The incomplete #73146 else branch still wrote the interrupt checkpoint into
the placeholder assistant row. Mid-tool steers then replayed that scaffold as
the model's own prior reply, which it echoed into a self-replicating ghost
loop. Carry the scaffold only on the user correction's api_content, matching
the assistant-tail branch.
* perf(gateway): stop spawning git for paths that cannot answer
The project tree probes every distinct session cwd, and on a long-lived
history most of those directories are deleted worktrees — `git -C` there
can only fail, at the price of a fork each. Stat first.
The second elision is `common_repo_root`: only repos have a common dir,
and the parallel warm never covers that probe because `resolve()` reaches
it only for cwds that already resolved. Every non-repo cwd was therefore
paying a serial `git` spawn on the discovery pass.
* perf(gateway): quit reading system prompts the project tree discards
`_project_tree_row` keeps about eighteen fields and drops the rest, but
the query behind it selected `s.*` plus the resolved system prompt — 37MB
of blob per build on my session history, read out of the B-tree and then
thrown away.
* perf(gateway): warm every path the project tree will resolve
The warm covered session cwds, but build_tree also resolves each declared
project folder and each discovered repo root. Those were the last probes
running one directory at a time while the sidebar showed a skeleton.
The warm covered session cwds, but build_tree also resolves each declared
project folder and each discovered repo root. Those were the last probes
running one directory at a time while the sidebar showed a skeleton.
`_project_tree_row` keeps about eighteen fields and drops the rest, but
the query behind it selected `s.*` plus the resolved system prompt — 37MB
of blob per build on my session history, read out of the B-tree and then
thrown away.
The project tree probes every distinct session cwd, and on a long-lived
history most of those directories are deleted worktrees — `git -C` there
can only fail, at the price of a fork each. Stat first.
The second elision is `common_repo_root`: only repos have a common dir,
and the parallel warm never covers that probe because `resolve()` reaches
it only for cwds that already resolved. Every non-repo cwd was therefore
paying a serial `git` spawn on the discovery pass.
* feat(desktop): resolve a session's pull request
A session row can say whether its work is open, merged or closed, and link
to it. The join is the session's own repo + branch, asked of GitHub in one
batched GraphQL request per repo (branch aliases, not a `gh pr list` page
that a busy repo crowds ours out of), through the remote-aware git facade so
a desktop on a remote gateway asks the backend's `gh`.
Two ways a session's branch can't answer, both covered:
- It ran on trunk. Fork PRs share our branch namespace, so asking about
`main` badges a stranger's PR onto it — trunk is never asked about, and
cross-repository PRs are dropped server-side either way.
- It worked in a worktree, so the branch it recorded at start isn't where
the PR came from. Creating a PR from the review pane binds the session to
the branch it actually used, and for sessions that predate that, the PR is
recovered from the transcript: `gh pr create` prints a bare PR url and
nothing else, so a tool result whose whole output is one is a claim rather
than a mention. Scanned read-only across profiles, once per session ever.
* refactor(desktop): one profile glyph
The rail, the profiles page and the session-row chip each drew the same
tinted initial square from scratch, so a row tag could disagree with the
rail about a profile's color. One component owns the square, its tint, and
the home icon the default profile gets instead of a letter.
* feat(desktop): sidebar filter menu
The sessions header's project/list toggle was one binary choice standing in
for a view. It becomes a menu: group by date, project or status; order by
updated, created, status, tokens or cost; show tokens, cost, PR, profile or
an always-visible timestamp per row; filter by status, pull request, project
or archived. Everything persists, and one reset puts it all back.
The pieces that make it read right:
- Status groups reuse the date dividers rather than inventing a second
separator, and a magnitude sort (tokens, cost) drops the calendar
entirely — "Today" above the priciest session you have ever had is a lie.
- Row metadata shares the trailing slot the kebab covers on hover, so only
the last fact steps aside and the number you switched on stays readable.
- A filter deepens the loaded page to 300 rows and hands the window back
when cleared, so "merged PRs" doesn't quietly answer for the last 50.
- Archived is a view of its own set, and dragging is still what picks a
manual order — the menu only offers a way back out of one.
The sessions header's project/list toggle was one binary choice standing in
for a view. It becomes a menu: group by date, project or status; order by
updated, created, status, tokens or cost; show tokens, cost, PR, profile or
an always-visible timestamp per row; filter by status, pull request, project
or archived. Everything persists, and one reset puts it all back.
The pieces that make it read right:
- Status groups reuse the date dividers rather than inventing a second
separator, and a magnitude sort (tokens, cost) drops the calendar
entirely — "Today" above the priciest session you have ever had is a lie.
- Row metadata shares the trailing slot the kebab covers on hover, so only
the last fact steps aside and the number you switched on stays readable.
- A filter deepens the loaded page to 300 rows and hands the window back
when cleared, so "merged PRs" doesn't quietly answer for the last 50.
- Archived is a view of its own set, and dragging is still what picks a
manual order — the menu only offers a way back out of one.
The rail, the profiles page and the session-row chip each drew the same
tinted initial square from scratch, so a row tag could disagree with the
rail about a profile's color. One component owns the square, its tint, and
the home icon the default profile gets instead of a letter.
A session row can say whether its work is open, merged or closed, and link
to it. The join is the session's own repo + branch, asked of GitHub in one
batched GraphQL request per repo (branch aliases, not a `gh pr list` page
that a busy repo crowds ours out of), through the remote-aware git facade so
a desktop on a remote gateway asks the backend's `gh`.
Two ways a session's branch can't answer, both covered:
- It ran on trunk. Fork PRs share our branch namespace, so asking about
`main` badges a stranger's PR onto it — trunk is never asked about, and
cross-repository PRs are dropped server-side either way.
- It worked in a worktree, so the branch it recorded at start isn't where
the PR came from. Creating a PR from the review pane binds the session to
the branch it actually used, and for sessions that predate that, the PR is
recovered from the transcript: `gh pr create` prints a bare PR url and
nothing else, so a tool result whose whole output is one is a claim rather
than a mention. Scanned read-only across profiles, once per session ever.
Fix: _LENGTH_CONTINUATION_DROPPED_TOOLS_PREFIX ended with '(' but
_get_continuation_prompt still had f'({tool_list})', producing
'((write_file)' instead of '(write_file)'. Removed the '(' from
the prefix constant — the parenthesis belongs in the interpolation.
Widened: promoted the empty-response nudge (line 6993,
'You just executed tool calls but returned an empty response...')
to _EMPTY_TOOL_RESPONSE_NUDGE constant and added it to the
classifier's recognition set. Same bug class — its
_empty_recovery_synthetic metadata flag doesn't survive SessionDB
projection either.
Test: added parametrize case for the empty-response nudge (7→8 cases).
E2E: verified byte-for-byte string equivalence for all nudge constants.
aed114a69 taught _is_synthetic_compression_user_turn to recognize the
max-iteration nudge as ephemeral runtime scaffolding rather than a human
turn, since its role="user" metadata flag doesn't survive SessionDB
projection and a crash/interrupt mid-turn can persist it durably — becoming
the compaction anchor / auto-focus topic in place of the real task.
conversation_loop.py's retry loop appends several more role="user" rows
with the exact same "ephemeral, metadata-tag-only" shape, none of them
recognized by the classifier:
- The three _get_continuation_prompt variants (length-continuation nudge,
tagged _length_continuation_nudge) — two fixed strings plus a third that
interpolates the dropped-tool-call list.
- _CODEX_INCOMPLETE_NUDGE (codex/responses reasoning-only retry).
- The codex ack-continuation nudge (acknowledgment-only reply re-prompt).
- The dropped-tool-call nudge (tagged _dropped_toolcall_nudge) — persisted
across up to 3 consecutive retries before the finalization pop-loop
strips it; an interrupt/crash before that pop can persist it same as the
max-iteration case.
Promote the previously-inline nudge strings to named module-level constants
in conversation_loop.py (single source of truth for both construction and
recognition), then extend the classifier to recognize all of them — exact
match for the five fixed-content nudges, a stable-prefix check for the
dropped-tool-call continuation variant (its tool list is interpolated so it
can't be exact-matched, same treatment TODO_INJECTION_HEADER already gets).
Imported lazily inside the classifier to avoid a module-load-order cycle —
conversation_loop.py already imports FROM context_compressor.py at call
time for the same reason.
Simplify registration/unregistration to match delegate_tool.py's
hasattr+getattr pattern instead of over-defensive try/except Exception
blocks. Delete inspect.getsource() change-detector test (breaks on
rename, proves nothing the behavioral test doesn't cover).
Net: -73 lines, +35 lines = -38 lines.
A background memory/skill review (agent/background_review.py) forks a
second, complete AIAgent in a daemon thread that deliberately shares the
live agent's own session_id for prompt-cache warmth. Nothing previously
stopped a user's next live turn from starting while that fork was still
mid-conversation, letting both stream against the same session_id and
credentials concurrently. That produced two observable failures:
- Doubled prompt-token accounting on the live turn's own calls (the two
concurrent request/response streams under one session_id confuse the
token-usage bookkeeping), triggering premature context compression.
- A lockup that a normal interrupt could not clear: the review fork is a
fully independent AIAgent with its own _interrupt_requested flag, and
was never added to the parent's _active_children list -- the only list
AIAgent.interrupt() actually walks for cross-agent cancellation -- so a
live-turn Ctrl+C had no propagation path to it at all.
Fix, three files:
1. agent/agent_init.py -- add _background_review_agent /
_background_review_lock tracking state to every AIAgent, mirroring the
existing _active_children pattern.
2. agent/background_review.py -- the review fork now registers itself on
the parent's _active_children right after construction (reusing the
same list/lock interrupt() already fans out to for real subagent
delegation), and unregisters on every exit path (success, the
tool-whitelist finally, and the outer exception safety-net). All
registration is defensive (getattr/try-except) so an AIAgent built
without going through agent_init.py's setup degrades to "no
cross-turn cancellation" instead of aborting the whole review.
3. agent/conversation_loop.py -- at the very start of every
run_conversation() turn, if a prior background review is still
in-flight, it is now proactively cancelled via interrupt() before the
live turn proceeds -- fire-and-forget, non-blocking, adds no latency.
Adds 3 regression tests to tests/run_agent/test_background_review.py,
confirmed to fail against the pre-fix code via a scripted revert.
Verified: ruff clean on all touched files; 66/66 background-review and
interrupt-propagation tests pass; 256/256 across turn_finalizer +
run_agent regression suites; no fork-only symbols in the diff.
Follow-ups from review of #82049:
- extract append_user_instruction() into agent/skill_commands so the
stable-prefix construction cannot drift between the skill and cron
builders (the registered prefix must stay a byte-prefix of the built
message); cron no longer imports the private _SINGLE_SKILL_INSTRUCTION
- add the startswith guard to the skill builder registration site,
matching the stronger cron guard
- rename _MAX_BYTES to _MAX_CHARS (sum(map(len, ...)) counts characters,
not bytes) and correct the comment
- collapse find_stable_prefix's two-lock dance into a single critical
section (scan is <=32 short-circuiting startswith calls, measured
2-4us; drops the snapshot copy and the TOCTOU re-check)
- document the split-shape lifetime (marked-endpoint window) in the
module docstring
- add a contract test for the helper's byte-prefix invariant
(mutation-checked)
Follow-up review of the builder-declared cache boundary (#81867) found three
ways the split could silently stop paying off, or keep paying more than it
should, on a long-lived gateway process.
Flattening no longer consults the registry. `strip_anthropic_cache_control`
matched the decorated split by looking the first block up in the prefix
registry, so a mid-turn failover that re-decorates a request built many
messages earlier (#72626) would fail to flatten once _MAX_ENTRIES newer
scaffolds had been registered in between, and would hand the next provider
the two-part shape instead of the canonical string. The split is now matched
by its shape: a marker on the *first* part of a user message is something no
other decoration produces (list content otherwise gets its marker on the last
part, and the two-part [static, volatile] split is role-gated to system), so
the ""-join stays provably byte-exact without any process state. This drops
`is_registered_stable_prefix` and one lock acquisition per stripped message.
Lookups now refresh LRU position. A scaffold fired every minute by cron could
be evicted by a burst of one-off skill invocations while still being the
hottest prefix in the process, silently reverting it to whole-message caching.
Registration now also evicts by total retained bytes (4 MiB). Entries hold
whole expanded skill bodies, so a 32-entry cap alone does not bound memory.
The newest entry is always kept, so a single oversized scaffold still gets a
boundary instead of disabling the split.
Tests: eviction-then-failover round-trip, LRU refresh on hit, byte-cap
eviction, and oversized-single-entry survival.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Webhook/cron skill invocations concatenate a large static scaffold
(activation note + expanded skill body) with a small volatile tail
(ticket payload, timestamps) into one user string, and the Anthropic
cache planner marked that whole string as a single atomic block — so a
few changed tail bytes forced a full cache rewrite on every invocation.
Instead of re-parsing scaffold marker strings out of the message at
request time (fragile when a payload or skill body quotes the marker),
the builders now register the exact stable-prefix bytes in a small
process-local LRU registry at construction time. The cache planner
splits a registered user string into [marked stable prefix, unmarked
volatile tail] request-locally; canonical session history stays a plain
string, and the failover stripper flattens the split back byte-exactly
via an O(1) registry lookup. Unregistered messages keep the existing
whole-message policy.
Covers the single-skill builder (webhook + slash command + TUI) and the
cron job prompt assembler (multi-skill, bundles, skipped-skill notice),
with registration guarded against injection-scanner sanitization.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
`read_window_below` enumerates through get-windows, which on Linux reads
`_NET_CLIENT_LIST_STACKING` via xprop. That is an X11 protocol, and Wayland
deliberately refuses to tell one application about another's windows. Under
XWayland it is worse than nothing: it finds the few legacy X11 clients and
silently misses every native Wayland window, which on a Hyprland desktop is
most of them — so the HUD floats over an app it cannot name.
Hyprland answers the question directly. `j/clients` on its command socket
returns every window with class, title, position, size, pid and focus history.
Ask it first when HYPRLAND_INSTANCE_SIGNATURE is set, fall back to get-windows
everywhere else, and keep the picking logic shared and unchanged.
Three things the provider has to get right, all covered by tests: order comes
from focusHistoryID rather than the list; windows on other workspaces are
dropped, since they share coordinates with the visible ones and would win the
overlap test; and our own window is left out, because focus history is not
stacking order — the HUD floats on top while the user works underneath it, so
slicing after ourselves would skip past the very app we are trying to report.
One request per tool call, opened and closed immediately: Hyprland evaluates
this socket synchronously and freezes until a five-second timeout on a
connection left hanging.
Folds the model-switch fix in with the untitled retry. They answer
different halves and each is wrong alone: counting alone left a session
that merely opened with machinery nameless forever, because nothing
reconsidered it, and the stored title alone would never title at all on a
store too old to report one. Skip only when both agree — past the opening
turn, and already named.
Counting a turn now judges a multimodal one on its text, so "here's a
screenshot, fix the login" counts as the question it is rather than
reading as machinery and undercounting the conversation.
Co-authored-by: yy28 <yy28@vip.sina.com>
Switching models before sending the first real message titled the session
"[System: The active model for this chat has…" instead of the user's actual
question.
`_append_model_switch_marker` persists its notice with `role="user"` because
strict OpenAI-compatible providers reject a system message that is not first
(#48338). Titling had no way to tell that apart from a genuine opening turn,
which caused two distinct failures:
1. `_MACHINE_PREFIXES` did not cover the marker. Its `[System: ` prefix
matches none of `[CONTEXT COMPACTION`, `[Runtime note:`, or `[SYSTEM]`
(different case, no closing bracket), so `is_titleable_user_message()`
returned True and the marker was formatted into the title.
2. `maybe_auto_title()` counted the marker as a user message. With the marker
present, the first real question arrived at `user_msg_count == 2` and the
`> 1` guard returned early, so the session was never titled at all and its
`title` column stayed NULL. Fixing only (1) would therefore have traded a
wrong title for a permanently missing one.
Add the marker prefix to `_MACHINE_PREFIXES` (kept in sync with
`tui_gateway.server._MODEL_SWITCH_MARKER_PREFIX`) and count only titleable
user messages when detecting the opening turn.
The guard stays narrow: ordinary user text that happens to start with
"[System:" still titles normally.
Adds 6 regression tests, verified to fail without the fix.
Two lookalike gaps found auditing the titler.
_MACHINE_PREFIXES missed the compressor's legacy summary opener and the
"[System note:" injections, so a compacted or resumed session could be
named after the note that carried it. Take the summary prefix from the
compressor that emits it rather than keeping a fourth local copy.
The fast-model exclude list covered embedders but not the other non-chat
siblings a provider names after its chat model — "gpt-4o-mini-tts"
satisfies the "-mini" rung and cannot answer a prompt.
An opener is not always titleable — an image with no caption, a compaction
handoff, a bare slash command — and those sessions stayed unnamed for
life, because the guard that stops re-titling a named session also stopped
the nameless one from ever asking again. Let a later turn name a session
that still has no title.
The derived title also ran the collision dedupe inline on the turn.
It is a slice of the user's own words, so it collides constantly — people
open sessions with "hi" — and resolving "hi #47" is a widening scan on the
critical path for a name the model replaces a second later. Decline it
there and let the background stage, which can afford the scan, pick it up.
The fast-model picker reads /v1/models to find the small model a provider
currently serves, and it asked anonymously. Most of those endpoints need a
key, so the fetch 401'd and the empty result read as "this provider has no
small model" — the picker fell back to its curated list and never noticed.
Worse, a failed fetch cached its empty result forever, so one bad moment
during startup disabled live model discovery for the life of the process,
and the processes that read this run for weeks. Give the failure an expiry
and pass the provider's credentials.
The bare family rungs (-mini, -flash, haiku) also picked whichever id
sorted first, which is the oldest generation a provider still serves:
gpt-3.5-mini over gpt-5.4-mini, claude-3-haiku over claude-haiku-4.5.
Compare the digit runs as numbers so the rung meant to keep us current
does.
Titling is two-stage — a slice of the user's own words lands inline, the
model's version replaces it a second later — and the platform rename lanes
fired on both. That is two rate-limited calls to reach one name, and
Discord allows two channel renames per ten minutes, so the throwaway could
be the one that survived. The callback now carries which stage it is, and
the lanes take the model's.
The relay lane also asked where the reply landed at title time, which is
before the model has answered: it polled the send-result cache for ten
seconds and read the timeout as "never auto-threaded", so any turn with
tool calls in it silently kept its raw thread name. Wait on the send
itself instead — the adapter already owns that cache, so it can say when a
reply arrives and, just as usefully, that one arrived carrying nothing.
The turn prologue titles every session, and it is shared by every agent —
including the ones no person is reading. A cron job already names its own
session after the job in its finally block, so the titler spent a side-LLM
call per fire to write the delivery scaffolding over it for the length of
the run. A delegated child's session is hidden from every picker, so a
batch at max_concurrent_children paid N title calls for N names nobody
opens.
Both are the same class of run that already sets skip_memory to stay off
the auxiliary path, so keep the titler off it too.
Every unsent tab was called "New session", so a row of them said nothing
about which was which. Name each one from its composer, using the same
first-line, word-boundary rule the backend's derive_title applies a moment
after the draft is finally sent — so the name the tab already shows is the
name it keeps.
The title moves with the composer, which is far faster than a pane
contribution should be re-registered. Panes can now render a tab label
instead of declaring one, so the label subscribes to its own key and a
rename repaints one string rather than the panes area.
A draft got no dot at all, so the one tab that has never done anything
looked identical to a settled session. Give it the faintest mark the app
has — a hollow outline, weakest claim in the dot's priority order, so the
first thing that actually happens speaks over it.
The row's own message_count is the tiebreaker for what counts as a draft:
a session RESUMING also holds an empty message list for a moment, and
calling that a draft flashes the wrong mark on a conversation with years
of history in it.
The chip was pointer-events: none until [data-slot='composer-rich-input']
had :focus, which made the only visible way out of HUD mode conditional
on the thing most likely to be broken when someone wants out. When focus
never lands (#81893 on macOS) you can neither type nor click your way
out: the HUD is a transparent always-on-top rectangle over the desktop
with no in-app dismiss.
It is now always clickable and dim (0.45) at rest, brightening on hover,
focus-visible, and composer focus. That keeps the original intent — not
a loud chip over the app behind — without gating the escape hatch on the
failure mode it exists for.
Salvaged from #82317 by @Ne0teric. The centering half of that PR is
dropped: #82233 already fixed the dock offset, and its 'translate: none'
is the exact literal Lightning CSS folds into 'transform', which is the
bug #82233 fixed.
Co-authored-by: Ne0teric <Ne0teric@users.noreply.github.com>
* fix(desktop): open HUD mode on the focused conversation's profile
The HUD is a full app renderer that adopted the PRIMARY backend's
profile at boot, so toggling HUD mode from a conversation on any other
profile resolved the session id against the wrong backend — the lookup
missed and the HUD fell back to the default profile's last session
(#82285).
- openHud() resolves the target's owning profile (session's stamped
owner, else the active gateway profile) and passes it through
hermes:hud:open.
- hudUrl() carries the profile in the query string next to win=hud;
the HUD renderer's gateway boot honors it as an override for both
getConnection() and profile adoption, so the window dials and adopts
the right backend from first paint.
- Retargeting a live HUD onto a session from a DIFFERENT profile
respawns the window against that profile's backend (a renderer adopts
its backend exactly once at boot; an in-place goto would repeat the
wrong-backend lookup).
No profile in the URL means no override — ordinary windows and
single-profile users boot exactly as before.
* refactor(desktop): extract the HUD renderer URL so its contract is tested
hudUrl() built the query string inline in main.ts, where the part that
actually breaks — `?win=hud&profile=` must sit BEFORE the '#' or
HashRouter eats it as the route — had no coverage. Move it next to
buildSessionWindowUrl's split (pure piece out of the monolith, unit
tested) and pin the contract: flag order, profile encoding, trailing
slash on the dev server, empty profile omitted, packaged file URL.
Co-authored-by: rainbowgore <rainbowgore@users.noreply.github.com>
* refactor(desktop): resolve the HUD's target profile through the existing ladder
openHud() had its own copy of "stamped owner, else active gateway, else
default" — the same ladder rememberedSessionProfile() already owns for
the remembered-navigation key, down to sessionMatchesStoredId and the
default fallback. One resolver per policy, so the two can't drift.
---------
Co-authored-by: Brooklyn Nicholson <brooklyn.bb.nicholson@gmail.com>
Co-authored-by: rainbowgore <rainbowgore@users.noreply.github.com>
* fix(desktop): keep the HUD on the session it was opened for
The HUD is a full app renderer, so the main window's cold-start
'restore last session' logic ran inside it: opening HUD on a blank new
chat (#/) navigated it to the remembered session instead of the new one,
because a blank draft has no stored id and the HUD boots at the default
route. Guard the restore/remember effect with isHudWindow() — the HUD's
destination is always chosen explicitly at open time.
Also stops the HUD from clobbering the main window's remembered
navigation while it is up.
* fix(desktop): use type-only import for the windows-store mock in HUD restore test
consistent-type-imports forbids inline import() type annotations; use the
established import type * as pattern (same as session-row.test.tsx).
Extend the packaged-app HUD geometry test from horizontal-only to full
containment: both axes for the dock and the input, plus an explicit
assertion that no percentage translate survives on the composer dock.
The vertical clipping reported on Windows (#82203) and macOS (#82214)
is the same escape class on the other axis, and the computed-translate
probe makes a future optimizer regression fail with a diagnosis instead
of a bare coordinate mismatch.
Two focus polish items from the first fully-working hand-off run
(ryanc, 2026-08-09):
1. The progress window came up backgrounded: the script is spawned via
`cmd start /min`, and Form.Show() + TopMost keeps it above other
windows without ACTIVATING it. Claim activation explicitly
(Form.Activate + SetForegroundWindow) right after Show.
2. The relaunched Desktop came up behind whatever the user had focused:
a WMI-spawned process starts unfocused and cannot take foreground by
itself. Since the hand-off owns foreground while its progress window
is up, delegate it: AllowSetForegroundWindow(new pid), poll up to 20s
for Electron's MainWindowHandle, then ShowWindow(SW_RESTORE) +
SetForegroundWindow. Best-effort at every step -- a focus failure
never affects the update result.
Sequence on success: progress window foreground during the update ->
window closes -> freshly relaunched Hermes.exe takes foreground.
Verified live on the incident machine: Add-Type shim compiles under
PS 5.1; WMI spawn + AllowSetForegroundWindow + MainWindowHandle poll +
ShowWindow all execute against a real spawned window. (In the bg test
shell SetForegroundWindow returns False by OS design -- only the
current foreground owner may delegate; the real flow's TopMost progress
window IS that owner.) PS parse clean, check-windows-footguns clean.
_PROVIDER_PREFIXES was a hand-maintained frozenset, so providers that ship
as plugins (bundled like fireworks, or user plugins under
$HERMES_HOME/plugins/model-providers/) were never recognised as
provider: prefixes in model strings, and metadata/context-window lookups
received the unstripped string. Mirror the _URL_TO_PROVIDER auto-extend
that already sits below it: add each registered profile's name and
aliases after discovery. The _OLLAMA_TAG_PATTERN guard keeps model:tag
strings intact.
Fixes#66106
First real-world run of the #82328/#82366 hand-off (2026-08-09, ryanc)
surfaced two defects:
1. The console window never closes after the update finishes -- and
closing it manually KILLS the freshly relaunched GUI. Root cause:
Start-DesktopRelaunch spawned Hermes.exe as a child of the console
PowerShell. Electron/Chromium calls AttachConsole(ATTACH_PARENT_
PROCESS) at boot, so the new Desktop latched onto the hand-off's
console: the console can't close while an attached process lives,
and closing it takes the attached GUI down with it. Fix: create the
process via WMI (Win32_Process.Create) -- parent becomes WmiPrvSE,
no console to inherit or attach, same detachment explorer.exe gives
a normal launch. Start-Process fallback retained (tethered Desktop
beats no Desktop).
2. Both the console and the progress box render hermes update's UTF-8
glyphs (checkmarks, arrows) as mojibake. PS 5.1 defaults redirected
child streams to the OEM codepage. Fix: StandardOutput/ErrorEncoding
= UTF8 on the child, PYTHONIOENCODING/PYTHONUTF8 so Python emits
UTF-8, and [Console]::OutputEncoding = UTF8 for our own echo.
Verified live on the incident machine: WMI-created process parents to
WmiPrvSE.exe (not the shell); UTF-8 glyph round-trip through the exact
ProcessStartInfo shape reads back byte-correct (15/15 chars). PS 5.1
parse clean, check-windows-footguns clean.
The #70716 regression fix changes popen_start_new_session from False to
True in the systemd-scope branch. Update the assertion in
test_wraps_in_systemd_scope_when_supervisor_and_available and the
docstring in test_systemd_post_spawn_failure_never_kills_gateway_process_group.
systemd-run --scope does not give the invoked process a new session: the
worker keeps the parent's session and inherits its controlling terminal.
When the parent is an interactive TUI on a pts (INVOCATION_ID present ->
is_gateway_supervisor_process()=True), every background spawn drops the
worker into the same session as the foreground process group; the spawn
then stops the whole session (SIGTTIN/SIGTTOU family), observed as 5 dead
TUIs in state T ("Arrêté") on 2026-08-08.
Fix: popen_start_new_session = True in the systemd-scope branch of
spawn_local. The worker (and the systemd-run wrapper) get a private
session while the scope cgroup isolation is preserved - the scope is
attached to the invoked process, not to the spawning session.
Verified: simulated TUI (INVOCATION_ID) + ProcessRegistry.spawn_local ->
worker in hermes-worker-*.scope with sid != simulator sid, exits cleanly,
simulator stays alive (previously: same sid -> stopped).
Review feedback on the #82328/#82366 hand-off, all four points plus the
missing progress GUI:
1. FAIL CLOSED. Both preflight gates aborted-open: a Desktop still alive
after 30s proceeded anyway, and a shim locked after 20s proceeded
with --force - both mutate a potentially locked install (the exact
Access-denied brick class). Now: desktop-alive -> exit 4, nothing
changed; shim-locked -> exit 5, nothing changed. Both relaunch the
Desktop so the user is never stranded.
2. TRUTHFUL COMPLETION. `hermes update` treats a Desktop GUI build
failure as non-fatal (warns, exits 0) - correct for CLI use, a lie
for a Desktop-driven update that then relaunches the OLD exe as
"success". The script now detects the warning in the update output,
retries the build once (`hermes desktop --force-build --build-only`),
and exits 6 with an honest message when it still fails.
3. MARKER OWNERSHIP. Cleanup now removes the marker only while OUR pid
still owns it - a handoff partner that rewrote the marker keeps its
claim (same rule as UpdateLock.release).
4. RESULT SURFACING. The script writes .hermes-update-result.json on
every exit path (ok, exit_code, message, branch, finished_at). New
electron/handoff-result.ts consumes it exactly once at the boot
update-gate: success logs, failure shows a real dialog pointing at
desktop-update-handoff.log. Stale (>30min) and malformed results are
consumed silently. Previously a failed detached update was
indistinguishable from "nothing happened" - the exact live report
that triggered this work.
5. PROGRESS UI. The old Tauri updater showed a window; the script ran
in a hidden console with zero feedback. It now shows a WinForms
progress window (marquee bar + streaming log) pumped via DoEvents
during the update; -NoUi keeps tests/headless sessions clean, and a
WinForms-unavailable session degrades to log-only.
Also: subprocess execution moved from Start-Process (ExitCode
unreliably $null under PS 5.1 even with the Handle workaround -
observed live: happy path reported "failed (exit )") to
System.Diagnostics.Process with synchronous stdout pumping, which
keeps the UI alive and the exit code real.
E2E on a real Windows box, sandbox HERMES_HOME + compiled fake
hermes.exe, all five paths:
- happy: exit 0, result {ok:true, "Update complete."}
- shim held open via O_RDWR: exit 5, nothing mutated, honest result
- desktop pid alive (60s ping child): exit 4 after the 30s gate
- update exits 0 printing "Desktop build failed" + rebuild fails:
exit 6, result names the stale build and the retry command
- foreign-owned marker: overwritten by step-0 claim, removed as owner;
ownership check verified in the cleanup path
vitest 18/18 (5 new handoff-result tests), typecheck 3 projects clean,
eslint clean, PS 5.1 parse + footguns + ASCII-only clean.
Remaining known gap (deliberate): the full click-to-relaunch lifecycle
through a REAL Desktop build still needs one live Windows verification
after this lands - tracked in the PR body.
Live failure on the first real use of #82328 (2026-08-09): clicking
Update closed the Desktop with "an updater will happen", then nothing.
desktop.log showed `launched repo hand-off script`, but
desktop-update-handoff.log was never created - PowerShell exited 0
without executing a single line.
Root cause, isolated by spawning the exact production shape against a
sandbox HERMES_HOME: `spawn('powershell', [..., '-File', script],
{ detached: true, stdio: 'ignore', windowsHide: true })` kills
powershell.exe during console-subsystem init, before -File processing.
Variant matrix: plain pipes -> runs; hide only -> runs; detached only ->
runs; detached+hide -> exits 0, script never starts. Unit tests and
foreground invocations can't see this class of bug.
Fix: wrapHandoffForDetachedConsole() routes the invocation through
`cmd /d /s /c start "" /min powershell ...` - `start` allocates the
script its own minimized console and fully detaches it; the cmd wrapper
exits immediately. Verified the wrapped form survives the full
detached+hidden production spawn.
Knock-on: child.pid is now the short-lived wrapper, not the script, so
the Electron-side marker pre-write can't represent the script. The
script now claims the update marker itself as step 0 (its own $PID,
byte-exact "<pid>\n<ts>\n" via WriteAllText - Set-Content emits CRLF
and would break the three readers' framing). The Electron pre-write is
kept as a bridge for the spawn window: the script overwrites it, and if
the script never starts the wrapper's dead pid reads as stale and
self-deletes (no wedge). `hermes update` adopts the script's claim via
update_lock.py's process-ancestry rule, unchanged.
E2E in exact production shape (cmd start wrapper, detached, hidden,
parent exits 1.5s after spawn) against a sandbox HERMES_HOME with a
compiled fake hermes.exe: script ran, claimed marker with its own pid
(fake observed "<script-pid>|<ts>|" LF-framed DURING the update),
desktop-pid wait worked, update invoked with correct argv, marker
removed on completion. vitest 13/13 (new wrapper-shape test), 3-project
typecheck clean, eslint clean, PS 5.1 parse + windows-footguns clean.
HexLab98
brooklyn!
hermes-seaeye[bot]
kshitij
pierrenode
Adam Durham
joaomarcos
yy28
Teknium
Ne0teric
Oliver Mee
Driss NAAMANE