## Thinking Path
> - Paperclip is the open source app people use to manage AI agents for
work
> - The telemetry subsystem (`packages/shared/src/telemetry`) flushes
event batches to an ingest endpoint; today the client drops batches
silently on 429 / 413 / 400 / network error with no re-queue and no
retry
> - The current behaviour is uncharacterised — there are no regression
tests pinning it, so any future retry work could accidentally change the
drop semantics without a test failing
> - Before adding retry logic it is critical to lock the current
baseline as an explicit, documented contract so regressions are
immediately visible
> - Additionally, the retry/cap design needs a config surface
(`maxEventsPerBatch`, `maxBodyBytes`, `maxPendingRetryBatches`,
`backoff`) so operators can tune behaviour without forking; that surface
should be defined and defaulted before the first consumer lands, not
after
> - This pull request adds characterisation tests that pin today's
silent-drop baseline and adds an optional, additive config surface with
resolved defaults — `client.ts` is untouched so no behaviour changes
> - The benefit is that the retry work in the follow-up PR can build on
a verified baseline, safe defaults, and a pre-wired config contract
rather than specifying everything at once
## Linked Issues or Issue Description
No pre-existing public GitHub issue. Underlying problem described below
(feature-request template).
**Problem or motivation**
The telemetry client in `packages/shared/src/telemetry/client.ts` has no
regression tests covering its error-path semantics. On 429 / 413 / 400 /
network failure the current code drains the whole queue via `splice(0)`
and discards the batch — silent drop, no retry. That behaviour is
correct for a best-effort client today, but nothing verifies it. When
retry logic lands, there is no safety net to confirm it does not
accidentally re-queue already-dropped batches. In parallel, the retry
design requires a typed config surface (`maxEventsPerBatch`,
`maxBodyBytes`, `maxPendingRetryBatches`, `backoff`) that the client and
its callers need to agree on before the first consumer is wired up.
**Proposed solution**
Two-phase additive scaffolding:
1. Lock the silent-drop baseline as pinned tests (temporary; Impl-2
replaces them when retry lands).
2. Add the config surface and defaults now so Impl-2 can consume them
immediately.
**Alternatives considered**
Writing the tests and config surface inside the retry PR — rejected
because it makes the retry diff much harder to review and loses the
regression-safety benefit of a committed baseline.
**Roadmap alignment**
Maintenance / client-correctness improvement; not a core roadmap
feature.
## What Changed
- **`client.test.ts`** — 5 new Phase-1 characterisation tests pinning
silent-drop on 429, 413, 400, and network-error. A
`stubFetchStatus(status)` helper DRYs the four drop cases. These are
temporary regression anchors: marked `TODO(impl-2): replace when retry
lands`.
- **`config.ts`** — `TELEMETRY_DEFAULTS` constant (frozen) +
`resolveCaps()` + `resolveTelemetryConfig()`. Provides a single exported
source of truth for defaults; nothing reads these yet.
- **`types.ts`** — 4 optional, additive `TelemetryConfig` fields:
`maxEventsPerBatch` (default 50), `maxBodyBytes` (default 524 288),
`maxPendingRetryBatches` (default 20), `backoff` (exp-with-jitter
shape). All optional and backward-compatible.
- **`index.ts`** — re-exports `TELEMETRY_DEFAULTS` and
`resolveTelemetryConfig` from the public package surface.
- **`client.ts`** — no changes (verified via `git diff --stat`).
## Verification
```bash
# Unit tests — 16 passed (5 new Phase-1 pins + 2 new Phase-2 config tests + 9 existing)
pnpm --filter @paperclipai/shared exec vitest run src/telemetry
# Type check — clean
pnpm --filter @paperclipai/shared typecheck
# Confirm client.ts untouched (zero output expected)
git diff HEAD~1 -- packages/shared/src/telemetry/client.ts
```
All three commands verified locally before push.
## Risks
**Low risk.** `client.ts` is untouched — no behaviour change. All four
modified files are additive:
- New tests cannot regress production behaviour.
- New config fields are `optional` and their defaults match the current
implicit constants in `client.ts`, so any future reader gets identical
semantics until it overrides them.
- `resolveTelemetryConfig` replaces no existing function; it is new.
- No envelope / wire change; no new PII / crypto / sink.
The `TODO(impl-2)` markers on the Phase-1 pins make their temporary
nature explicit in code.
## Model Used
- **Provider:** Anthropic
- **Model:** Claude Sonnet 4.6 (`claude-sonnet-4-6`)
- **Context window:** 200k tokens
- **Capabilities:** Tool use, code execution, agentic task completion
via Paperclip agent SDK
- **Mode:** Standard (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: Harold Kim <harold@paperclip.ing>
Co-authored-by: Paperclip <noreply@paperclip.ing>