Review follow-up on the salvaged AAC + markdown changes:
- Fix an inaccurate comment claiming the STT layer has a sniff-and-remux
fallback (verified: no such fallback exists; the ffmpeg-absent path caches
raw ADTS and STT may reject it).
- Type the _markdown_to_signal wrapper as tuple[str, list[str]] to match the
shared helper instead of a bare tuple.
- Replace the hardcoded /home/pi/... test fixture with a runtime-generated
ADTS AAC sample so the remux round-trip actually runs in CI (skips only
when ffmpeg is absent) instead of always-skipping.
Android Signal delivers voice notes as raw ADTS AAC frames, which
share the `0xFF 0xFx` sync word with MPEG-1/2 Layer 3 (MP3). The
`_guess_extension` byte-signature test in gateway/platforms/signal.py
was matching both, so ADTS AAC was being misclassified as MP3 — saved
to disk with the wrong extension and rejected by every major STT API
(Groq, OpenAI) because their server-side format sniffers inspect the
actual codec, not the file extension.
Two changes:
1. Tighten the MP3 vs ADTS disambiguator. ADTS packs `ID`,
`layer`, and `protection_absent` into bits 3-0 of byte 1, where
`ID=0` and `layer=00` for AAC. Real MP3 has `ID=1` and
`layer` in {01, 10, 11}. The mask `0xF6` against target `0xF0`
cleanly separates them.
2. Remux raw ADTS AAC to MP4 container at the cache step via
`ffmpeg -c:a copy`. Single demux/remux, no re-encode, no quality
loss, sub-100ms on a Pi 5. The cached file is a normal `.m4a`
that all major STT providers accept. ffmpeg is a transitive
dependency of many other Hermes features (TTS, video skills) so
this isn't a new install requirement; the remux degrades
gracefully to a no-op if ffmpeg is missing.
The new helper `_remux_aac_to_m4a` is unit-tested with a real
Android voice note from the audio cache that originally triggered
the bug, plus synthetic ADTS frames for the byte-level
disambiguator and garbage-input graceful failure.
Closes the gap that broke transcription for any Android Signal user
sending voice messages to Hermes.
Route Signal send paths through shared markdown formatting helpers and render markdown bullets consistently as Unicode bullets. Add coverage for Signal formatting and send_message integration.
The raft platform plugin's check_raft_requirements() logged a WARNING every
time it returned False. Since check_fn is called on every load_gateway_config()
(~every 10s during normal gateway operation), users who don't have the raft
CLI installed get their logs flooded with no way to suppress it — hermes plugins
disable doesn't work for bundled platform plugins, and platforms.raft.enabled:
false doesn't gate the check_fn call.
Fix: make check_raft_requirements() a silent predicate (return True/False
only, no logging), matching the convention documented and used by other
platform adapters (e.g. teams/adapter.py). The caller in
gateway/platform_registry.py create_adapter() already emits its own warning
when requirements aren't met and an adapter is actually requested — that's the
correct place for a user-facing warning (fires once per connect attempt, not
once per config load).
Fixes#49234
When a tool call itself restarts the gateway (docker restart, systemctl
restart, and similar), the process is terminated mid-call — before the
tool result is persisted and before the orderly drain rewind can run. The
transcript tail is left as an assistant(tool_calls) with no matching tool
answer. On resume the model re-issues the unanswered call, taking the
gateway down again — an infinite loop (#49201).
Source fix: _build_gateway_agent_history now strips a trailing
assistant(tool_calls) block that has no tool answers
(_strip_dangling_tool_call_tail), so there is nothing for the model to
re-execute. This complements _strip_interrupted_tool_tails, which only
handles the case where a tool result row exists with an interrupt marker.
Cognitive backstop: the resume-pending system note now states that any
restart command in the history already ran and must not be re-executed or
verified, and the empty-message auto-resume startup turn reports recovery
and asks for instructions instead of the nonsensical "address the user's
NEW message" (there is no new message on that turn).
Reimplements the intent of #49243 by @JoaoMarcos44 at the replay layer.
Fixes#49201
_build_fal_payload and _build_fal_edit_payload assemble the request and then
filter it down to the model's supports / edit_supports whitelist. That filter
also covers prompt (and image_urls for edits), which every FAL endpoint
requires. Today all model configs happen to list those keys, but a single
config that omits one would silently produce a request with no prompt or no
source images — a broken generation with no error.
Always keep the mandatory keys regardless of the whitelist so a missing
whitelist entry can only drop optional knobs, never the prompt or the images.
The god-file Phase 4 refactor (094aa85c37) moved agent construction into
CLIAgentSetupMixin, which set the atexit shutdown reference with a bare
`global _active_agent_ref`. After extraction that global binds the *mixin
module's* namespace, not cli.py's. cli._run_cleanup reads
cli._active_agent_ref to decide whether to fire the memory provider's
on_session_end hook — and it stayed None for the whole session, so the
`if _active_agent_ref:` branch was dead and on_session_end never ran on
/exit. Custom memory providers silently lost end-of-session extraction.
Fix: publish the reference onto the cli module explicitly
(`import cli as _cli; _cli._active_agent_ref = self.agent`), using the
deferred-import pattern already established in the mixin.
Regression test asserts cli._active_agent_ref is populated by the mixin's
publish line and guards against a relapse to the bare `global` form. The
existing shutdown tests passed only because they hand-assigned the ref,
which is exactly what masked this.
Teams overrode send_image/send_image_file but not send_video, send_voice,
or send_document — so when the gateway dispatched a video/voice/document
reply to a Teams chat it fell through to the base-class text fallback and
sent the local file path as plain text (same broken-UX class as the LINE
URL-image gap in #49298).
Extract the existing send_image attachment logic into a shared
_send_media_attachment helper (remote URL by reference, local file as a
base64 data URI, MIME guessed from the path) and route all four media
kinds through it. 5 new tests cover remote-URL, local-file base64,
no-app, and missing-file paths.
Simplify pass on the picker-persist coverage:
- Stub list_picker_providers + resolve_display_context_length so the
tests no longer make real outbound HTTP calls (OpenRouter catalog +
Ollama /api/show) during picker setup and confirmation rendering.
Runtime drops from ~11s to ~0.4s and the tests are now deterministic.
- Collapse the two positive persist cases into one parametrize over the
config seed (nested-dict vs flat-string), asserting the nested-dict
invariant in both.
- Assert the in-memory session override is applied in the --session
case, closing a 'passes for the wrong reason' gap (config untouched
AND the switch still took effect).
- _FakePickerResult -> types.SimpleNamespace.
Mutation re-checked on the final test: both persist cases fail on
pre-fix slash_commands.py; the --session case passes on both.
Add regression coverage for the picker persist fix: drive the real
_handle_model_command with a fake picker-capable adapter that captures
the on_model_selected callback, fire a 'tap', and assert config.yaml is
written (bare /model), left untouched (--session), and that a flat-string
model: is coerced to a nested dict on a tap.
Mutation-checked: the persist and coercion assertions fail on pre-fix
slash_commands.py and pass on the fix.
Auto-generated session titles already rename the Telegram forum topic via
the title_callback path, but the /title command only wrote the session
title to the database. On a Telegram topic lane the visible topic kept its
auto-assigned name, so a user who ran /title to override it saw no change.
Propagate the user-chosen title to the topic by calling the existing
_schedule_telegram_topic_title_rename helper on a successful /title set. It
already no-ops off Telegram topic lanes and when auto-rename is disabled.
MCP Streamable HTTP servers that garbage-collect idle sessions on a short
TTL (e.g. Unreal Engine's editor MCP, ~15s) were unusable: the keepalive
was hardcoded at 180s, so the session was always dead by the time it ran,
and every idle tool call then landed on an expired session and paid the
full reconnect path (observed hangs of 113-143s until interrupt, bounded
only by the 300s tool_timeout).
Two coordinated, backward-compatible changes:
- Add per-server `keepalive_interval` (config.yaml, not an env var per the
contribution rubric). Default 180s — byte-identical to the old hardcoded
value when unset — floored at 5s. Servers with short session TTLs set it
below their TTL so the session stays warm.
- Switch the keepalive probe from `list_tools()` to `ping` (the MCP base
protocol liveness primitive). On large servers `list_tools` pulled ~1 MB
every cycle (830 tools = 1,068,041 bytes); `ping` is ~55 bytes and works
uniformly across tool/prompt/resource servers. Tool-list changes still
arrive out-of-band via notifications/tools/list_changed -> _refresh_tools.
`ping` is an OPTIONAL utility, so to guarantee zero regression for a
tool-capable server that doesn't implement it: the first -32601 latches
`_ping_unsupported` and the probe falls back to the pre-ping `list_tools`
path for that connection (no reconnect loop). The latch resets on each
fresh connection (_discover_tools, all transport paths) so a server that
gains ping support after a reconnect is re-probed with the cheap path.
Non-(-32601) ping errors propagate as genuine liveness failures.
Verified end-to-end against a live Unreal MCP server (idle 22s past the
~15s TTL -> post-idle tool call returns in 0.31s, no teardown) and with a
simulated ping-less tool server driving the real keepalive loop (ping once,
list_tools thereafter, no reconnect). 25/25 unit tests pass.
Note: a separate upstream defect (modelcontextprotocol/python-sdk#2604)
still tears down the whole session when one tool-call POST returns 4xx;
that is not addressed here.
* fix(discord): hydrate channel context when replying to a message
Replying to a message in a free-response (non-mention, threads-off)
channel previously received only the 500-char "[Replying to: ...]"
snippet — the history-backfill gate fired only for mention-gated
channels and threads, so a reply got no surrounding channel context.
Replies now route through the same _fetch_channel_context hydration
that threads use. When the user replied to a specific (often older)
message, a reply-anchored window is scanned ending at that message so
the agent sees the exchange around what was pointed at, even when the
target sits before the self-message partition. The two windows are
merged chronologically and de-duplicated by message id.
Also hardens the recent-window scan to skip non-conversational status
bumps before the self-message partition check, and makes author-name
resolution defensive against partial/deleted authors.
* fix(discord): duck-type reply-target resolution instead of isinstance(discord.Message)
The e2e suite stubs the discord module, so discord.Message is a MagicMock
and isinstance(_resolved, discord.Message) raises 'isinstance() arg 2 must
be a type'. Any object with an int .id works as a scan anchor, so resolve
the reply target by duck-typing on .id and fall back to a _Snowflake from
the reference message_id.
Second review pass (Codex + Hermes subagent). Codex reproduced a real race with
a two-thread harness; both converged on the remaining issues.
- Generation-aware publish (fixes a lost-update race): two refresh callers (the
late-refresh daemon and the between-turns prologue around turn 1) could each
compute a snapshot outside the lock; a SLOWER caller holding an OLDER registry
generation could acquire the publish lock after a newer caller and clobber it,
deleting just-landed tools. refresh_agent_mcp_tools now captures
registry._generation before computing and refuses to publish a stale set;
agent._tool_snapshot_generation tracks the published generation.
- Context-engine routing names (_context_engine_tool_names) are now staged on a
local and published atomically with the snapshot, and only claimed when this
rebuild actually appended the schema — matching agent_init's dedup so a
registry/plugin tool of the same name keeps its own dispatch. (Previously
mutated live, before the publish lock, and on no-change refreshes.)
- CLI /reload-mcp: self.enabled_toolsets is resolved once at startup, so a
server newly ENABLED in config mid-session wasn't picked up (TUI already
re-resolved). Merge now-connected MCP server names into the override (unless
the user pinned all/*), mirroring startup, and keep self.enabled_toolsets in
sync. Closes the CLI/TUI parity hole.
- ACP (acp_adapter/server.py) routed through the shared helper — it was a 5th
sibling rebuild that re-injected memory tools but NOT context-engine tools and
bypassed the atomic/name-diff path (inert today, fragile).
- mcp_startup._resolve_discovery_timeout pulls its default from DEFAULT_CONFIG
(single source of truth) instead of a stale hardcoded 5.0 literal.
- Tests: stale-generation-no-clobber, _skip_mcp_refresh honored, timeout
fallback uses DEFAULT_CONFIG.
Consolidated findings from three independent reviewers (Codex, Claude Code, a
Hermes subagent w/ the hermes-agent-dev skill):
- BLOCKING: refresh_agent_mcp_tools rebuilt only the registry subset, silently
dropping post-build-injected memory-provider (mem0/honcho/…) and context-
engine (lcm_*) tools on every refresh. Now additive-preserving: re-applies
the same injectors agent_init uses, staged on locals and published atomically.
- Re-injection now honors the #5544 enabled_toolsets gate for context-engine
tools, so a restricted-toolset platform can't get lcm_* leaked back in.
- Atomic read-diff-publish under one lock: the returned `added` set and the
(tools, valid_tool_names) pair are consistent even under concurrent callers
(no half-swap, no TOCTOU).
- background_review fork opts out (_skip_mcp_refresh) so its byte-identical
tools[] cache parity with the parent is preserved.
- CLI /reload-mcp routed through the shared helper (was a 4th divergent copy
with the same clobber bug + missing disabled_toolsets).
- Explicit reloads (TUI RPC + CLI) pass enabled_override so a server the user
just enabled in config this session is picked up; automatic paths reuse the
agent's build-time selection.
- mcp_discovery_timeout default 5.0 -> 1.5s: correctness now comes from the
between-turns refresh, so the startup wait is only a small turn-1 UX bump
rather than a heavy dead-server latency penalty.
- has_registered_mcp_tools checks registered TOOLS (not connected servers) so a
zero-tool/prompt-only server doesn't make the per-turn hook fire forever.
- Tests: rewrote the thread-safety test to actually exercise the write path
(alternating tool sets), added the #5544-gate regression, the memory/context
preservation regression, and a "callable next turn via valid_tool_names"
contract; removed a dead monkeypatch line.
A slow MCP server (HTTP/OAuth, 2-6s cold connect) that finishes connecting
after the agent's one-time tool snapshot was uncallable for the rest of the
session. The merged pre-first-turn late-refresh only helps during the dead air
before the user's first keystroke; once a turn starts it bails to protect the
prompt cache, so a user who types before the server connects never gets the
tools without a manual /reload-mcp.
Refresh the snapshot in the per-turn prologue (build_turn_context), before this
turn's first API call assembles tools=. This is cache-safe by construction: the
refresh only ever extends a fresh request prefix at a turn boundary, never
mutates the cached prefix of an in-flight turn. So late tools become callable on
the user's NEXT turn automatically, with no /reload-mcp and no cache cost.
- tools/mcp_tool.py: has_registered_mcp_tools() — cheap guard so sessions with
no MCP servers (the common case) skip the rebuild entirely.
- agent/turn_context.py: call the shared refresh_agent_mcp_tools() helper at the
top of the prologue when MCP servers are registered.
- tests: 3 contract tests through the real build_turn_context (adds late tool;
skipped when no servers; no snapshot churn when unchanged).
.hermes/plans/: SPEC + PLAN documenting the root cause, the cache-safety
constraint, and why the existing fixes (#48403/#41630/#42802) don't close it.
MCP servers that connect after the agent's one-time tool snapshot were
invisible for the whole session. Two root causes, fixed together:
1. The startup discovery wait was a flat 0.75s. HTTP/OAuth servers
commonly take 2-6s on a cold connect, so they missed the window and
their tools never entered the agent's snapshot. `thread.join(timeout)`
already returns the instant discovery completes, so raising the bound
costs ~0s for the common case (no MCP / fast servers) and only ever
blocks for a genuinely-pending server, capped so a dead server can't
freeze startup. The bound is now configurable via
`mcp_discovery_timeout` (config.yaml, default 5.0s).
2. Three call sites duplicated the agent tool-snapshot rebuild (the TUI
`reload.mcp` RPC, the gateway reload, and the TUI late-binding refresh
thread), and the late-refresh detected changes by tool COUNT — missing
an equal-size add/remove swap. Consolidated into one shared
`tools.mcp_tool.refresh_agent_mcp_tools(agent)` helper that diffs by
tool NAME, mutates the agent under a lock (thread-safe), and respects
the agent's own enabled/disabled toolsets.
The late-binding refresh keeps its pre-first-turn cache-safety guard:
it never rebuilds the tool list once a turn has started, so the cached
prompt prefix is never invalidated mid-conversation.
Tests: new tests/tools/test_refresh_agent_mcp_tools.py covers the
name-based diff, in-place mutation, agent-scoped filtering, thread
safety, and the config-driven discovery bound (incl. instant-return
when nothing is pending). 75 passed across the touched areas.
next(iter(frozenset)) picked a different blocklist var each run
(PYTHONHASHSEED-dependent), hurting reproducibility. sorted()[0]
keeps the invariant-style assertion (any real blocklisted var)
while making failures reproducible.
Follow-up to salvaged PR #49207.
Wires support for the MCP `elicitation/create` request (Python SDK 1.11+)
so MCP servers can ask the user to confirm sensitive operations
mid-tool-call (payment authorization, OAuth confirmation, etc.) instead
of failing closed or requiring out-of-band biometrics.
Behavior:
- `tools/mcp_tool.py` adds `ElicitationHandler`, attached per server task
and passed to `ClientSession` as `elicitation_callback`. Form-mode
requests route through the existing approval system; URL-mode requests
decline cleanly (out of scope for this pass).
- `tools/approval.py` adds `request_elicitation_consent()`, which dispatches
to whichever surface owns the active session — `_await_gateway_decision`
for Telegram / Slack / etc. (so the approval prompt lands on the right
platform), `prompt_dangerous_approval` for CLI / TUI. Fails closed on
timeout, missing notify_cb, or exception.
- The MCP tool wrapper snapshots `contextvars.copy_context()` into
`MCPServerTask._pending_call_context` before each `session.call_tool`
and clears it after. The recv-loop task that dispatches incoming
`elicitation/create` requests does not inherit the agent task's
contextvars (HERMES_SESSION_PLATFORM and friends), so without the
bridge `_is_gateway_approval_context()` returns False on every
gateway session and the elicitation falls through to a CLI prompt
that has no TTY → fail-closed decline. The handler now reads the
snapshot via its `owner` back-reference and replays it through
`Context.copy().run(...)` so attribution survives the task hop.
Tests (`tests/tools/test_mcp_elicitation.py`):
- form-mode accept / decline / cancel
- URL-mode declined without prompting
- exception in approval system → decline
- timeout in approval → cancel
- context-bridge regression tests (replay observed in consent call,
missing-context fallback, multiple-replay safety, owner with
cleared `_pending_call_context`)
Verified end-to-end against pay's MCP server on macOS: agent message
arrives via Telegram, agent calls `mcp_pay_curl` against a paid endpoint,
pay returns 402, ElicitationHandler routes the approval prompt back to
the originating Telegram chat, user replies in TG, the curl tool signs
and completes.
Platforms tested: macOS 14 (darwin/arm64). No Unix-only syscalls
introduced; Windows footgun checker passes on the touched files.
Cron no_agent and pre-check scripts ran with the full gateway/agent
environment, allowing scripts under HERMES_HOME/scripts/ to read provider
credentials. Apply _sanitize_subprocess_env like terminal and MCP paths
(SECURITY.md section 2.3).
Add regression test asserting blocklisted provider vars are absent in the
child process.
Sets the Telegram bot's short description (the line under its name) to
"Online" on gateway connect and "Offline" on clean disconnect, gated
behind extra.status_indicator (off by default).
Telegram bots have no presence/online dot — that's a user-account
feature the Bot API doesn't expose for bots. The short description is
the closest available surface, so this gives users a way to tell whether
the gateway is up from the bot's profile.
- New extra.status_indicator flag (+ status_online/status_offline text
overrides), read in __init__ via config.extra — no config-schema change.
- _set_status_indicator() helper: best-effort, swallows API errors so it
never blocks connect/disconnect; truncates to Telegram's 120-char cap.
- Wired Online after _mark_connected(), Offline at top of disconnect()
while the bot HTTP client is still alive.
- 9 unit tests + Telegram docs section.
Requested by @ilTrumpista, cc @Teknium.
The image-too-large reactive shrink (try_shrink_image_parts_in_messages)
conflated two independent constraints: it always rejected a resize whose
re-encoded bytes were >= the original, even when the shrink was driven by a
PIXEL-DIMENSION cap (Anthropic many-image 2000px) rather than the byte budget.
Downscaled screenshot PNGs routinely re-encode LARGER in bytes, so the
dimension-correct result was discarded and the image left oversized -> the
provider re-rejected on retry and the session wedged forever.
Fix: track which constraint triggered the shrink (bytes vs dimension) and gate
the accept on the SAME axis.
* dimension path: accept the result as long as it is now within max_dimension,
regardless of byte size (verify via Pillow; fall back to the byte gate only
when the re-encode can't be decoded).
* bytes path: still require bytes to shrink, but ALSO re-check the per-side cap
when it's active — _resize_image_for_vision returns a best-effort, possibly
over-cap blob when it exhausts its halving budget on a very-high-aspect
image, so a byte-shrink alone can leave it over the dimension cap and
re-brick on retry.
Extend the unshrinkable-oversized guard to the pixel axis so a partial shrink
doesn't burn the one-shot retry.
Single shared agent path -> fixes CLI, TUI, and gateway alike.
Adds a real-Pillow runnable proof (repro_48013_image_shrink_brick.py) that
reproduces the issue's per-image table (bricks 3/5 before, passes 5/5 after)
plus unit invariants for the dimension and bytes accept/reject paths,
partial-progress accounting, and the bytes-path still-over-cap regression
surfaced by adversarial review.
Closes#48013
The classic CLI status bar could appear twice after a horizontal terminal
resize — two bars at two widths with two different elapsed readings.
Root cause: prompt_toolkit's Application._on_resize() calls renderer.erase(),
which does cursor_up(_cursor_pos.y) + erase_down() using the _cursor_pos.y
cached from the LAST render at the OLD width (renderer.py:745). On a column
shrink the terminal reflows the already-painted full-width chrome into extra
physical rows, so the cached y undershoots: cursor_up doesn't climb past the
reflowed rows and erase_down leaves the old bar stranded ABOVE the live
origin. The next paint stacks a fresh bar below it. The existing post-resize
suppression hides the NEW bar for ~0.35s but never erases the already-reflowed
OLD one, so the ghost survives the whole window. Ctrl+L / /redraw clears it,
confirming a viewport wipe is the fix.
Fix: on a WIDTH change, _recover_after_resize now routes through the same
recovery as Ctrl+L — _clear_prompt_toolkit_screen(rebuild_scrollback=False)
(CSI 2J, visible viewport only) + _replay_output_history() — BEFORE delegating
to prompt_toolkit's resize. Banner-safe: 2J never touches scrollback history
(that's CSI 3J, which we don't send here), so the startup banner is preserved.
Rows-only resizes skip the clear (no reflow → no ghost) to avoid an extra
repaint. Tracks _last_resize_width to distinguish the two.
Tests: replace the now-obsolete 'never clears on resize' assertion with two
tests — rows-only resize delegates without clearing; width change clears the
viewport + replays and never wipes scrollback.
The classic CLI status bar could vanish for the rest of a session: any
terminal reflow (SIGWINCH from a tmux pane change, SSH window restore, font
zoom) set _status_bar_suppressed_after_resize=True, but the flag was ONLY
cleared on the next *submitted* user input. Resize then sit idle and the
bottom chrome rendered at height 0 on every repaint — even with the
refresh clock ticking — so the bar was gone until you typed and hit enter.
Fix: _recover_after_resize now schedules a debounced unsuppress timer that
clears the flag and repaints once the reflow settles (~0.35s), so the bar
returns on its own during idle. The next-submit clear stays as a fast path.
Fails open: any error in scheduling clears the flag immediately rather than
leaving the bar stuck hidden.
Adds a Raft platform adapter as a bundled plugin (plugins/platforms/raft/)
connecting Hermes to Raft as an external agent via a wake-channel bridge.
The adapter starts a loopback HTTP endpoint, spawns 'raft agent bridge' as a
child process, and injects content-free wake hints into the gateway session
pipeline. The agent reads/sends messages through the Raft CLI; the adapter
never touches message bodies or delivery cursors. Activity observer hooks
report tool/LLM/session lifecycle events via a bounded at-most-once queue.
Auto-enables when RAFT_PROFILE is set.
Cherry-picked from PR #47629. Authored by skyzh (@xxchan).
Manual verification surfaced a second bypass class beyond the standalone
config loaders: several code paths bridge config.yaml values into os.environ
(HERMES_TIMEZONE, HERMES_REDACT_SECRETS, HERMES_MAX_ITERATIONS, TERMINAL_*,
network.force_ipv4, ...) by reading the raw user YAML, so the env the whole
process reads carried the USER's value even when an administrator pinned it —
e.g. a managed timezone was overridden because gateway/run.py wrote the user's
timezone into HERMES_TIMEZONE, and _resolve_timezone_name() checks the env var
first.
Wired the shared apply_managed_overlay() into every config→env bridge:
- gateway/run.py module-level startup bridge (timezone, redact_secrets,
max_turns, terminal, display, gateway.strict, ...)
- gateway/run.py _reload_runtime_env_preserving_config_authority (the per-turn
re-bridge that keeps config authoritative over reloaded .env — must keep
MANAGED authoritative on every turn, not just startup)
- hermes_cli/main.py early security.redact_secrets / network.force_ipv4 bridge
(runs before load_config is usable, at import time)
- hermes_cli/send_cmd.py top-level scalar config→env bridge
Verified end-to-end against a writable managed dir (12/12 checks incl. timezone,
logging, model, skin, gateway settings, write-guard) and in a clean process the
gateway per-turn bridge writes HERMES_TIMEZONE=<managed>. Adds an
order-independent regression test for the bridge overlay.
The skin bug was one instance of a class: several subsystems build their
config dict directly from config.yaml instead of routing through
hermes_cli.config.load_config (which carries the managed merge), so they
silently ignored administrator-pinned values. Audited every config.yaml
reader and fixed the behavioral-read bypasses:
- gateway/config.py load_gateway_config (messaging gateway: session_reset,
quick_commands, stt, model, ...)
- gateway/run.py _load_gateway_config (its read_raw_config fast path also
skipped the merge — read_raw_config returns raw user YAML)
- tui_gateway/server.py _load_cfg (new TUI + desktop backend: skin,
reasoning_effort, service_tier, provider_routing)
- cron/scheduler.py (scheduled-job model/reasoning/toolsets/provider_routing)
- hermes_logging.py (logging.level/max_size_mb/backup_count)
- hermes_time.py (timezone)
- hermes_cli/doctor.py (memory-provider diagnostic reads effective config)
All route through a new shared managed_scope.apply_managed_overlay() helper
that mirrors _load_config_impl (env-only expansion so a user ${VAR} can't
shadow a managed literal, root-model-string normalization, leaf-merge) and is
fail-open. cli.py's earlier inline fix is refactored onto the same helper.
Write-back paths (slash_commands, telegram/yuanbao dm_topics, profile
distribution) are deliberately left reading raw user YAML — overlaying managed
values there would persist them into the user file. The dashboard
(web_server.py) already routes through load_config and needed no change.
TUI loader caches the RAW config so _save_cfg never writes managed values to
disk. Adds test_managed_scope_overlay.py (helper) and
test_managed_scope_loaders.py (per-surface integration); mutation-checked.
cli.py's load_cli_config() builds CLI_CONFIG independently of
hermes_cli.config._load_config_impl (it reads config.yaml directly and merges
into hardcoded defaults), so the Phase 2 managed merge never reached the
interactive CLI/TUI surface. Symptom: a managed display.skin (and any other
display/CLI pref read from CLI_CONFIG) was silently ignored by the TUI while
`hermes config`/`doctor`/write-guards — which go through load_config — correctly
honored it. Found via manual testing: the skin engine kept using 'default'.
Fix: overlay the managed config last in load_cli_config(), mirroring
_load_config_impl — expand against the process env only (so a user ${VAR} can't
shadow a managed literal), normalize the root model key so a managed
`model: x/y` string can't clobber the dict shape callers expect, then
leaf-merge. Fail-open so managed scope can never block CLI startup.
Adds tests/hermes_cli/test_managed_scope_cli_config.py locking that CLI_CONFIG
honors managed values, preserves user siblings, and is inert with no scope.
- show_config prints an administrator header naming the managed source and
lists the pinned config/env keys when a scope is active (silent otherwise).
- hermes doctor gains a managed_scope_check under Configuration Files that
reports the resolved managed dir + pinned key counts, and flags a
HERMES_MANAGED_DIR redirect (the documented foot-gun).
- set_config_value hard-rejects a managed config key (D2) and names the
source, exiting non-zero.
- save_env_value / remove_env_value refuse a managed env key.
- save_config strips managed leaves from a bulk write (mechanical safety net)
with a warning, so the unmanaged remainder still persists.
New _strip_dotted_keys helper drives the bulk-save pruning. All guards are
distinct from and layered after the existing is_managed() package-manager
write-lock.
load_hermes_dotenv now loads the managed-scope .env after user/project .env
and external secret sources, with override=True, so managed env values beat
the user .env and any pre-existing shell export. Reuses the existing dotenv
fallback + credential-sanitization path. Fail-open: no managed dir/.env is a
no-op and any error is swallowed so managed scope never blocks startup.
_load_config_impl now deep-merges the managed config.yaml on top of the
expanded user config so managed leaves win while sibling keys stay
user-controlled (leaf-level merge, D3). Managed values are expanded against
the process env only, never user-defined ${VAR}, so a user can't shadow a
managed literal. The managed file's (mtime,size) is folded into the load
cache key so editing it invalidates the cache. This inverts the usual
env-over-config precedence for pinned keys by design (see design doc §4.1).
New hermes_cli/managed_scope.py resolves a system-level managed directory
(HERMES_MANAGED_DIR override > /etc/hermes), parses managed config.yaml/.env
with fail-open semantics, and exposes is_key_managed/is_env_managed helpers.
The system default is ignored under pytest and HERMES_MANAGED_DIR is added to
the conftest env scrub so a real managed scope can't leak into the suite.
Not wired into the load paths yet (Phases 2-3).
release_stale_claims and detect_stale_running call _terminate_reclaimed_worker
and then release the task claim unconditionally, even when the termination did
not actually kill the worker. _terminate_reclaimed_worker already reports this
via its "terminated" flag, but the callers ignore it.
When a worker is parked in uninterruptible (D) state — for example throttled by
a cgroup memory.high limit — a pending SIGTERM/SIGKILL cannot be delivered until
the throttle lifts, so the kill is a no-op. The dispatcher then frees the claim
and spawns a fresh worker beside the still-alive one. Repeated every dispatch
tick this accumulates duplicate workers without bound, deepening the memory
pressure that caused the throttle in the first place — a self-reinforcing
runaway.
Fix: gate both automatic reclaim paths on _worker_survived_termination(). When
we attempted to kill our own host-local worker and it is still alive, defer the
reclaim (_defer_reclaim_for_live_worker extends the claim a short grace and
emits a reclaim_deferred event) instead of releasing. This guarantees at most
one live worker per task and is self-correcting: not spawning a duplicate is
what relieves the pressure so the pending signal lands and the worker dies, and
the next tick reclaims cleanly. Non-host-local claims and the operator-driven
reclaim_task() path keep their existing force-release behaviour.
Related: #41448 (concurrent dispatchers amplify this by doubling reclaim
frequency); #42858 (kill the worker rather than orphan it on archive).
Tests: defer-when-worker-survives, reclaim-when-killed,
release-when-not-host-local, and the detect_stale_running path.
Two robustness gaps from community review (#44919):
1. Windows dead-path: replaced bespoke fcntl.flock with gateway.status
_try_acquire_file_lock / _release_file_lock — already cross-platform
(msvcrt on Windows, fcntl on POSIX). Added _release_singleton_lock
helper.
2. Lock fd never released: stored handle is now released explicitly in
both exit paths — CancelledError handler and normal while-loop exit.
Allows in-process stop/restart (tests, embedded use).
Also tightened docstrings — 'corrupt the SQLite DBs' is now specific
(wal_autocheckpoint=0 + concurrent manual WAL checkpoints can corrupt
index pages), matching the module's own concurrency claims.
The gateway's embedded dispatcher has no guard against more than one dispatcher
running concurrently. dispatch_in_gateway defaults to true, so a second gateway
for the same profile (a restart race where the old process is slow to exit) — or
any deployment that runs multiple profile gateways with the default — starts a
second dispatcher loop. As #41448 describes, concurrent dispatchers each run
release_stale_claims() against the same boards, double reclaim frequency, and
re-dispatch slow workers before they finish. In practice they also corrupt the
shared kanban SQLite DBs under concurrent write load.
Add _acquire_singleton_lock(): an exclusive, non-blocking fcntl.flock at the
machine-global kanban root (kanban_home()/kanban/.dispatcher.lock — the board is
shared across profiles by design, so this serialises every gateway, not just one
profile). The first gateway to start its dispatcher holds the lock for its
process lifetime; any other gateway finds it contended, logs, and skips
dispatching while still running for messaging. Falls back to config-only control
on non-POSIX or filesystems without flock.
This is more robust than a per-profile guard because the documented model is
"one dispatcher sweeps all boards" — the contention is across profiles, not just
within one. Closes#41448.
Test: lock is exclusive (held, then contended while held, then held again after
release).
PR #49056 set the default to 0, which reverts the #45592 idle-clock fix:
without a periodic invalidate, prompt_toolkit stops repainting the bottom
chrome during idle and the status bar goes stale/disappears after a turn.
Restore 1.0 as the default for everyone. The config knob stays — users on
emulators where the per-second redraw fights auto-scroll (#48309) can set
display.cli_refresh_interval: 0 to opt out.
- _guard_named_profile_under_multiplexer: when the default gateway is running
with gateway.multiplex_profiles=on, a named-profile 'hermes gateway run' hard
-errors (pointing at the multiplexer) instead of double-binding that
profile's platforms. Inert unless all hold: this invocation is a named
profile, a default-profile gateway is alive, and its config has multiplexing
on. --force overrides. Wired into run_gateway's guard chain.
- write_runtime_status gains served_profiles: the secondary-adapter startup
records [active] + multiplexed profiles into runtime_status.json so
'hermes status' can show per-profile coverage without a second probe. Absent
for single-profile gateways.
Tests: served_profiles round-trips and is absent by default; guard is inert for
the default profile / under --force / when no default gateway is running.
Bring up adapters for every profile the gateway serves, not just the active
one. Keeps self.adapters as the default/active profile's map (the ~93 existing
self.adapters[...] sites are untouched) and adds secondary profiles under
self._profile_adapters[profile][platform].
- _start_secondary_profile_adapters loops profiles_to_serve(multiplex=True),
skips the active profile (handled by the primary startup loop), and for each
other profile loads its gateway config and creates+connects its enabled
adapters under that profile's _profile_runtime_scope (home + secret scope).
- Each secondary adapter gets _make_profile_message_handler(profile): stamps
source.profile (when unset) before delegating to the shared _handle_message,
so the agent turn and session key resolve to that profile.
- Same-platform credential-conflict detection: _adapter_credential_fingerprint
hashes the adapter's bot token (salted, truncated — never logs the token);
two profiles claiming the same (platform, token) refuse the duplicate with a
clear error naming both, since one token can't be polled twice.
- Port-binding hard-error: a SECONDARY profile that enables a port-binding
platform (webhook, api_server, msgraph_webhook, feishu, wecom_callback,
bluebubbles, sms) is a config error and aborts startup via MultiplexConfigError
— the default profile owns the single shared HTTP listener and serves every
profile through the /p/<profile>/ prefix, so a second bind can only collide.
Distinct from a transient connect failure (which logs + stays alive to retry):
a config error writes gateway_state=startup_failed and exits cleanly with an
actionable message (names the profile, the platform, and the fix). There is no
valid reason to bind a second port once you've opted into a multiplexer.
- Shutdown tears down secondary adapters alongside the primary ones.
- Defensive getattr guards keep partial-construction unit tests (stop(),
_run_agent on bare instances) working.
No-op when multiplex_profiles is off (self._profile_adapters stays empty).
Tests: fingerprint stability/log-safety/distinctness, profile message-handler
stamping (and not overriding an already-stamped source), port-binding hard-error
raises + names the profile/platform, non-binding platform is not rejected, and
the guard set covers every TCP-binding adapter.
