by this environment.
## Checklist
- [x] I have included a thinking path that traces from project context
to this change
- [x] I have specified the model used (with version and capability
details)
- [x] I have checked ROADMAP.md and confirmed this PR does not duplicate
planned core work
- [x] I have searched GitHub for duplicate or related PRs and linked
them above
- [x] I have either (a) linked existing issues with `Fixes: #` / `Closes
#` / `Refs #` OR (b) described the issue in-PR following the relevant
issue template
- [x] I have not referenced internal/instance-local Paperclip issues or
links (only public GitHub `#NNN` / `github.com/paperclipai/paperclip`
URLs)
- [x] My branch name describes the change (e.g. `docs/...`, `fix/...`)
and contains no internal Paperclip ticket id or instance-derived details
- [x] I have run tests locally and they pass
- [x] I have added or updated tests where applicable
- [x] I have updated relevant documentation to reflect my changes
- [x] I have considered and documented any risks above
- [x] All Paperclip CI gates are green
- [x] Greptile is 5/5 with no open P2s, recommendations, or follow-ups
- [x] I will address all Greptile and reviewer comments before
requesting merge
Co-authored-by: Paperclip <noreply@paperclip.ing>
## Thinking Path
> - Paperclip is the open source app people use to manage AI agents for
work.
> - A growing part of that work runs in sandboxed environments rather
than on the operator's local machine.
> - Today sandbox providers can start fresh workspaces and run probes,
but they do not have a shared contract for capturing and reusing
prepared sandbox state.
> - Operators need a way to set up tools, credentials, and project
dependencies once, then reuse that prepared image for later agent runs.
> - This pull request adds reusable sandbox custom images across the
provider contract, server runtime, and board UI.
> - It also keeps probes and sandbox copy flows aligned with
pre-authenticated/custom-image environments.
> - The benefit is faster, more reliable sandbox runs without repeatedly
rebuilding the same environment setup.
## Linked Issues or Issue Description
No public GitHub issue was found for this change. Inline feature request
follows.
### Problem or motivation
Sandboxed agents need reusable prepared runtime state so repeated runs
do not require manual setup every time. Operators often need system
packages, CLIs, SDKs, dependency caches, credentials, and project
tooling available before an agent can work productively.
### Proposed solution
Add a provider-level custom-image capability, server-side setup/capture
lifecycle, Daytona/fake provider support, and board UI controls for
creating, testing, selecting, and deleting custom images.
### Alternatives considered
Leaving this as provider-specific setup outside Paperclip would keep the
control plane blind to image state and would not give agents consistent
environment metadata. Re-running setup commands for every lease is
simpler, but slower and less reliable for interactive or credentialed
setup.
### Roadmap alignment
Checked `ROADMAP.md`; this aligns with the Cloud / Sandbox agents
roadmap area and does not duplicate any related public issue or PR found
by search.
Additional context:
- Subsystem affected: cross-cutting (`packages/db`, `packages/shared`,
`packages/plugins`, `server`, `ui`).
- Duplicate search: searched GitHub for `sandbox custom image` and
`sandbox template environment`; no related public issues or PRs were
found.
## What Changed
- Added custom-image shared types, validators, constants, API paths, and
database schema/migration.
- Added server services/routes for custom-image templates and setup
sessions, including runtime cleanup and provider metadata handling.
- Extended plugin/sandbox provider capabilities for interactive setup,
template capture, and template deletion.
- Implemented custom-image support in the fake sandbox provider and
Daytona provider.
- Updated environment runtime/config handling so active custom images
flow into leases, probes, and agent execution.
- Added board UI controls and API client support for custom-image setup,
capture, selection, status, and error states.
- Hardened sandbox copy/probe behavior for insecure clipboard contexts
and pre-authenticated sandbox images.
- Added targeted coverage across shared validators, DB schema, server
routes/services, provider plugins, adapter probes, and UI flows.
## Verification
- `pnpm install --frozen-lockfile --ignore-scripts`
- `pnpm vitest run
packages/adapters/claude-local/src/server/test.probe.test.ts
packages/adapters/claude-local/src/server/test.ts
packages/adapters/codex-local/src/server/test.remote.test.ts
packages/adapters/codex-local/src/server/test.ts`
- `pnpm --filter @paperclipai/adapter-claude-local typecheck`
- `pnpm --filter @paperclipai/adapter-codex-local typecheck`
- `pnpm vitest run
packages/db/src/environment-custom-images-schema.test.ts
packages/shared/src/environment-custom-images.test.ts
packages/shared/src/validators/plugin.test.ts
server/src/__tests__/environment-custom-images-service.test.ts
server/src/__tests__/workspace-runtime.test.ts
packages/plugins/sandbox-providers/daytona/src/plugin.test.ts
ui/src/pages/CompanyEnvironments.test.tsx
ui/src/pages/CompanySettings.test.tsx`
- `pnpm -r typecheck`
- `pnpm build`
- `rm -rf packages/db/dist && pnpm test:run`
- Public-safety scan of the final diff found no internal Paperclip issue
links, private instance URLs, or real secret patterns.
## Risks
- Adds a database migration and new environment runtime tables, so
migration ordering and rollback need care.
- Provider implementations may differ in how reliably they can
capture/delete images; unsupported providers surface capability-gated UI
states.
- Custom-image state can contain operator-prepared tooling and
credentials inside the provider image, so providers must enforce their
own access controls and cleanup semantics.
- Broad surface area across shared contracts, server runtime, plugins,
adapters, and UI means CI and Greptile review should be watched closely.
> For core feature work, check [`ROADMAP.md`](ROADMAP.md) first and
discuss it in `#dev` before opening the PR. Feature PRs that overlap
with planned core work may need to be redirected — check the roadmap
first. See `CONTRIBUTING.md`.
## Model Used
OpenAI GPT-5 (`gpt-5`) via Codex CLI with tool use and code execution.
Assisted with branch cleanup, conflict resolution, local verification,
and PR preparation. Earlier branch implementation work was assisted by
Paperclip-managed Claude/Codex agents.
## Checklist
- [x] I have included a thinking path that traces from project context
to this change
- [x] I have specified the model used (with version and capability
details)
- [x] I have checked ROADMAP.md and confirmed this PR does not duplicate
planned core work
- [x] I have searched GitHub for duplicate or related PRs and linked
them above
- [x] I have either (a) linked existing issues with `Fixes: #` / `Closes
#` / `Refs #` OR (b) described the issue in-PR following the relevant
issue template
- [x] I have not referenced internal/instance-local Paperclip issues or
links (only public GitHub `#NNN` / `github.com/paperclipai/paperclip`
URLs)
- [x] My branch name describes the change (e.g. `docs/...`, `fix/...`)
and contains no internal Paperclip ticket id or instance-derived details
- [x] I have run tests locally and they pass
- [x] I have added or updated tests where applicable
- [x] I have updated relevant documentation to reflect my changes
- [x] I have considered and documented any risks above
- [ ] All Paperclip CI gates are green
- [ ] Greptile is 5/5 with no open P2s, recommendations, or follow-ups
- [x] I will address all Greptile and reviewer comments before
requesting merge
---------
Co-authored-by: Paperclip <noreply@paperclip.ing>
## Thinking Path
> - Paperclip is the open source app people use to manage AI agents for
work.
> - Local adapters are the bridge between Paperclip's control plane and
provider CLIs such as Claude Code and Codex.
> - Those adapters can run either on the host machine or inside a
remote/sandbox execution target.
> - Sandbox probes need to validate the same auth/config path that real
sandbox execution will use.
> - The previous probe paths could surface misleading Claude errors,
rely on host-only Codex state, or upload far more Codex home state than
the probe needed.
> - This pull request fixes the Claude and Codex sandbox probe/runtime
behavior together while keeping provider-specific sandbox image work out
of scope.
> - The benefit is faster, clearer adapter health checks that better
match real sandbox execution.
## Linked Issues or Issue Description
No public GitHub issue was found for this exact bug during duplicate
search.
Bug report:
**What happened?**
Sandboxed Claude/Codex adapter tests could diverge from real runtime
auth/config behavior. Claude sandbox probes could show the leading
stream init line instead of the real final error, and Codex sandbox
probes could upload full managed home state or mask a sandbox-local
login with an empty uploaded `CODEX_HOME`.
**Expected behavior**
Sandbox probes should exercise the remote runtime contract, preserve
useful sandbox credentials, avoid relying on unrelated host state, and
report actionable probe failures.
**Steps to reproduce**
1. Configure a remote/sandbox execution target for `claude_local` or
`codex_local`.
2. Run the environment Test/probe path where host credentials differ
from the sandbox's runtime credentials or the managed Codex home
contains session history.
3. Observe that probe behavior can differ from the actual sandbox
runtime path or surface an unhelpful Claude stream initialization line.
**Paperclip version or commit**
Current `master` before this PR, based on `4a2447da3`.
**Deployment mode**
Local development/control-plane deployment with remote sandbox execution
targets.
Related search performed:
- Public issues: `Claude sandbox probe`, `Codex CODEX_HOME sandbox`
returned no matches.
- Public PRs: `Claude Codex sandbox probe`, `codex home sandbox`,
`claude auth sandbox` returned no matches.
## What Changed
- Made Claude sandbox Test probes materialize the same Paperclip-managed
Claude config seed path used by sandbox execution.
- Preserved sandbox-local Claude credentials when materializing remote
Claude config and expanded auth-required detection for `/login` API-key
failures.
- Improved Claude hello-probe diagnostics so the final result/error is
surfaced instead of the unhelpful stream init event, with transient
upstream failures downgraded to warnings.
- Changed Codex probe behavior to upload only minimal auth/config files
instead of the full managed `CODEX_HOME`.
- Let Codex sandbox probes leave `CODEX_HOME` unset when the host has no
credentials, so pre-authenticated sandbox images can be tested directly.
- Excluded bulky host-local Codex session/shell state from sandbox
runtime home uploads.
- Switched the Codex local default model away from the
ChatGPT-unsupported `gpt-5.3-codex` option.
- Added regression coverage for Claude parsing/probe paths, Codex
adapter metadata/argument/probe behavior, and server-level Claude
sandbox environment behavior.
## Verification
Passed locally:
- `pnpm install --frozen-lockfile`
- `pnpm vitest run
packages/adapters/claude-local/src/server/parse.test.ts
packages/adapters/claude-local/src/server/test.probe.test.ts
server/src/__tests__/claude-local-adapter-environment.test.ts`
- `pnpm vitest run packages/adapters/codex-local/src/index.test.ts
packages/adapters/codex-local/src/server/codex-args.test.ts
packages/adapters/codex-local/src/server/test.remote.test.ts`
- `pnpm --filter @paperclipai/adapter-claude-local typecheck`
- `pnpm --filter @paperclipai/adapter-codex-local typecheck`
- `pnpm --filter @paperclipai/server typecheck`
- `git diff --check`
## Risks
- Adapter configuration behavior is sensitive to local vs sandboxed
execution mode, so review should focus on environment detection,
argument construction, and any state written during probe/test runs.
- The Codex default-model change may affect newly created agents that
rely on the adapter default instead of an explicit model.
- Excluding Codex session/shell state from sandbox uploads should be
safe for fresh sandbox runs, but reviewers should confirm no runtime
resume path depends on that host-local state.
- Provider-specific setup/capture behavior is intentionally left to
separate work.
## Model Used
OpenAI GPT-5 Codex via Paperclip `codex_local`; tool-enabled local
coding session with terminal access. Context window size was not exposed
by the runtime.
## Checklist
- [x] I have included a thinking path that traces from project context
to this change
- [x] I have specified the model used (with version and capability
details)
- [x] I have checked ROADMAP.md and confirmed this PR does not duplicate
planned core work
- [x] I have searched GitHub for duplicate or related PRs and linked
them above
- [x] I have either (a) linked existing issues with `Fixes: #` / `Closes
#` / `Refs #` OR (b) described the issue in-PR following the relevant
issue template
- [x] I have not referenced internal/instance-local Paperclip issues or
links (only public GitHub `#NNN` / `github.com/paperclipai/paperclip`
URLs)
- [x] My branch name describes the change (e.g. `docs/...`, `fix/...`)
and contains no internal Paperclip ticket id or instance-derived details
- [x] I have run tests locally and they pass
- [x] I have added or updated tests where applicable
- [x] I have updated relevant documentation to reflect my changes
- [x] I have considered and documented any risks above
- [x] All Paperclip CI gates are green
- [x] Greptile is 5/5 with no open P2s, recommendations, or follow-ups
- [x] I will address all Greptile and reviewer comments before
requesting merge
---------
Co-authored-by: Paperclip <noreply@paperclip.ing>
## Thinking Path
> - Paperclip is the open source app people use to manage AI agents for
work.
> - Sandboxed agent runs can spend meaningful time preparing a remote
workspace before the agent transcript shows useful output.
> - Operators need short, current progress text for those setup phases,
but that text should not become durable run history.
> - The existing live-run websocket path already carries run updates to
the UI, so the backend can reuse that channel instead of adding polling.
> - This pull request adds an ephemeral runtime-progress contract, a
process-local status store, and heartbeat integration for
sandbox-managed runs.
> - The benefit is a clearer active-run experience without database
migrations or persistent progress rows.
## Linked Issues or Issue Description
Refs #248
No exact public GitHub issue was found for this status-message plumbing.
The underlying problem is that active sandboxed runs currently have
setup phases, such as workspace sync and restore, where the operator
cannot see concise current progress through the live run state. This PR
addresses that gap for the backend/runtime layer while keeping progress
messages ephemeral.
GitHub search performed for related or duplicate work: `sandbox runtime
status`, `sandbox restore index`, and `runtime progress`. No direct
duplicate PR was found.
## What Changed
- Added shared runtime-progress types and the `heartbeat.run.progress`
live event type.
- Added a process-local heartbeat run runtime-status store with TTL,
bounded/redacted messages, and terminal cleanup.
- Threaded runtime progress callbacks through heartbeat execution and
active/live run serialization.
- Emitted sandbox-managed runtime phase updates for sync, adapter
startup, restore/export, and finalization paths.
- Added backend and adapter-utils tests for ephemeral status behavior,
terminal cleanup, live serialization, and sandbox progress callbacks.
## Verification
- `pnpm install --frozen-lockfile`
- Local PII scan before push: high-confidence secret patterns, internal
issue links, local user paths, and private URL patterns checked across
all three split diffs; no real secrets or internal links found. The only
secret-like text is an intentional fake test fixture (`sk-test-secret`).
- `git diff --check origin/master..feat/sandbox-runtime-status`
- `pnpm exec vitest run
server/src/services/heartbeat-run-runtime-status.test.ts
server/src/__tests__/heartbeat-runtime-state.test.ts
server/src/__tests__/agent-live-run-routes.test.ts
packages/adapter-utils/src/sandbox-managed-runtime.test.ts` — 4 files,
23 tests passed.
- `pnpm run typecheck` passed on both top stacks that include this
branch: `feat/sandbox-status-ui` and `fix/sandbox-restore-index-sync`.
- `pnpm run build` passed on both top stacks that include this branch;
Vite reported existing CSS `::highlight` and chunk-size warnings.
- `pnpm run test:run` was attempted on `fix/sandbox-restore-index-sync`;
it failed in two unrelated broad-suite tests. One depends on this host's
Git default branch behavior, and one depends on local Claude
model-discovery environment. The changed focused suites above pass.
## Risks
- Runtime progress is process-local by design, so status disappears
after TTL, terminal cleanup, or server restart.
- Clients that do not consume `heartbeat.run.progress` simply keep
existing behavior.
- Message redaction is intentionally generic; overly specific phase
details should stay out of runtime-progress payloads.
> For core feature work, check [`ROADMAP.md`](ROADMAP.md) first and
discuss it in `#dev` before opening the PR. Feature PRs that overlap
with planned core work may need to be redirected — check the roadmap
first. See `CONTRIBUTING.md`.
## Model Used
OpenAI GPT-5 via Codex coding agent, with shell/tool execution in a
local worktree. Exact context-window metadata is not exposed by the
runtime.
## Checklist
- [x] I have included a thinking path that traces from project context
to this change
- [x] I have specified the model used (with version and capability
details)
- [x] I have checked ROADMAP.md and confirmed this PR does not duplicate
planned core work
- [x] I have searched GitHub for duplicate or related PRs and linked
them above
- [x] I have either (a) linked existing issues with `Fixes: #` / `Closes
#` / `Refs #` OR (b) described the issue in-PR following the relevant
issue template
- [x] I have not referenced internal/instance-local Paperclip issues or
links (only public GitHub `#NNN` / `github.com/paperclipai/paperclip`
URLs)
- [x] My branch name describes the change (e.g. `docs/...`, `fix/...`)
and contains no internal Paperclip ticket id or instance-derived details
- [x] I have run tests locally and they pass
- [x] I have added or updated tests where applicable
- [x] I have updated relevant documentation to reflect my changes
- [x] I have considered and documented any risks above
- [x] All Paperclip CI gates are green
- [x] Greptile is 5/5 with no open P2s, recommendations, or follow-ups
- [x] I will address all Greptile and reviewer comments before
requesting merge
---------
Co-authored-by: Paperclip CTO <cto@paperclip.local>
Co-authored-by: Paperclip CTO <noreply@paperclip.ing>
## Thinking Path
> - Paperclip is the open source app people use to manage AI agents for
work.
> - Local adapters can run agents against sandboxed execution targets by
syncing the workspace and selected runtime assets into the sandbox.
> - The Codex local adapter includes a managed Codex home asset so
sandboxed Codex runs can use the expected auth, config, skills, and
session state.
> - That asset follows symlinks, which is useful for real Codex home
content but unsafe for transient launcher directories.
> - Transient `tmp` and `.tmp` directories can contain symlinks to large
host binaries, so the sandbox archive can inline large executable
targets instead of just the small home directory content.
> - This pull request excludes transient Codex home directories from the
sandbox home asset while preserving the required Codex home files.
> - The benefit is a much smaller and more predictable sandbox setup
upload without changing the runtime files Codex actually needs.
## Linked Issues or Issue Description
No public issue was found for this exact sandbox archive-size bug.
Bug description:
- What happened: sandboxed `codex_local` runs sync the managed Codex
home as a `home` asset with `followSymlinks` enabled. If transient Codex
home dirs such as `tmp` or `.tmp` contain symlinks to a large host
binary, the archive can inline that binary and make `Syncing home to
sandbox` much larger than the managed home directory itself.
- Expected behavior: sandbox setup should include the Codex home files
needed for auth, config, skills, and session continuity, but should not
archive transient launcher scratch directories.
- Reproduction shape: create a managed Codex home with normal
auth/config/skills files and a `tmp/arg0` or `.tmp` symlink to a large
host executable, then start a sandboxed `codex_local` run. The home
asset archive grows by the symlink target size.
- Version/commit: observed on local `master` before this change.
- Related public context: #5028 covers a different managed Codex home
reliability issue around stale auth files; this PR addresses sandbox
archive bloat from transient symlink targets.
## What Changed
- Excluded `tmp` and `.tmp` from the Codex `home` asset that is uploaded
for sandboxed runs.
- Added regression coverage proving transient symlinked home dirs are
excluded from the tar while required auth/config/skills files remain
included.
- Kept `followSymlinks` behavior for the rest of the Codex home asset so
existing non-transient symlink behavior is preserved.
## Verification
- `git diff --check`
- Local PII/secret pattern scan over the committed diff
- `pnpm exec vitest run
packages/adapter-utils/src/sandbox-managed-runtime.test.ts`
- `pnpm --filter @paperclipai/adapter-utils typecheck`
- `pnpm --filter @paperclipai/adapter-codex-local typecheck`
## Risks
Low risk. The exclusion is limited to transient Codex home scratch
directories, and the regression test verifies the files needed in the
sandbox are still archived. The main compatibility risk is if a user
intentionally placed required persistent Codex state under `tmp` or
`.tmp`; those paths are treated as volatile scratch space by this
change.
## Model Used
OpenAI Codex coding agent based on GPT-5, with shell, git, and GitHub
CLI tool use. Exact hosted model build and context-window size were not
exposed in the local adapter runtime.
## Checklist
- [x] I have included a thinking path that traces from project context
to this change
- [x] I have specified the model used (with version and capability
details)
- [x] I have checked ROADMAP.md and confirmed this PR does not duplicate
planned core work
- [x] I have searched GitHub for duplicate or related PRs and linked
them above
- [x] I have either (a) linked existing issues with `Fixes: #` / `Closes
#` / `Refs #` OR (b) described the issue in-PR following the relevant
issue template
- [x] I have not referenced internal/instance-local Paperclip issues or
links (only public GitHub `#NNN` / `github.com/paperclipai/paperclip`
URLs)
- [x] My branch name describes the change (e.g. `docs/...`, `fix/...`)
and contains no internal Paperclip ticket id or instance-derived details
- [x] I have run tests locally and they pass
- [x] I have added or updated tests where applicable
- [x] I have updated relevant documentation to reflect my changes
- [x] I have considered and documented any risks above
- [x] All Paperclip CI gates are green
- [x] Greptile is 5/5 with no open P2s, recommendations, or follow-ups
- [x] I will address all Greptile and reviewer comments before
requesting merge
Co-authored-by: Paperclip <noreply@paperclip.ing>
## Thinking Path
> - Paperclip orchestrates AI agents for zero-human companies
> - The `codex_local` adapter runs `codex exec` as a child process and
streams JSONL events from its stdout
> - NEE-79 caught a real-world `codex_local` orphan: codex sat in
`read()` on stdin for 1h+, no JSON events emitted past startup, no LLM
call in flight. The adapter had no inactivity timer; the only safety net
was the platform-level 1h silent-run detector
> - This is precisely the failure shape NEE-80 scoped: an
adapter-detected fault that should be killed *by the adapter* and
surfaced as `failed`, not waited out for an hour
> - This pull request adds an output-inactivity watchdog inside the
codex-local adapter that resets on every parsed JSONL event from stdout,
kills the child via SIGTERM → 5s grace → SIGKILL when it fires, and
resolves the run with a structured watchdog failure
> - The benefit is that NEE-79-class hangs shorter than 1h stop reaching
the platform-level safety net — they fail fast and visibly at the
adapter layer, with diagnostic logs that don't require host shell access
## Linked Issues or Issue Description
No existing GitHub issue covers this; describing the underlying bug
in-PR per the bug report template
(`.github/ISSUE_TEMPLATE/bug_report.yml`):
- **What happened?** A `codex_local` run sat in `read()` on stdin for
over an hour: codex emitted its startup JSONL events, then nothing — no
further events, no LLM call in flight, no exit. The adapter kept the
child alive indefinitely; the run was only reaped by the platform-level
1h silent-run safety net, an hour after it had effectively died.
- **Expected behavior:** The adapter should detect that its child has
stopped producing output long before the platform-level safety net, kill
it, and surface the run as `failed` with a diagnostic that explains what
happened.
- **Steps to reproduce:** Run any `codex_local` issue where `codex exec`
hangs after startup (e.g. codex blocks reading stdin and never emits
another JSONL event). Observe the run stays alive until the 1h platform
safety net fires.
- **Paperclip version or commit:** reproduced on master prior to this
branch.
- **Agent adapter(s) involved:** codex_local.
Related PRs found while searching for duplicates (none implement an
event-aware inactivity watchdog inside `codex_local`):
- #4004 — generic idle + wall watchdogs for `runChildProcess` in
adapter-utils; complementary, operates below the JSONL parse layer and
is not codex-event-aware
- #4742 — fail-fast on codex-local *startup* hang; this PR covers the
post-startup hang class
- #6861 — zombie-run termination for codex-local; reaping after exit,
not inactivity detection
- #7811 — the analogous output-idle timeout for grok-local
## What Changed
- `packages/adapters/codex-local/src/server/watchdog.ts` *(new)* — pure
watchdog primitive with injectable timers/clock:
`resolveCodexInactivityTimeout`, `createCodexInactivityWatchdog`,
`formatWatchdogErrorMessage`. Default `7 * 60_000` ms; honors `null` as
the disabled escape hatch
- `packages/adapters/codex-local/src/server/execute.ts` — `runAttempt`
now wraps `onSpawn` to capture `pid`/`processGroupId`, feeds stdout
chunks through `noteStdoutChunk`, and on watchdog fire sends SIGTERM to
the process group, schedules SIGKILL after 5s, and returns an
`AdapterExecutionResult` with `exitCode: null`, `signal:
SIGTERM|SIGKILL`, `errorMessage: "watchdog: no codex output for {N}m
{S}s"`, `errorCode: "codex_output_inactivity_watchdog"`. With
`errorMessage` set and `timedOut: false`, `heartbeat.ts:5860` maps the
run to `outcome === "failed"` (not `cancelled`)
- `packages/adapters/codex-local/src/index.ts` — `agentConfigurationDoc`
documents `outputInactivityTimeoutMs` (number ms; `null` disables;
non-positive falls back to default with a warning log at spawn)
- `packages/adapters/codex-local/src/server/watchdog.test.ts` *(new)* —
13 tests covering acceptance criteria 2 and 3 plus supporting cases:
fires after silence, no-fire across 12× (threshold − 1s) cycles,
multi-event chunks, non-JSON ignoring, single-fire idempotency,
formatter shape, full resolution table, `null` → disabled
-
`packages/adapters/codex-local/src/server/watchdog.integration.test.ts`
*(new)* — real Node subprocess that prints one JSONL event then sleeps;
`runChildProcess` reaps it within `threshold + 6s`; signal is SIGTERM or
SIGKILL; `parsedEventCount === 1` (acceptance criteria 1 and 4)
## Verification
```
$ pnpm --filter @paperclipai/adapter-codex-local typecheck
> tsc --noEmit
# clean
$ pnpm --filter @paperclipai/adapter-codex-local exec vitest run
✓ src/server/quota-spawn-error.test.ts (1 test)
✓ src/server/codex-home.test.ts (3 tests)
✓ src/server/codex-args.test.ts (3 tests)
✓ src/server/watchdog.test.ts (13 tests)
✓ src/server/parse.test.ts (9 tests)
✓ src/server/execute.remote.test.ts (4 tests)
✓ src/ui/parse-stdout.test.ts (3 tests)
✓ src/ui/build-config.test.ts (1 test)
✓ src/server/watchdog.integration.test.ts (1 test)
Test Files 9 passed (9)
Tests 38 passed (38)
```
Acceptance criteria check:
1. ✅ Simulated child emits one event then sleeps → killed at threshold;
result `errorMessage` matches `watchdog: no codex output for {N}m {S}s`
(`watchdog.integration.test.ts`)
2. ✅ Child emits events every (threshold − 1s) → not killed
(`watchdog.test.ts: "does not fire when events arrive every (threshold -
1s)"`)
3. ✅ `outputInactivityTimeoutMs: null` disables the watchdog
(`resolveCodexInactivityTimeout` returns `disabled`; `execute.ts` skips
watchdog construction and logs a startup warning)
4. ✅ Real subprocess reaped well within `threshold + 6s` — 250 ms
threshold, 290 ms wall clock in CI
5. ✅ Adapter-level fault → `outcome === "failed"` per
`heartbeat.ts:5860`. NEE-79-class hangs <1h get caught at the adapter,
not the platform-level safety net
Post-rebase verification (head `2a8703fab`, rebased onto master
`69a368ed5`):
```
$ pnpm --filter @paperclipai/adapter-codex-local typecheck # clean
$ pnpm --filter @paperclipai/adapter-codex-local exec vitest run
Test Files 11 passed (11)
Tests 64 passed (64)
```
## Risks
- Low risk. Behind a default-on watchdog that only fires after 7m of
zero parsed JSON events. Operators can disable it with
`outputInactivityTimeoutMs: null` for known-slow tasks
- The kill path reuses the same `process.kill(-pgid, signal)` pattern
that `runChildProcess` already uses for its terminal-result cleanup, so
signal semantics match the existing code path
- `timedOut: false` is preserved on watchdog fire — the platform-level
timeout outcome is unchanged, only the `failed`-vs-success
classification flips. No behavioral shift for already-failing runs
- Sandbox/SSH execution targets: the watchdog fires and emits the
structured log, but the kill is best-effort because remote pids aren't
owned by this process. The platform-level 1h safety net still applies.
Out of scope for NEE-81 by design
## Model Used
- Provider: Anthropic Claude
- Model: claude-opus-4-7
- Mode: Claude Code (extended thinking, tool use)
## Checklist
- [x] I have included a thinking path that traces from project context
to this change
- [x] I have specified the model used (with version and capability
details)
- [x] I have checked ROADMAP.md and confirmed this PR does not duplicate
planned core work
- [x] I have searched GitHub for duplicate or related PRs and linked
them above
- [x] I have either (a) linked existing issues with `Fixes: #` / `Closes
#` / `Refs #` OR (b) described the issue in-PR following the relevant
issue template
- [x] I have run tests locally and they pass
- [x] I have added or updated tests where applicable
- [ ] If this change affects the UI, I have included before/after
screenshots (N/A — no UI changes)
- [x] I have updated relevant documentation to reflect my changes
(`agentConfigurationDoc` for `outputInactivityTimeoutMs`)
- [x] I have considered and documented any risks above
- [ ] All Paperclip CI gates are green (re-running on the rebased head)
- [x] Greptile is 5/5 with no open P2s, recommendations, or follow-ups
(P2 addressed; review threads resolved)
- [x] I will address all Greptile and reviewer comments before
requesting merge
---------
Co-authored-by: Neeraj Kumar Singh <b.nirajkumarsingh@hotmail.com>
Co-authored-by: Devin Foley <devin@paperclip.ing>
Co-authored-by: Paperclip <noreply@paperclip.ing>
## Thinking Path
> - Paperclip is the open source app people use to manage AI agents for
work
> - The `codex_local` adapter isolates local Codex runs by assigning
managed `CODEX_HOME` state per company and per agent
> - PR #8272 tightened that isolation, but it left a gap: once
`CODEX_HOME` became explicit, the adapter treated it like a user-managed
override and skipped auth seeding
> - That meant newly isolated agents could launch with no usable
`auth.json`, hit OpenAI unauthenticated, and fail with `401 Missing
bearer`
> - Users who had already persisted one of those broken managed homes
could remain stranded even after config changes unless Paperclip
repaired the home itself
> - This pull request teaches Paperclip to seed managed homes correctly,
backfill already-stranded managed homes on startup, and reject
credential-less managed homes before they reach the provider
> - The benefit is that affected managed `codex_local` agents recover
automatically after upgrade and restart, without manual `CODEX_HOME`
surgery
## Linked Issues or Issue Description
- Fixes#497
- Refs #5028
- Related PR: #8272
- Related PR: #8399
## What Changed
- Distinguished Paperclip-managed `CODEX_HOME` paths from genuine
external overrides and always seeded auth into managed homes, even when
`CODEX_HOME` is explicit in config.
- Wrote API-key-backed `auth.json` files for managed homes when
`OPENAI_API_KEY` is configured, otherwise symlinked the shared Codex
auth for subscription/OAuth flows.
- Added a startup reconciliation pass that backfills already-isolated
managed homes created by the broken release so upgrade plus restart
repairs stranded agents automatically.
- Preserved previously resolved API-key auth when the stored
`OPENAI_API_KEY` binding is secret-backed and startup cannot resolve the
secret value directly.
- Hardened the managed-home preflight to require a credential-bearing
`auth.json`, not just file presence, and documented the recovery
behavior.
- Added regression tests covering managed-home seeding, fail-fast
behavior, and server-side startup reconciliation.
## Verification
```bash
pnpm exec vitest run packages/adapters/codex-local/src/server/codex-home.test.ts packages/adapters/codex-local/src/server/execute.auth.test.ts server/src/__tests__/codex-auth-reconciliation.test.ts server/src/__tests__/codex-local-execute.test.ts server/src/__tests__/server-startup-feedback-export.test.ts
pnpm --filter @paperclipai/adapter-codex-local typecheck
pnpm --filter @paperclipai/server typecheck
pnpm check:tokens
pnpm exec vitest run server/src/__tests__/server-startup-feedback-export.test.ts
```
GitHub Actions `PR` workflow is green on latest head `e5b1e08d4`,
including policy, typecheck, test shards, build, e2e, serialized server
suites, and canary dry run. Greptile Review is green on latest head with
0 comments added. No UI changes.
## Risks
- Startup reconciliation now mutates persisted managed Codex homes at
boot. Risk is low because it only touches Paperclip-managed
company/agent home paths and no-ops when a home already has usable auth.
- Genuine external `CODEX_HOME` overrides remain intentionally
self-managed, so those users still own repair steps inside their custom
home.
- Hosts with neither shared Codex auth nor an explicit per-agent API key
now fail earlier with a clearer adapter error instead of surfacing a
downstream `401`, which changes timing but not capability.
## Model Used
- OpenAI Codex, GPT-5-based coding agent in a local Codex session; exact
served model ID/context window were not exposed to the session. Tool use
and code execution were enabled.
## Checklist
- [x] I have included a thinking path that traces from project context
to this change
- [x] I have specified the model used (with version and capability
details)
- [x] I have checked ROADMAP.md and confirmed this PR does not duplicate
planned core work
- [x] I have searched GitHub for duplicate or related PRs and linked
them above
- [x] I have either (a) linked existing issues with `Fixes: #` / `Closes
#` / `Refs #` OR (b) described the issue in-PR following the relevant
issue template
- [x] I have not referenced internal/instance-local Paperclip issues or
links (only public GitHub `#NNN` / `github.com/paperclipai/paperclip`
URLs)
- [x] My branch name describes the change and contains no internal
Paperclip ticket id or instance-derived details
- [x] I have run tests locally and they pass
- [x] I have added or updated tests where applicable
- [x] If this change affects the UI, I have included before/after
screenshots
- [x] I have updated relevant documentation to reflect my changes
- [x] I have considered and documented any risks above
- [x] All Paperclip CI gates are green
- [x] Greptile is 5/5 with no open P2s, recommendations, or follow-ups
- [x] I will address all Greptile and reviewer comments before
requesting merge
---------
Co-authored-by: Paperclip <noreply@paperclip.ing>
## Thinking Path
> - Paperclip is the control plane for running AI-agent companies, so
long-running remote work needs to stay observable to human operators.
> - Cloud / sandbox agents are an active roadmap area, and their
workspace sync path is part of the runtime substrate every remote coding
run depends on.
> - In the sandbox and SSH execution-target flows, Paperclip logged that
sync had started, then often went silent for the full transfer window.
> - That made large remote syncs feel stalled and also hid a real
performance problem in the command-managed sandbox upload path.
> - The first part of this pull request threads a throttled
progress-reporting surface through the adapter execution-target stack so
sync and restore work can emit meaningful updates.
> - The second part fixes the command-managed sandbox transport itself:
it removes the old serial 32KB append bottleneck, but also falls back
away from the single-stream path when a provider-backed sandbox runner
cannot surface mid-flight stdin progress.
> - The result is that sandbox and SSH transfers are both faster and
more observable, including the live Daytona-style sandbox case that
previously only emitted `0%` and `100%`.
## Linked Issues or Issue Description
No public GitHub issue exists for this bug, so it is described inline
below following the bug report template.
### What happened
- Remote sandbox and SSH workspace syncs could spend a long time
transferring data while only logging a start line (`Syncing workspace
and runtime assets to sandbox environment`) and, at best, a terminal
line.
- In the command-managed sandbox path, the original upload
implementation also paid a large performance cost by appending base64
data in many small sequential remote writes (thousands of serial 32KB
round-trips on a large workspace).
- After the initial transport rewrite, live provider-backed sandbox runs
still only emitted `0%` and `100%` because the single-stream stdin RPC
buffered progress until completion.
### Expected behavior
- Long-running sandbox and SSH syncs should periodically report how much
of the transfer is complete (a percentage and/or MB transferred) so an
operator can tell the run is healthy and making progress rather than
stuck.
- The main sandbox upload path should not be artificially slow.
- A transfer that fails partway should leave an explicit failure marker
in the log rather than a dangling intermediate percentage.
### Steps to reproduce
1. Run an agent against a sandbox (command-managed) or SSH
(remote-managed) execution target with a non-trivial workspace.
2. Watch the run log during the workspace/runtime asset sync phase.
3. Observe that the log shows the sync start line and then stays silent
for the full transfer (live provider-backed sandbox runs only show `0%`
then `100%`).
### Paperclip version or commit
- Branch `PAPA-825-provide-status-updates-when-syncing-sandboxes` off
`master`.
### Deployment mode
- Self-hosted / local instance using sandbox (command-managed) and SSH
(remote-managed) execution targets, including provider-backed sandbox
runners.
## What Changed
- Added shared throttled runtime progress reporting and threaded
`onProgress` through the adapter execution-target surface and adapter
`execute.ts` entrypoints.
- Added sync and restore progress reporting for the command-managed
sandbox path and the SSH/remote-managed path, including git
import/export progress where totals are known.
- Reworked command-managed sandbox transfer behavior so uploads use the
faster single-stream path when appropriate, but fall back to chunked
progress-emitting writes when the runner cannot expose mid-stream stdin
progress.
- Marked provider-backed environment sandbox runners as not supporting
single-stream stdin progress so live sandbox runs emit meaningful
intermediate updates instead of only `0%` and `100%`.
- Emit an explicit terminal failure marker (`failed at NN% (x/y MB)`)
when an SSH/tar transfer rejects, so a failed sync no longer leaves a
dangling intermediate percentage in the log.
- Run the SSH sync/restore size estimate (local directory walk / remote
`du` probe) concurrently with the transfer instead of awaiting it before
opening the pipe, so progress instrumentation no longer adds startup
latency proportional to workspace file count.
- Added and extended focused regression coverage for runtime progress
throttling and the new failure marker, command-managed sandbox
transfers, sandbox orchestration, SSH transfer progress, and environment
execution-target wiring.
## Verification
- `pnpm exec vitest run
packages/adapter-utils/src/runtime-progress.test.ts
packages/adapter-utils/src/ssh-fixture.test.ts
packages/adapter-utils/src/command-managed-runtime.test.ts
packages/adapter-utils/src/sandbox-managed-runtime.test.ts`
- `pnpm exec vitest run
packages/adapter-utils/src/command-managed-runtime.test.ts
server/src/__tests__/environment-execution-target.test.ts`
- `npx tsc --noEmit` for `packages/adapter-utils`
## Risks
- The provider-backed sandbox fallback now prefers chunked
command-managed writes when progress hooks are active, so
small-to-medium uploads may trade some raw throughput for observable
intermediate progress on runtimes that cannot surface true mid-stream
stdin progress.
- Progress percentages on tar-based transfers still depend on estimates
in some cases, so operators may briefly see MB-only lines before the
estimate resolves, then near-final clamping before the terminal `100%`
line.
- This PR changes shared execution-target behavior used by multiple
adapters, so regressions would most likely appear in remote runtime
setup/teardown flows rather than in a single adapter.
## Model Used
- Initial implementation: OpenAI GPT-5.4 via Codex local agent
(`codex_local`), high reasoning mode.
- Observability follow-ups (failure marker, concurrent size estimate,
added tests): Claude Opus 4.8 via Claude Code (`claude_local`).
## Checklist
- [x] I have included a thinking path that traces from project context
to this change
- [x] I have specified the model used (with version and capability
details)
- [x] I have checked ROADMAP.md and confirmed this PR does not duplicate
planned core work
- [x] I have searched GitHub for duplicate or related PRs and linked
them above
- [x] I have either (a) linked existing issues with `Fixes: #` / `Closes
#` / `Refs #` OR (b) described the issue in-PR following the relevant
issue template
- [x] I have not referenced internal/instance-local Paperclip issues or
links (only public GitHub `#NNN` / `github.com/paperclipai/paperclip`
URLs)
- [x] I have run tests locally and they pass
- [x] I have added or updated tests where applicable
- [ ] If this change affects the UI, I have included before/after
screenshots
- [ ] I have updated relevant documentation to reflect my changes
- [x] I have considered and documented any risks above
- [ ] All Paperclip CI gates are green
- [ ] Greptile is 5/5 with no open P2s, recommendations, or follow-ups
- [ ] I will address all Greptile and reviewer comments before
requesting merge
---------
Co-authored-by: Paperclip <noreply@paperclip.ing>
Jannes Stubbemann
Nicky Leach
Devin Foley
Dotta
Jon Mischo
Neeraj Kumar Singh B