Fleet audit showed these task skills are commonly needed across users;
shipping bundled per Teknium's direction. Docs and tests follow the
bundled paths.
Surfaced during the post-merge review pass on our own #81113 follow-up:
cached_fetch_api_models gained _cache_entry_valid (numeric-'at'
validation) but its sibling cached_provider_model_ids still did
float(entry.get('at', 0)), which raises ValueError/TypeError on a
hand-edited or corrupted provider_models_cache.json row and propagates
uncaught into the /model picker call sites. Same fix, same helper:
corrupt rows are now a cache miss (live fetch), never an exception.
Both wrappers now share the identical validity predicate, closing the
divergence the 'mirrors' docstring promised away.
Also two test nits from the same review: unused OrderedDict import
dropped and the drain-order assertion strengthened to pin LRU-first
FIFO order in tests/gateway/test_agent_cache_pressure.py.
Mutation-checked: restoring the raising float() form makes the new
corrupt-at tests fail.
Rescope: hermes verify fills only the runtime-smoke gap and plugs into
the pieces Hermes already has instead of standing beside them.
- agent/verification_evidence.py: record_verify_run() — explicit ledger
write for hermes verify results (shared _insert_evidence factored out
of record_terminal_result). Passing runs mark the workspace passed
like scripts/run_tests.sh; failures are recorded; --phase/--skip-start
runs are recorded as targeted scope.
- hermes_cli/verify_cmd.py: record results into the ledger on completion
(fail-silent, HERMES_SESSION_ID attribution); on the detect path merge
detect_project_facts verify commands the recipe missed into the
recipe's test list (never applied to a saved manifest).
- agent/verification_stop.py: recipe-aware nudge — when the workspace
has a runnable recipe (start command or .hermes/environment.json),
suggest hermes verify --json as the preferred full check; cheap,
try/except-guarded detection that can never break the nudge path.
- agent/verify/recipes.py: document layer ownership (coding_context =
cheap prompt facts; verify/recipes = deep runtime recipe).
- tests/verify/test_ledger_and_nudge_integration.py: 17 tests covering
ledger pass/fail recording, the closed edit->nudge->verify->satisfied
loop, recipe-aware nudge wording + fail-silence, and the facts merge.
vision_analyze reads container-only images by exec-reading them inside the
sandbox, but unlike terminal_tool it never triggered environment creation. Under
a non-local backend (ssh, docker, ...), a session whose first action was
vision_analyze on a remote path failed with 'no active sandbox session' until an
unrelated terminal command happened to establish the connection.
Add terminal_tool.ensure_task_env(task_id), a public lazy get-or-create that
reuses the terminal tool's own creation machinery, and call it from
image_source._resolve_container_fallback before the in-sandbox read. Extract the
ssh/container config-dict builders so both paths derive settings identically
(no duplication). Best-effort and fail-closed: a failed bring-up leaves the
existing 'no active sandbox' error intact, never a host read.
Fixes#62825
The modal/browserbase test file replaces tools.environments.base with a
SimpleNamespace stub; terminal_tool now imports EnvironmentConnectionError
from that module, so the stub must provide it too.
Connection-class infrastructure failures on remote terminal backends (SSH
host unreachable/timed out, Docker daemon down or missing, remote file
sync failing on a dead link) previously surfaced to the model as raised
RuntimeError tracebacks. The model got a stack blob with no guidance and
the failure was indistinguishable from a tool bug.
Now:
- New EnvironmentConnectionError(RuntimeError) in tools/environments/base.py
carrying a reason + retry_hint. Subclassing RuntimeError keeps every
existing catcher working.
- ssh.py classifies connect-refused, connect-timeout, scp, remote mkdir,
bulk upload/download, and remote rm failures as connection errors.
- docker.py classifies all four _ensure_docker_available() failure paths
(missing exe, non-executable exe, daemon timeout, `docker version`
failure).
- terminal_tool catches EnvironmentConnectionError and returns a
structured tool result the model can act on:
{"status": "degraded", "reason": ..., "retry_hint": ..., "exit_code": -1}
The failed backend is evicted from the environment cache so a later
call retries from scratch — recovery is automatic once the backend is
reachable again.
- Config gate terminal.degraded_mode: warn|fail (default warn) in
config.yaml, bridged as TERMINAL_DEGRADED_MODE across all four bridge
sites (cli.py env_mappings, gateway/run.py _terminal_env_map,
TERMINAL_CONFIG_ENV_MAP, DEFAULT_CONFIG). "fail" preserves the
historical error+traceback tool result.
- Command failures (nonzero exit, command-not-found) are NOT touched —
only infrastructure failures classify as degraded.
Tests: tests/tools/test_terminal_degraded_mode.py (15 tests) covering
exception classification for ssh+docker, structured degraded results,
no-caching of degraded envs, recovery after the backend returns,
nonzero-exit results unaffected, fail-mode preservation, invalid-mode
fallback to warn, and the four-site config bridge invariant.
Inspired by: Claude Cowork degraded-backend behavior (idea-level,
docs-only evidence).
Adds a bounded, read-only health probe per CONFIGURED tool backend, run
only when the user explicitly passes `--live` (real network calls):
- Firecrawl: credit-usage metadata GET (auth check, no scrape spend)
- FAL: models metadata GET (never a generation)
- Browser: headless launch + about:blank + close (full cleanup)
- MCP: initialize + tools/list per configured server (reuses the
`hermes mcp test` machinery in mcp_config._probe_single_server)
- TTS/STT: provider models/voices list GET (openai/groq/elevenlabs);
local providers (edge/piper/faster-whisper/...) skipped
Invariants:
- Opt-in only: zero probes without --live (default False)
- Bounded: sequential, per-probe timeout (doctor.live_probe_timeout,
default 10s, config.yaml knob)
- Never mutates state; unconfigured backends skip with a note
- Failure isolation: every probe wrapped in a catch-all; a probe crash
can never break the doctor run; failures append to the issues summary
New: hermes_cli/doctor_live.py, tests/hermes_cli/test_doctor_live.py
(23 tests, probes mocked at the HTTP/client seam).
Wired: --live flag in subcommands/doctor.py; run_doctor calls
maybe_run_live_checks after all static checks.
Coordination: PR #70124 (--probe-routes) probes LLM routes; this flag
probes TOOL backends — different surface, no code-region collision
(the run_doctor hook here sits at the end-of-run summary, not the
API Connectivity section #70124 extends).
Inspired by: paradigmxyz/centaur tool-health-smoke (MIT/Apache-2.0);
sibling: #70124 (LLM route probes — different surface)
Per-task `output_schema` (JSON Schema object) on task items plus the
top-level single-goal form — a one-time static addition to the tool
schema (never varies per call).
- Child side: the schema is appended to the child's context as an
explicit OUTPUT CONTRACT block before spawn.
- Completion side: the parent validates the child's final answer with
jsonschema; on failure it sends exactly ONE bounded retry turn
carrying the validation errors verbatim (no schema re-paste).
- Result entries gain schema_valid (+ schema_retries, and schema_errors
on final failure) ONLY when a schema was requested; schema-less calls
keep a byte-identical result shape.
- Malformed schemas are rejected loudly at dispatch (coerce_output_schema
meta-validates via jsonschema's validator_for/check_schema).
- New helpers in tools/delegation_output_schema.py: coerce, contract
block, fence/prose-tolerant extraction+validation, retry message.
Pattern from: github/copilot-cli ctx.agent(prompt,{schema}) — PATTERN
ONLY, zero code/prompt text copied (proprietary); proven consumer:
delegate-task-output-patterns skill.
Tests: tests/tools/test_delegate_output_schema.py (24 tests — valid
first try, invalid->retry->valid, invalid twice -> schema_valid false +
errors surfaced, retry-exception degrade, no-schema legacy shape pin,
dispatch rejection, contract plumbing). Delegation suite: 221/221 green.
Add an optional `region: [x1, y1, x2, y2]` parameter to vision_analyze
(pixel coordinates in the ORIGINAL image space). The crop is applied
with Pillow BEFORE the downscale/embed-cap pipeline, so the cropped
region gets the full resolution budget — a zoom for reading small text
or UI details after a full shot.
- New `_crop_image_region` helper: clamps out-of-bounds coordinates to
the image, rejects zero-area/inverted/malformed regions with an error
naming the actual image dimensions so the model can retry sensibly.
- Wired into both the native fast path (`_vision_analyze_native`) and
the legacy aux-LLM path (`vision_analyze_tool`).
- Schema gains one static optional param (byte-stable thereafter); the
description documents the intended flow: full shot first, then zoom.
- No region supplied = behavior unchanged (regression-guarded).
Tests: tests/tools/test_vision_region.py (11 tests — crop applied,
clamping, zero-area rejection with dims, malformed input, pre-downscale
full-budget zoom, schema shape, handler pass-through, no-region
unchanged). Widened one narrow fake_native stub in test_vision_tools.py
to be kwargs-tolerant.
Ported from: QwenLM/qwen-code zoom-image.ts (Apache-2.0)
write_file/patch targeting AGENTS.md, CLAUDE.md, SOUL.md, .cursorrules, or a
project-local .hermes config dir now ALWAYS prompt the human for approval —
even under --yolo/auto-approve — and fail closed when no human channel
exists. These files steer future agent behavior, so an injected write to
them is a prompt-injection persistence vector.
Design:
- New _check_protected_instruction_write() in tools/file_tools.py, a
sibling of _check_sensitive_path that returns approval-required rather
than a hard error. It realpaths before matching (symlink lesson from
#41351), matches basenames case-insensitively in ANY directory, rejects
'./x/../AGENTS.md' traversal via normpath, and gates files whose
immediate parent dir is `.hermes` (project-local config) while exempting
the authoritative ~/.hermes home (governed by its own guards).
- Approval is ONE-OPERATION only: no session/permanent persistence, no
yolo bypass — intentionally does not route through _run_approval_gate.
Gateway sessions get the button round-trip with allow_permanent and
allow_session both False; CLI uses the per-thread approval callback;
no channel at all = BLOCKED (fail closed).
- Multi-file V4A patches: ONE protected file gates the ENTIRE patch (a
single prompt lists all protected targets; deny applies nothing).
- Config: security.protected_instruction_files (default true) and
security.protected_instruction_extra_patterns (fnmatch on basename).
Config read failure keeps the gate ON.
Tests: 22 new cases in tests/tools/test_file_write_safety.py covering the
adversarial checklist — deny/approve/yolo-bypass attempt, symlink at a
protected target, case variants, relative traversal, arbitrary-directory
basenames, project-local .hermes, checkout-nested-under-~/.hermes
non-gating, patch replace + V4A multi-file atomicity, gateway round-trip,
fail-closed with no human, config off/extra patterns.
Ported from: RooCodeInc/Roo-Code RooProtectedController (Apache-2.0);
companion: #58631 (terminal vector), symlink lesson from #41351.
Adds a per-server `trust: full|untrusted` config key
(mcp_servers.<name>.trust). On an untrusted server, every write-capable
tool call — any tool whose discovery-time annotations do not carry
readOnlyHint=True — routes through the existing approval surface
(tools.approval.request_elicitation_consent, same lazy-import +
surface-routing pattern the MCP elicitation handler uses) before the RPC
fires. Denied/cancelled/errored approvals fail closed: the RPC never
runs, including the lazy first-use server spawn.
Design points:
- Classification happens at CALL TIME from metadata captured at
DISCOVERY (_record_tool_trust_metadata in _register_server_tools and
the lazy cache-registration path). No toolset/schema mutation, so the
toolset stays byte-stable and prompt caching is preserved.
- readOnlyHint is a server-supplied HINT: on an untrusted server a lying
server can at most skip approval for tools it claims read-only — it
can never widen access. Trust tiering itself is operator config.
- Missing/malformed annotations => write-capable (fail closed).
- Unrecognized trust values => untrusted (fail closed); missing key =>
full (backward compatible, documented in mcp-config-reference).
- The schema cache now persists readOnlyHint so lazy-registered servers
gate identically on next startup without spawning.
Tests: tests/tools/test_mcp_trust_gating.py (11 tests, TDD red->green):
approval invoked + accept proceeds, deny/cancel blocks RPC, readOnlyHint
=true skips gate, trusted/unconfigured servers skip gate, explicit
readOnlyHint=false gated, approval exception fails closed, trust
normalization, discovery-time capture (SDK objects and cached dicts).
Ported from: cloudflare-os classifyTool() (Apache-2.0), corroborated by
Claude Cowork (idea-level).
MCP servers generated from Rust/TypeScript union types commonly emit
closed value sets as const unions:
{"anyOf": [{"const": "red"}, {"const": "green"}, {"const": "blue"}]}
Strict tool-calling backends reject or mishandle these; the equivalent
property-level enum form is universally supported. Add
collapse_const_unions() to tools/schema_sanitizer.py and wire it into
the _normalize_mcp_input_schema discovery pipeline after the nullable
strip.
Rules:
- Collapse only when EVERY non-null branch is a pure const of the same
primitive type (bool never merges with integer).
- Mixed unions, non-uniform const types, and mismatched declared types
pass through untouched.
- A single {"type": "null"} branch is tolerated: consts -> enum,
null -> nullable: true hint (matches strip_nullable_unions, which
leaves null+multi-const unions alone by its one-non-null-branch rule).
- Outer title/description/default/examples carried onto the replacement.
- Deterministic, branch-order-preserving, non-mutating — applied at
discovery only, so schemas stay byte-stable per conversation.
Ported from: block/goose tool_schema_normalize.rs (Apache-2.0)
Two small config-gated features:
1. Kanban orphaned-card reconciliation (kanban.reconcile_orphans, default
true, config.yaml): a running card with broken claim bookkeeping
(claim_lock or claim_expires NULL — crash mid-claim, manual SQL, DB
restore) is invisible to all existing recovery paths
(release_stale_claims requires claim_expires NOT NULL,
detect_crashed_workers requires host-local lock + pid,
detect_stale_running is config-disabled by default) and shows Running
forever. New reconcile_orphaned_running() pass in kanban_db.py runs
each dispatch_once tick: requeues orphans to ready with an explanatory
comment, closes any leaked run, emits a 'reconciled' event, and defers
when the recorded PID is still alive on this host (never requeue
beside a live worker). Surfaced via DispatchResult.reconciled_orphans.
2. Per-server MCP identity header (mcp_servers.<name>.identity_header,
config.yaml): optional {name, value_from: static|profile, value}
mapping; the header is attached to that server's HTTP/SSE transport
requests. 'static' sends the config value; 'profile' resolves the
active Hermes profile name once at connect time (no per-call
mutation). Explicit per-server headers of the same name (any casing)
win. Invalid blocks warn-and-ignore; stdio servers warn-and-ignore.
Tests: tests/gateway/test_kanban_reconcile_orphans.py (9),
tests/tools/test_mcp_identity_header.py (13), all written first (RED)
then implemented (GREEN). No new HERMES_* env vars.
Inspired by: openai/symphony tracker reconciliation (Apache-2.0) +
Poke per-user MCP identity (idea-level).
Resumable ESTOP sentinel at $HERMES_HOME/ESTOP that halts NEW work only:
- agent/estop.py: sentinel engage/disengage/is_engaged (single stat, no
caching), optional reason + timestamp stored as JSON, paused_reply()
notice, check_paused() log-once-per-engagement helper. Corrupt/empty
sentinel still pauses (fail safe); a `touch ~/.hermes/ESTOP` works.
- cron/scheduler.py: tick() skips dispatch while engaged (logged once per
engagement, not per tick). Due jobs simply wait for the next tick after
resume — in-flight runs are never touched.
- gateway/kanban_watchers.py: dispatcher skips auto-decompose and worker
spawning while engaged; zombie reaping still runs and running workers
finish naturally.
- gateway/run.py: new gateway turns (post-auth, non-internal) get a brief
"Hermes is paused" reply instead of an agent run. Internal events
(in-flight background completions) bypass the gate.
- hermes_cli/subcommands/pause.py: `hermes pause [--reason]` and
`hermes resume`, wired into main() and _BUILTIN_SUBCOMMANDS.
- hermes_cli/status.py: `hermes status` shows a PAUSED banner (one stat).
- tests/test_estop.py: 20 tests — sentinel lifecycle, reason surfacing,
log-once, cron skip + resume, kanban gate, gateway paused reply +
internal bypass, CLI idempotence, builtin-set parity, status line.
Never kills in-flight work; resumable with no restart. Footprint ladder:
CLI command only, no new model tool, no new env vars.
Ported from: gastownhall/gastown estop.go (MIT); related prior art:
#26778 (/panic — kill/exit semantics, deliberately different: ours is a
resumable pause), #44617 (interrupt in-flight cron — out of scope here).
Each serialized result entry now carries cost_usd (rounded to 6 dp)
and cost_status (the child's session_cost_status — 'estimated',
'reported', 'included', or 'unknown') alongside tokens/api_calls/
duration, so the parent model can see what each delegation cost.
The internal _child_cost_usd field is still stripped before
serialization and the parent session cost rollup is untouched.
Tool schema is unchanged (byte-stable).
Inspired by: Perplexity Agent API result shape (idea-level)
Reject malformed tasks=[...] batches before any child agent is spawned:
- exact-duplicate goals (case/whitespace-normalized), error names both
task indices
- placeholder goals: bare 'TODO', bare 'task N', unexpanded <...> or
{...} template markers, or goals shorter than 10 chars after strip
- 1-task batches, with an error pointing the model at the single
`goal` form instead
All checks are batch-only — the single-goal form is exempt by design
(short goals like goal="test" are valid there). Error strings are
actionable: each tells the model exactly how to fix the call.
Tool schema is unchanged (byte-stable); validation is runtime-only in
the existing batch-validation region.
Existing tests using terse batch goals ("A"/"B"/"C") updated to
realistic distinct goals per the new contract.
Inspired by: MoonshotAI/kimi-code agent-swarm.md validation rules (MIT)
Validate a job's configuration BEFORE any agent machinery is constructed:
- missing provider API key (AuthError from a read-only
resolve_runtime_provider probe; skipped when a fallback_providers chain
is configured, since auth-fallback may rescue the run)
- attached skill not ready (skill_view readiness_status=setup_needed —
missing required env vars / commands / credential files)
- delivery platform unknown or unconnected (deliver=local/origin/all are
never checked; gateway-config load failures fail open)
On a failing check run_job returns a [blocked_config]-marked error without
constructing AIAgent/MCP/etc, so a misconfigured job never burns an LLM
call. run_one_job records last_status='blocked_config' and delivers the
alert exactly ONCE across ticks (persisted preflight_alerted bit — the
alert-once shape from the #73506 dead-pin auto-pause); the next healthy
run clears the marker so a future break re-alerts. Every preflight check
fails open: only an affirmative misconfiguration verdict blocks.
Config: cron.preflight (default true); `cron.preflight: false` restores
the old fail-during-run behavior. Documented in the cron user guide and
config defaults.
mark_job_run gains an optional status= override (unblocked call shape
unchanged) and drops preflight_alerted on any successful run.
Tests: tests/cron/test_preflight_config.py (blocked_config + no agent +
single alert across two ticks, healthy job unaffected, recovery clears
dedup, fallback-chain rescue, opt-out restores old behavior, skill
readiness miss, unknown delivery platform, deliver=local never loads
gateway config). Full tests/cron/ + cronjob tool suite green (525 tests).
Ported from: paperclipai/paperclip execution-semantics §5 (MIT);
in-repo precedent: #27948, #73506
- cron/notepad.py: SQLite-backed cron_notepad(job_id, key, value,
updated_at) store in its own profile-local db (cron/notepad.db),
following the executions.py connection/transaction pattern. APIs:
set_note/get_note/delete_note/list_notes/clear_notepad +
render_notepad_section. Documented size caps: 16KB per value,
128-char keys, 64KB per job total; oversized writes raise ValueError.
- cron/scheduler.py: inject non-empty notepads into the job prompt at
the context_from data-injection seam as a clearly-labeled
"Job notepad (persistent across runs)" section that also documents
the CLI write path. Empty notepad renders "" — byte-stable prompts
for jobs that never use the feature.
- hermes_cli/cron.py + hermes_cli/subcommands/cron.py:
`hermes cron notepad <job_id> [get|set|delete|list]` under the
existing cron subcommand tree (no new top-level command, no new
model tool — the agent writes via terminal + CLI).
- tests/cron/test_notepad.py: CRUD, durability, cap enforcement,
prompt injection, byte-stable empty case, read-failure resilience,
CLI handler + dispatch (TDD; watched fail first).
Inspired by: Amp (Sourcegraph) cron notepad (idea-level, proprietary —
zero code).
Add monitor-mode cron jobs: a cheap monitor source (monitor_script or
monitor_url) runs on every tick BEFORE any agent machinery is built.
Its output is hashed as exact bytes and compared to the hash stored
from the last agent-triggering tick:
- unchanged -> agent run suppressed entirely (no LLM, no delivery);
the tick is recorded as a silent no_change run visible in the
executions ledger doc
- changed -> a MONITOR CHANGE DETECTED block (capped unified diff of
previous vs current output + the new output) is injected into the
prompt via the existing extra_prompt seam, then a normal agent run
- first run -> always runs the agent with a baseline block
- source failure -> delivered as an ERROR alert, never treated as a
change; the stored hash is untouched so recovery to prior output
still suppresses
Implementation:
- cron/monitor.py (new): hash/diff/URL-fetch/state persistence.
monitor_script reuses _run_job_script (same ~/.hermes/scripts/
containment + interpreter rules); monitor_url is a bounded GET
(30s, 256KB, http/https only). Output is exact bytes by design —
scripts should emit stable output (documented).
- cron/jobs.py: additive job fields monitor_script / monitor_url /
monitor_state {last_output_hash, last_changed_at}. JSON job records
need no migration. create-time validation: sources are mutually
exclusive and incompatible with no_agent=True.
- cron/scheduler.py: one tight monitor gate in run_job between the
no_agent short-circuit and the LLM path (outside sibling-lane
regions). State persists in jobs.json + a per-job snapshot file, so
suppression survives scheduler restarts.
- tools/cronjob_tools.py: additive optional monitor_script/monitor_url
params on the cronjob tool (create + update, empty string clears),
path containment validated at the API boundary, surfaced in
_format_job.
- hermes_cli: --monitor-script/--monitor-url on `hermes cron create`
and `hermes cron edit`; `hermes cron list` shows the monitor source
and last-changed time.
Tests (tests/cron/test_monitor_kind.py, TDD): unchanged suppresses,
changed injects diff, first run always runs, hash persists across
module reload (restart), script failure is error-not-change with hash
untouched, create/update validation, tool wiring + path-escape reject.
Inspired by: ChatGPT Work monitor tasks (idea-level, docs-only);
enabler: #80774
Answers "what would the approval system do with this command?" without
executing it, prompting anyone, or persisting anything. Composes the
REAL runtime evaluators from tools/approval.py in the same order as
check_all_command_guards: container-skip gate, hardline blocklist,
sudo-stdin guard, user approvals.deny rules, yolo/mode-off bypass,
permanent command_allowlist, dangerous-pattern detection. Because the
same functions run — including _command_detection_variants's
normalization/de-obfuscation path — an obfuscated command gets exactly
the verdict its plain form would get at runtime, and the output shows
the normalized-variant trace the detectors actually evaluated.
- hermes_cli/approvals_test.py: evaluate_command() + text/JSON output.
Script-friendly exit codes: 0 allow, 1 usage, 2 ask-approval, 3 deny
(hardline / sudo-stdin / user deny rule).
- hermes_cli/subcommands/approvals.py: `test` subparser with --env-type
(default local), --json, and a REMAINDER command (dest command_words —
NOT "command", which main.py's startup path reads as the top-level
subcommand name).
- hermes_cli/approvals_suggest.py: dispatch `test` and mention it in the
bare-`hermes approvals` usage text.
- tests/hermes_cli/test_approvals_test.py: verdict matrix (benign /
hardline / dangerous / user-deny from config / container skip /
mode=off vs hardline), obfuscated==plain verdict parity with
normalized trace, spy proof that the real runtime detectors are the
ones invoked, read-only invariants (nothing executed; prompt and
persistence paths rigged to explode), JSON shape, dispatcher and
parser wiring.
Read-only by construction: only detection/matching functions are
called; the approval gate, prompts, gateway notify, and allowlist
writers are never reached.
Inspired by: Amp `permissions test` (idea-level, proprietary — zero code)
- Give cached_fetch_api_models the same stale-while-revalidate tier as
cached_provider_model_ids: TTL-expired entries within the 7d window are
served instantly while a background refresh rewrites the cache —
without this, every /model open an hour into the session re-blocked on
the live probe (#72762's stall class, deferred).
- Generalize _spawn_swr_refresh(cache_key, refresh_fn) so non-slug
custom:<base_url> keys reuse the same inflight-dedupe scaffolding;
slug behavior unchanged (default refresh_fn preserved).
- Convert the missed sibling site: acp_adapter/server.py
_named_custom_provider_catalogs() live-probed every custom_providers
row's /v1/models per ACP catalog build.
- Extract _cache_entry_valid() (the fp/models predicate existed 4x) and
validate 'at' is numeric so hand-edited/corrupt cache JSON degrades to
a live fetch instead of raising through the picker's blanket except.
- Flatten the dead api_mode conditional (fetch_api_models declares
api_mode=None; branch was behaviorally inert).
- Tests: 4 new guards (stale-serve, stale-window cutoff, generalized SWR
write-through, corrupt-at degradation) — stale-serve and corrupt-at
mutation-checked; 2 existing tests updated for the new behavior.
Custom OpenAI-compatible endpoints (named custom_providers rows, bare
provider: custom, and per-endpoint-map entries) called fetch_api_models()
directly at three call sites in model_switch.py, with no disk cache — unlike
first-class providers, which go through cached_provider_model_ids(). Every
plain /model open live-probed the active custom endpoint's /v1/models,
regardless of how recently it had already been probed.
Adds cached_fetch_api_models() in hermes_cli/models.py: a TTL disk-cache
wrapper keyed on custom:<base_url> (custom endpoints have no
PROVIDER_REGISTRY slug to key on) and fingerprinted on api_key/api_mode/
headers, with the same stale-beats-nothing fallback policy as
cached_provider_model_ids(). Routes all three probe call sites through it.
Since prewarm_picker_cache_async() already calls list_authenticated_providers()
with probe_custom_providers defaulting True, this also fixes the endpoint
being warmed on boot (populating the disk cache) instead of that work being
discarded on every open — any custom endpoint (an LLM gateway, a
self-hosted vLLM/SGLang server, etc.), not just one specific provider.
Fixes#72762. Salvaged from #72810 per review feedback: extracts just the
verified custom-endpoint cache fix with real cache-contract test coverage
(hit/stale/rotation/refresh/fallback), leaving the credential-pool and
Copilot-token-exchange costs described in the issue for separate follow-up.
- Drain the eviction plan (pop + del) before trim_memory: the batch
thread previously held every evicted agent in its local list while
gc.collect + malloc_trim ran, so the in-pass trim freed almost
nothing, the next tick re-read a still-high RSS, and the valve
over-evicted an extra batch of warm prompt caches per cycle.
- Clear _db_flush_scan_prefix in _release_evicted_agent_soft: it is a
shallow copy of the flushed transcript (stamped on every successful
flush) sharing every message dict — and pressure-evictable agents
have flushed by definition, so it pinned the multi-MB content strings
on exactly the agents the valve targets.
- Config-read failure now falls back to resolve_agent_cache_bounds({})
instead of bare AgentCacheBounds(): the dataclass default disables
the pressure pass, but an absent config section means 'auto' — a
transient read failure must not permanently switch off the OOM valve.
- protect_recent: false (YAML bool; False == 0) keeps the default MRU
protection instead of silently disabling it.
- 'No evictable session' warning now distinguishes sessions blocked on
un-flushed persistence (e.g. session DB never initialized — NFS
HERMES_HOME) from mid-turn agents, so operators can diagnose why the
valve isn't shedding instead of being pointed at running turns.
- _cgroup_limit_bytes checks the process's own cgroup (via the existing
gateway.cgroup_cleanup._own_cgroup_path) before the root files, so a
systemd unit's MemoryHigh=/MemoryMax= is detected — the root
memory.high/max read 'max' on those deployments.
- Tests: 5 new guards; drain-before-trim and scan-prefix-clear
mutation-checked (revert each fix -> its guard fails).
The per-session agent cache is capped at 128 entries with a 1h idle TTL, and
neither bound knows how many bytes it holds. Each cached agent pins
_session_messages -- the full transcript including tool output, tens of MB on
a session with 100+ tool calls -- so a gateway serving many chats keeps every
warm transcript resident: agents that took a turn inside the TTL are never
idle-swept, and the idle sweep additionally defers finalizable sessions until
they expire. RSS climbs until the cgroup throttles and SIGTERM can no longer
flush inside systemd's stop timeout.
Add the missing bound. Each session-expiry watcher tick compares the process's
anonymous RSS against a budget and, when over, sheds LRU agents through the
same soft-eviction path the cap enforcer uses, then runs malloc_trim so the
freed arenas actually return to the OS. Evicted sessions rebuild their
transcript from the persisted session on the next turn.
Three classes of session are never shed: agents mid-turn, the most recently
used ones, and any session whose transcript has not finished reaching disk
(_last_flushed_db_idx vs len(_session_messages) -- the same divergence the FTS
write-corruption guard reacts to when it preserves live history).
memory_high_mb defaults to "auto", deriving the budget from the cgroup limit
the gateway runs under, so a MemoryHigh/MemoryMax on the unit is respected
without a second number to keep in sync. The two existing bounds become
configurable alongside it under agent.agent_cache.
protect_recent is clamped to half the cache: a couple of sessions can exhaust
the budget on their own, and a fixed MRU guard would then protect everything
and leave the gateway climbing with nothing it would shed.
Fixes#80764
When the Nous Portal returns paid_service_access.allowed=false with
reason=member_spend_cap_exceeded, Hermes was falling through to the
generic 'no active subscription or usable credits' message — even
though the user has ample purchased credits and the real blocker is
an org-level per-member spend cap.
This adds a dedicated branch that surfaces the actual cause: names the
spend cap, shows the cap/spend amounts, and tells the user to ask their
org admin to raise it. Also adds member_spend_cap_exceeded to the
billing error code set so the error classifier and auth error formatter
route it through the Nous entitlement message path.
Review finding: the close block's comment promises a raising
session_db.close() is swallowed with the flag already cleared (no re-close
on a second agent.close()), but nothing pinned it. One test with a raising
_RecordingDB proves both halves.
Follow-up to the review on the session.resume ownership fix. Closing the
pre-transfer early returns left two gaps, both real.
1. The transfer had no owner on the other side. Once ownership moved to the
agent, teardown ran AIAgent.close() (via _teardown_session on session.close
and the orphaned-session reaper), which called session_db.end_session() —
that finalizes the session ROW, not the connection. A successfully resumed
profile session kept its dedicated handle, its db/-wal/-shm fds and its
background token-writer thread for the life of the gateway.
AIAgent now carries an explicit _owns_session_db, defaulting False so the
SHARED launch handle — which outlives every agent and backs every other live
session — is still never closed there. Only the dedicated-open sites set it,
at the point ownership actually changes hands.
2. session.resume was not the only profile-scoped open with no close on its
failure paths. Covered here with the same flag, via a _transfer_db_to_agent
helper that refuses the transfer unless the agent really holds that handle:
- the deferred builder (_start_agent_build), including the session-reaped-
mid-build case, where the built agent is discarded and never torn down, so
transferring to it would leak exactly as before;
- session.branch's branch_db;
- the compute host's per-profile open;
- AIAgent's own lazy open in _get_session_db_for_recall, which no other
object ever references and so was unconditionally abandoned.
Where a handle has already reached a registered session, the drop is
unconditional and the transfer is best-effort on top: a refused transfer leaves
the old leak, which is survivable, whereas closing under a live session is the
permanent "Cannot operate on a closed database" break the original patch exists
to avoid.
Tests: tests/tui_gateway/test_session_db_ownership_teardown.py (new, 14).
11 of the 14 fail without this change; the 3 that pass are the "must NOT close"
guards, which hold in both directions by design.
Review-pass follow-ups (three parallel reviewers, findings verified):
- hermes_state_search.py list_recent_user_messages now drops legacy
standalone compaction handoffs in the decode loop (SQL can't see them:
durable role=user, no display_kind). Closes the /undo N pairing skew
where the in-memory count (new predicate) and the DB soft-delete pick
(old predicate) targeted different turns on legacy sessions. Fetches
with headroom so the requested limit is still honored. 3 new tests,
mutation-checked (no-op'ing the skip fails 2/3).
- _should_skip_model_call_for_reference_handoff: single drive-check scan
(was two — once inside the restore helper, once after); the restore
helper no longer re-scans and its return value now decides the verdict.
- _final_response_from_messages replaced by the _HANDOFF_SKIP_FINAL_RESPONSE
constant it always returned (parameter was unused).
- _handoff_carries_live_user_content delegates to the canonical
_strip_context_summary_handoff_message — also fixes the edge where a
merged-shaped row with an EMPTY preserved prior tail was wrongly
treated as carrying live content.
- Site-level guard test for rollback.restore with a legacy handoff row
(predicate-in-context, complements the unit tests).
Pin #80622 invariants: handoff alone must not drive a model call after
stop, pending real users are restored, and synthetic compaction rows are
never treated as user-originated turns. Also give micro-compaction
enough passes to pay back the longer SUMMARY_PREFIX marker overhead.
The rough preflight estimate intentionally overestimates, but not by a
fixed margin: CJK text is counted at ~1.7x its o200k cost and
Responses-mode reasoning replay blobs at several times their billed
cost. Heavy sessions show rough estimates 2-3x real usage and compact
at 35-55% of the real window, stalling turns for minutes and discarding
detail (churn), because the defer guard only tolerated 5% rough growth
and sessions that never compressed had no baseline at all.
Pair every request's rough estimate (note_request_rough_estimate,
recorded in the conversation loop right after the pressure estimate)
with the provider's real prompt_tokens in update_from_response(), then
defer preflight while projected real usage — last real + rough growth
since that reading — stays under the threshold. Rough growth is itself
an overestimate of real growth, so the projection is an upper bound and
deferring below the threshold is safe; the provider's context-overflow
handler remains the backstop.
The baseline no longer ratchets on defer: it is refreshed by the
response pairing, and advancing it without a matching real reading
would shrink apparent growth and defer on stale data.
Review pass on the idle-unload feature found two material concurrency
bugs; both fixed here with a regression guard:
1. Unload-vs-use null deref (HIGH): _transcribe_local re-read the
module global _local_model at the transcribe call site. An idle
unload firing between the model load and transcribe() evaluated
None.transcribe → AttributeError → user-visible 'Local
transcription failed'. The window was real: the idle timer was only
touched AFTER a successful transcription, so a voice note arriving
exactly as the timeout expired raced the watcher directly.
Fix: bind a strong local reference under the model lock and use it
for the whole transcription (the watcher can null the global at any
time; this in-flight call keeps its instance — the generator holds
self, so no use-after-free). Also touch the idle timer at the START
of transcription so a long in-flight transcribe can't be counted as
idle time. The CUDA-fallback retry path gets the same treatment
(locked global write, local ref use).
2. Watcher replacement race + response-path join (MEDIUM/HIGH): the
old design stopped and re-started the watcher after EVERY
transcription with an unlocked set/join(5)/clear/start sequence on
shared globals. Two concurrent voice messages could interleave to
leave TWO live watchers (one with a stale, shorter timeout — a
raised unload_after_idle_seconds could still unload on the old
value), and the join(timeout=5) sat on the user-visible response
path (a watcher blocked on _local_model_lock during a concurrent
multi-second model load stalls the reply up to 5s).
Fix: single long-lived watcher under a management lock — started
only when none is alive (per-transcription cost: one lock + one
is_alive check), re-reads the configured timeout from config every
cycle (config edits now apply within one 30s interval, without
waiting for the next voice message — previously undocumented), and
stands down without unloading when the timeout is set to 0
mid-idle.
Tests: 17 now — idempotent start (same thread, no churn), config
re-read + stand-down-when-disabled, and the race guard
(unload firing mid-transcription must not fail the in-flight call).
The race guard is mutation-verified: reverting the fix (re-reading
the global at the call site) makes it fail with the exact NoneType
error; the fixed code passes.
The local faster-whisper model singleton (_local_model) is loaded once
and never released — the 'base' model holds ~370 MB of RAM/VRAM for
the entire lifetime of the process, even when no voice messages arrive
for hours or days. On long-running gateway processes (especially with
local LLMs competing for the same GPU) this is wasteful.
Add a config-driven idle unload: after stt.local.unload_after_idle_seconds
(default 0 = never) of no transcription activity, a lightweight daemon
thread sets _local_model = None so the Python GC can reclaim the
ctranslate2 objects. The next voice message reloads the model
transparently (the existing lazy-load path handles it).
The watcher:
- Checks every 30s whether idle time exceeds the configured threshold
- Acquires _local_model_lock before unloading (prevents races with
concurrent transcriptions that are mid-load)
- Exits immediately if the model is already None (unloaded by another
path, e.g. the CUDA fallback eviction)
- Is restarted by each transcription with the current config value,
so changing stt.local.unload_after_idle_seconds in config.yaml takes
effect on the next voice message without a process restart
Default is 0 (never unload) — zero behavior change for existing users.
Recommended value for gateway processes: 300 (5 minutes).
15 tests: config resolution (garbage/negative/None fallbacks), unload
safety (already-None, lock acquisition), touch timestamp, watcher
lifecycle (unload after timeout, no unload within timeout, exits when
model already None, stopped on new start). Existing STT test suite
unchanged.
Three-reviewer pass (reuse / quality / efficiency) on the trim diff;
four findings folded:
1. Short-clip input gate (efficiency, HIGH): the trim previously paid
the full ffmpeg encode before the <10%-saving discard check — every
dense conversational voice note burned 3 subprocess spawns + a
complete re-encode on the synchronous response path for nothing.
New _CLOUD_TRIM_MIN_INPUT_SECONDS=12 gate: below it, savings can't
matter (a >=10% saving is ~1s of audio, and several providers bill
a per-request minimum anyway — Groq bills 10s minimum), so the
whole pipeline is skipped using the duration we already probed.
Typical 5-10s voice notes now pay 1 ffprobe (~50ms), not 3 spawns +
encode (~0.3-1s; multi-second on small-VPS gateway hosts).
2. Shared encode profile (reuse, HIGH): the trim's ffmpeg command
duplicated _transcode_audio_for_stt's encode byte-for-byte (same
16kHz/mono/AAC-32k/faststart args, same subprocess.run kwargs).
Extracted _STT_M4A_ENCODE_ARGS + _run_ffmpeg_stt_encode(ffmpeg,
in, out, audio_filter=None); both call sites now share one owner,
so codec/bitrate/timeout changes can't drift between the paths.
3. is_truthy_value for the enable flag (quality, MEDIUM): raw
bool(cfg.get(...)) treated a YAML string "false" as enabled — the
exact bug class utils.is_truthy_value (already imported, already
used by is_stt_enabled and the xai/elevenlabs flags) exists for.
4. All-silence guard scales with keep_ms (quality, LOW): the fixed
0.3s floor equals the default keep window, so an output consisting
solely of one kept pause could pass as "speech"; now
max(0.3, 2*keep_seconds).
Also: _probe_audio_duration docstring documents it as the canonical
sync seconds-probe (gateway/run.py and the Telegram adapter carry
local variants of the same ffprobe invocation).
Tests: 24 now — YAML-string-false disables; short clips skip the
encode entirely (encoder mock asserted not-called); E2E fixtures
moved past the input gate. E2E re-verified: 13.2s note -> 6.2s
(-53%), 8s clip skipped with 1 probe.
Local faster-whisper gets Silero VAD (bf8004e3a) so silence never
reaches the model. Cloud providers got no such protection: the raw
file uploads untouched, so every second of silence in a voice note is
paid for twice — upload time and per-audio-minute billing — and cloud
Whisper hallucinates junk tokens on silent stretches exactly like
local Whisper did before the VAD hardening. A 13s voice note with two
long pauses is billed as 13s of audio to transcribe ~6s of speech.
Close the gap client-side: before uploading to a built-in cloud
provider (groq/openai/mistral/xai/elevenlabs/deepinfra), collapse long
pauses with ffmpeg's silenceremove filter, keeping
stt.cloud_trim_keep_ms (default 300) of every pause so word boundaries
and natural pacing survive. Uses ffmpeg, already a dependency of this
exact path via _transcode_audio_for_stt — no new dependency.
The trim is strictly best-effort — ALL of these upload the original
untouched, transcription never fails because of the trim:
- stt.cloud_trim_silence: false
- ffmpeg/ffprobe missing, trim failure, or timeout
- trimmed result ~empty (mostly-silence clip: the provider, not a
client-side dB heuristic, decides whether it contains speech)
- trim saves <10% (re-encoding for nothing)
Command-type and plugin providers are deliberately NOT trimmed: they
may wrap local CLIs that want the original bytes or run their own VAD.
E2E (real ffmpeg + faster-whisper): 13.2s voice note with 7s pause ->
6.2s upload (-53%); transcript of trimmed audio matches the original
on both utterances. Dense-speech and all-silence WAVs correctly fall
back to the original. 22 unit+E2E tests; STT/voice suite failures
identical to upstream/main baseline (all pre-existing).
_download_image() and _download_video() both used client.get() +
response.content, buffering the entire media body into memory before
checking the size cap. A server that omits Content-Length could send
an arbitrarily large payload, causing OOM.
Extract _stream_download_to_file() shared helper: streams via
client.stream() + aiter_bytes(), writes chunks to a temp file, enforces
the running byte count against the cap after each chunk, and atomically
replaces onto the destination on success. Cleans up the temp file on
failure. Uses utils.atomic_replace() for cross-device/symlink safety.
Malformed Content-Length values are now caught and ignored instead of
crashing with ValueError; the streaming cap is the authoritative guard.
Approach adapted from PR #10440 by @WuKongAI-CMU (closed as stale —
14923 commits behind, reverted 32 commits of vision_tools.py evolution
including SSRF-safe client, retry classification, and lazy imports).
Closes#10440
_fts_teardown_trash_step deleted rows via 'WHERE key IN (SELECT key
LIMIT N)' — each chunk's subquery re-scanned from the start of the
table, so chunk k skipped past (k-1)xN already-deleted rows: O(n²)
total row visits. On a v22 shadow table with ~230K rows that is on the
order of 10^8 row visits, turning optimize-storage teardown into a
multi-hour grind on slow disks, with a write lock held per chunk.
Single-column INTEGER-PK trash tables now drain via a fts_teardown_<tbl>_progress
high-water marker mirroring fts_rebuild_step: each chunk claims rows
past the marker (SELECT ... WHERE key > ? ORDER BY key LIMIT N), deletes
the claimed range, and publishes the new marker in the same transaction.
Per-chunk work is bounded → O(n) total.
TEXT-PK tables (the FTS config shadow table, pk like 'version') and
compound-key tables fall back to the legacy chunked delete — those are
small by construction.
Fixes#79324
Follow-up to the #80376 salvage:
- TurnLeaseToken.degraded is dead code since acquire() started raising
TurnLeaseTimeoutError: the only constructor site passes the default and
repo-wide there are zero external readers. Remove the field, its ctor
param, the repr segment, and the always-False guards in rebind()/release();
the class docstring's 'retained for compatibility' claim described
consumers that do not exist.
- Revert pure black-rewrap churn in test_config_env_bridge_authority.py and
test_turn_lease.py (hunks on functions this change does not touch), keeping
the functional Windows-env/encoding additions.
- Turn the default-value test into an invariant: config default must equal
gateway.turn_lease.DEFAULT_LEASE_WAIT instead of pinning the 1800 literal.
- Rewrap the dangling 'Released' comment line in gateway/run.py.
Verified: 20/20 targeted gateway tests, ruff clean; mutation check — with
gateway/turn_lease.py reverted to pre-fix main the module fails, restored it
passes.
teknium1