Add a standing-goal slash command that keeps Hermes working toward a
user-stated objective across turns until it is achieved, paused, or
the turn budget runs out. Our take on the Ralph loop — cf. Codex CLI
0.128.0's /goal.
After each turn, a lightweight auxiliary-model judge call asks 'is
this goal satisfied by the assistant's last response?'. If not, and
we're under the turn budget (default 20), Hermes feeds a continuation
prompt back into the same session as a normal user message. Any real
user message preempts the continuation loop automatically.
Judge failures fail OPEN (continue) so a flaky judge never wedges
progress — the turn budget is the real backstop.
### Commands
- `/goal <text>` — set a standing goal (kicks off the first turn)
- `/goal` or `/goal status` — show current state
- `/goal pause` — pause the continuation loop
- `/goal resume` — resume (resets turn counter)
- `/goal clear` — drop the goal
Works on both CLI and gateway platforms via the central CommandDef
registry.
### Design invariants preserved
- **Prompt cache**: continuation prompts are regular user-role
messages appended to history. No system-prompt mutation, no toolset
swap.
- **Role alternation**: continuation is a user turn, never injected
mid-tool-loop.
- **Session persistence**: goal state lives in SessionDB.state_meta
keyed by `goal:<session_id>`, so `/resume` picks it up.
- **Mid-run safety**: on the gateway, `/goal status|pause|clear` are
allowed mid-run (control-plane only); setting a new goal requires
`/stop` first so we don't race a second continuation prompt against
the current turn.
### Files
- `hermes_cli/goals.py` (new, 380 lines) — GoalManager + judge + state
- `hermes_cli/commands.py` — CommandDef entry
- `hermes_cli/config.py` — `goals.max_turns` default
- `hermes_cli/web_server.py` — dashboard category merge
- `cli.py` — /goal handler + post-turn continuation hook in
process_loop
- `gateway/run.py` — /goal handler + post-turn continuation hook
wrapping _handle_message_with_agent
- `tests/hermes_cli/test_goals.py` (new, 26 tests) — judge parsing,
fail-open semantics, lifecycle, persistence, budget exhaustion
- `website/docs/reference/slash-commands.md` — docs entry
* docs(sidebar): collapse exploding skills tree to a single Skills node
The Skills sub-tree in the left sidebar expanded to 200+ entries
(22 bundled categories + 15 optional categories, every skill a page).
That's most of the nav on a first visit — docs for the actual product
get drowned in it.
Collapse the sidebar to:
Skills
godmode (hand-written spotlight)
google-workspace (hand-written spotlight)
Bundled catalog (reference/skills-catalog — table of all bundled)
Optional catalog (reference/optional-skills-catalog — table of all optional)
Per-skill pages still generate and are still reachable at their URLs;
they're linked from the two catalog tables and from the Skills overview
page. They just don't appear in the left nav anymore.
sidebars.ts goes from 649 lines to 247. generate-skill-docs.py loses
the bundled/optional sidebar render helpers.
Also picks up incidental generator output drift on current main
(comfyui skill content refresh; 4 new skill pages for
devops-kanban-orchestrator, devops-kanban-worker,
productivity-here-now, productivity-shopify; two catalog refreshes).
These are what the generator produces on main today — keeping them
committed avoids the next docs build showing 'working tree dirty'.
* docs(sidebar): drop godmode and google-workspace spotlight pages
Keep the Skills sidebar node strictly principled: two catalog links,
nothing else. There was no rule for which skills got spotlight pages
and which got auto-generated pages — just that these two happened to
be hand-written first.
Both pages still build and are still reachable at
/docs/user-guide/skills/godmode and
/docs/user-guide/skills/google-workspace. They're linked from the
catalog tables and the Skills overview page.
Sidebar Skills node now:
Skills
├── Bundled catalog
└── Optional catalog
hermes update had two interactive [Y/n] prompts with no bypass:
1. Config migration (after new env/config options are added)
2. Autostash restore (when uncommitted work was stashed before pull)
hermes uninstall already has --yes/-y; mirrors that.
Under --yes:
- Config-migrate prompt → auto-yes, migrate_config(interactive=False)
so new config fields are applied but API-key prompts are skipped
(user runs 'hermes config migrate' later for those). Matches
gateway-mode semantics.
- Stash-restore prompt → auto-yes, git stash apply runs automatically.
Closes the 'can I hermes update -y, No ! Fix' gap reported by @murelux.
Adds opt-in auto-deletion for slash-command reply messages like
"New session started!", "Restarting gateway…", "Stopped.", and
YOLO toggles. After the TTL elapses the gateway calls the adapter's
delete_message; on platforms without a delete API (everything except
Telegram today) the TTL is silently ignored and the message stays.
Requested on Twitter by @charlesmcdowell — tool-call bubbles are useful
real-time, but system notices clutter the thread once the agent finishes.
Implementation:
- EphemeralReply(str) sentinel in gateway/platforms/base.py. Subclasses
str so existing 'X' in response / response.startswith(...) checks in
tests and call sites keep working unchanged; isinstance() still
distinguishes it for the send path.
- _process_message_background and both busy-session bypass paths
(in base.py) call _unwrap_ephemeral() on the handler return, send
the unwrapped text, and schedule a detached delete task when the
TTL > 0 AND the adapter class overrides delete_message.
- display.ephemeral_system_ttl (default 0 = disabled) in DEFAULT_CONFIG.
Handler can pass ttl_seconds explicitly to override.
- Wrapped the highest-noise return sites: /new, /reset, /stop,
/yolo on/off, /restart success + "already in progress". Draining
notices and /help output left as plain strings — those are
informational and users want to read them.
Backward-compat: default TTL 0 → no scheduling, no behavior change
for existing users. Platforms without delete_message silently no-op.
When the curator consolidates skill X into umbrella Y, any cron job
that listed X in its skills field would fail to load X at run time —
the scheduler logs a warning and skips it, so the scheduled job runs
without the instructions it was scheduled to follow.
cron.jobs.rewrite_skill_refs(consolidated, pruned) now updates jobs
in-place: consolidated names route to the umbrella target (dedup
when umbrella is already present), pruned names are dropped.
agent.curator._write_run_report calls it after classification,
best-effort so a cron-side failure never breaks the curator itself.
Results are recorded in run.json (counts.cron_jobs_rewritten + full
cron_rewrites payload), a separate cron_rewrites.json for convenience
when jobs were touched, and a section in REPORT.md.
Reported by @tombielecki.
DeepSeek V4 Pro tightened thinking-mode validation and rejects empty-string
reasoning_content with HTTP 400:
The reasoning content in the thinking mode must be passed back to the API.
run_agent.py injected "" at three fallback sites — the tool-call pad in
_build_assistant_message and both injection branches of
_copy_reasoning_content_for_api (cross-provider poison guard + unconditional
thinking pad). All three now emit " " (single space), which satisfies the
non-empty check on V4 Pro without leaking fabricated reasoning.
Also upgrades stale empty-string placeholders on replay: sessions persisted
before this change have reasoning_content="" pinned at creation time; when
the active provider enforces thinking-mode echo, the replay path now rewrites
"" -> " " so existing users don't 400 on their first V4 Pro turn after
updating. Non-thinking providers still round-trip "" verbatim.
Updates 9 existing assertions + adds 2 regression tests (stale-placeholder
upgrade, non-thinking verbatim preservation).
Refs #15250, #17400.
Closes#17341.
The user-visible /compress banner and the post-compression last_prompt_tokens
writeback both counted only the raw message transcript (chars/4). With a 15KB
system prompt and 30 tool schemas (~26KB), a 4-message transcript that looks
like ~45 tokens to the transcript-only estimator is really ~10.5K tokens of
request pressure — a 234x gap.
Two user-facing consequences:
- Banner shows 'Compressing … (~45 tokens)…' while compression is actually
firing on 10K+ tokens of real pressure, confusing users about why
compression triggered (reported by @codecovenant on X; #6217).
- Post-compression last_prompt_tokens writeback omits tool schemas, so the
next should_compress() check compares real usage against a stale
underestimate — compression triggers late, potentially past the model's
context limit on small-context models (#14695).
Swap estimate_messages_tokens_rough() for estimate_request_tokens_rough()
at every user-visible banner and at the post-compression writeback.
estimate_request_tokens_rough() already existed for exactly this purpose
and includes system prompt + tool schemas.
Touched call sites:
- run_agent.py: post-compression last_prompt_tokens writeback, post-tool
call should_compress() fallback when provider usage is missing
- cli.py: /compress banner + summary
- gateway/run.py: gateway /compress banner + summary
- tui_gateway/server.py: TUI /compress status + summary
- acp_adapter/server.py: ACP /compact before/after
Left intentionally alone:
- Session-hygiene fallback and the 'no agent' /status path in gateway/run.py
— no agent instance is in scope to query for system prompt/tools, and the
existing 30-50% overestimate wobble on hygiene is safety-accepted.
- Verbose-mode 'Request size' logging — informational only, already counts
system prompt via api_messages[0].
Also relabels the feedback line from 'Rough transcript estimate' to
'Approx request size' so the metric label matches what it actually measures.
Credits: diagnoses from @devilardis (#14695) and @Jackten (#6217);
user report @codecovenant on X (2026-04-30).
Closes#14695Closes#6217
When a user types /steer <text> on an ACP session that isn't actively
running a turn (and there's no interrupted-prompt salvage available),
_cmd_steer silently appended to state.queued_prompts and replied
"No active turn — queued for the next turn". That looks identical to
/queue output even though the user never typed /queue — @EddyLeeKhane
reported this as "/steer never works, gets queued instead".
Rewrite the payload to a plain user prompt before the slash-intercept
fires, matching the gateway's idle-/steer fallthrough in
gateway/run.py ~L4898.
`hermes update` ran the config migration (11 → 17) successfully then
crashed at `agent/skill_utils.py:340` during the post-migration
skill-config prompt. User @FlockonUS reported this on Twitter.
Root cause: `get_missing_skill_config_vars` in hermes_cli/config.py
only guarded the import of `discover_all_skill_config_vars`, not the
call. Any runtime exception inside the skill scan (malformed SKILL.md,
unreadable external skill dir, etc.) propagated up through
`migrate_config` and aborted `hermes update` after the version bump.
Wrap the call in try/except so skill-config prompting — which is a
post-migration nicety — can never block the migration itself.
The initial guardrail PR consolidated failure classification by pointing
display._detect_tool_failure at the new classify_tool_failure helper,
which was strictly broader: it flagged any JSON result with
"success": false / "failed": true / non-empty "error", plus plain-text
"traceback" and "error:" prefixes. That would uptick the user-visible
[error] tag on tools that return {"success": false} as a benign signal
(memory fullness, todo state, etc.) and feed the failure-streak counter
at the same time.
Restore display._detect_tool_failure to its pre-PR semantics verbatim.
Tighten classify_tool_failure (the guardrail's internal safety-fallback
used only when callers don't pass failed=) to match _detect_tool_failure
exactly, so the two never disagree. Production callers in run_agent.py
already pass an explicit failed= derived from _detect_tool_failure, so
the guardrail counter is driven by the same signal the CLI shows.
_process_message_background snapshotted callback_generation from the
interrupt event at the TOP of the task — before the handler ran.
_hermes_run_generation is only set on the event by
GatewayRunner._bind_adapter_run_generation during
_handle_message_with_agent, which runs DURING the handler await. The
early snapshot always captured None, which then flowed into
pop_post_delivery_callback(..., generation=None) in the finally block.
In pop_post_delivery_callback, generation=None with a tuple-registered
entry (generation, callback) bypasses the ownership check — it pops and
fires the callback regardless of which run owns it. Result: a stale run
could fire a fresher run's post-delivery callback (e.g. a
background-review notification attributed to the wrong turn).
Fix: move the snapshot into the finally block, after the handler has
run and _hermes_run_generation has been bound to the current run.
Regression test added: simulates a stale handler at generation=1 and a
fresher callback registered at generation=2. Pre-fix: snapshot=None →
pop fires the generation=2 callback under generation=1's ownership
("newer" fires). Post-fix: snapshot=1 → pop skips the mismatched
entry, callback stays in the dict for the correct run to claim.
Verified: test FAILS on current main (captures "newer" in fired list),
PASSES with this fix.
Salvaged from PR #12565 (the callback-ownership portion only; the
/status totals portion was already fixed on main in 7abc9ce4d via #17158).
Co-authored-by: Oxidane-bot <1317078257maroon@gmail.com>
Widens #16528 to two sibling sites that had the same quoted-boolean
bug: a YAML string "false" (or "0", "no", "off") silently evaluated
truthy under bool() / if-check.
- gateway/run.py _load_show_reasoning: is_truthy_value wrap
- tools/skill_manager_tool.py _guard_agent_created_enabled: is_truthy_value wrap
- regression tests for both
SELECT in get_messages_as_conversation() was missing finish_reason, so
assistant messages round-tripped through replay (including /branch copies)
silently dropped the provider's stop signal. Adds it to the SELECT, restores
it on assistant rows, and locks it in with a round-trip test.
When running on a host with sudoers NOPASSWD configured for the current
user, interactive Hermes sessions were unnecessarily entering the
password prompt path before executing sudo commands. Outside Hermes,
`sudo -n true` exits 0 for that user.
Add `_sudo_nopasswd_works()` that probes `sudo -n true` and, when it
succeeds, lets `_transform_sudo_command()` return the command unchanged
with no stdin password. The probe:
- Is scoped to the `local` terminal backend only, so Docker/SSH/Modal
and other remote backends do not inherit host sudo state.
- Re-probes every call (no process-lifetime cache) so an expired sudo
timestamp cannot silently make a later command block waiting for a
password that Hermes never prompts for.
- Is bypassed entirely when `SUDO_PASSWORD` is configured or a cached
password already exists, preserving existing explicit-password flows.
Co-authored-by: Junting Wu <juntingpublic@gmail.com>
The fix for this bug (isinstance guard) was merged via commit 3ff9e010,
but test coverage was not included. Adding 4 tests:
- dict metadata with hermes keys (normal case)
- string metadata (bug case — previously caused AttributeError)
- None metadata
- missing metadata key
Proves token A's detected capabilities do not leak to token B after the
fix in the preceding commit. Before the fix this test would have seen
both tokens return token A's cached value.
_capability_cache was a single module-level dict shared across all
tokens. If the bot token rotates or multiple tokens are used in one
process, capabilities detected for token A would be returned for
token B, causing wrong schema gating and incorrect runtime behavior.
Replace the single Optional cache with a Dict keyed by token so each
token gets its own isolated capability entry.
_SupervisorRegistry.get_or_start() returned an existing supervisor
whenever the cdp_url matched, without checking if the supervisor's
thread or event loop was still alive. A crashed supervisor would be
silently reused, causing missed dialog/frame updates.
Now checks both _thread.is_alive() and _loop.is_running() before
returning the cached instance. An unhealthy supervisor is torn down
and recreated, matching the existing URL-changed code path.
_get_peer() and _get_or_create_honcho_session() accessed _peers_cache
and _sessions_cache without holding _cache_lock, while other paths
in the same class use the lock consistently. Under concurrent tool
calls or prefetch threads, this can produce stale reads or lost
cache updates.
Wrap both unguarded cache read sites in _cache_lock. Network calls
(honcho.peer() and honcho.session()) remain outside the lock to
avoid holding it during I/O.
Three int() calls in HonchoClient.from_global_config() parsed
dialecticMaxChars, messageMaxChars, and dialecticMaxInputChars
directly without guards. A malformed value in honcho.json would
raise ValueError and abort provider initialization entirely.
Add _parse_int_config() helper following the existing
_parse_context_tokens() pattern, and replace all three raw
int() calls with it.
Add two operator-facing toggles for inbound Feishu admission, enabling
bot-to-bot scenarios such as A2A orchestration and inter-bot
notifications:
FEISHU_ALLOW_BOTS=none|mentions|all (default: none)
Accept messages from other bots. `mentions` requires the peer
bot to @-mention Hermes; `all` admits every peer-bot message.
FEISHU_REQUIRE_MENTION=true|false (default: true)
Whether group messages must @-mention the bot. Override per-chat
via `group_rules.<chat_id>.require_mention` in config.yaml.
Defaults preserve prior behavior. Self-echo protection is always on:
when the bot's identity is unresolved (auto-detection failed and
FEISHU_BOT_OPEN_ID unset), peer-bot messages are rejected fail-closed
to avoid feedback loops.
Admitted peer bots bypass the human-user allowlist
(FEISHU_ALLOWED_USERS) to match existing Discord behavior; humans
still need an explicit allowlist entry. yaml feishu.allow_bots is
bridged to the env var so the adapter and gateway auth layer share
one source of truth.
Resolving peer-bot display names requires the
application:bot.basic_info:read scope; without it, peers still route
but appear as their open_id.
Test: tests/gateway/test_feishu_bot_admission.py covers the admission
pipeline, group-policy bot-bypass, hydration, and event-dispatch
plumbing as a parametrized matrix.
Change-Id: I363cccb578c2a5c8b8bf0f0a890c01c89909e256
reset_session() creates a fresh SessionEntry with created_at == updated_at,
but get_or_create_session() bumps updated_at on the next inbound message,
causing _is_new_session in _handle_message_with_agent to evaluate False.
The topic/channel skill auto-load gate (group_topics, channel_skill_bindings)
silently skips the first message after a manual reset.
Add an is_fresh_reset flag on SessionEntry, set by reset_session() and
consumed once by the message handler. Kept distinct from was_auto_reset
because that flag also drives a 'session expired due to inactivity'
user-facing notice and a context-note prepend — both wrong for an
explicit /new or /reset.
Persisted through to_dict/from_dict so the flag survives gateway
restart between /reset and the next message.
Fixes#6508
Co-authored-by: warabe1122 <45554392+warabe1122@users.noreply.github.com>
Co-authored-by: willy-scr <187001140+willy-scr@users.noreply.github.com>
/status was reading session_entry.total_tokens from the in-memory
SessionStore (gateway/session.py), which the agent never writes to —
so the token count was always 0.
The agent already persists token deltas to the SQLite SessionDB
(run_agent.py:11497) for every platform with a session_id. Route
/status through that single source of truth instead of duplicating
token writes into a second store.
Fix:
- gateway/run.py: _handle_status_command now calls
self._session_db.get_session(session_id) and sums the five token
component columns (input/output/cache_read/cache_write/reasoning).
Falls back to 0 when no SessionDB is configured or no row exists.
- Two new regression tests covering the populated-row and
missing-row paths.
Co-authored-by: Hermes <127238744+teknium1@users.noreply.github.com>
Minor follow-up to the native-image-buffer isolation fix. The write site
in _prepare_inbound_message_text was calling build_session_key directly,
while every other call site in gateway/run.py uses the _session_key_for_source
helper — which consults session_store._generate_session_key first and falls
back to build_session_key. Keeping the write key and consume key on the
same helper prevents key drift if the session store ever overrides the
default keying behavior.
_SLASH_WORKER_TIMEOUT_S and _pool used raw float()/int() on env vars
at module level. A non-numeric value (e.g. HERMES_TUI_SLASH_TIMEOUT_S=abc)
raises ValueError during import, preventing TUI gateway from starting
with no useful error message.
Wrap both parses in try/except with safe fallbacks:
- HERMES_TUI_SLASH_TIMEOUT_S: fallback to 45.0s
- HERMES_TUI_RPC_POOL_WORKERS: fallback to 4 workers
sqlite3 can only bind str/bytes/int/float/None to query parameters.
Multimodal message content is a list of parts (text + image_url), which
raised 'Error binding parameter 3: type list is not supported' in
append_message and replace_messages.
In the CLI/TUI this surfaced as a visible crash when users pasted
screenshots. In the gateway it was silently swallowed by a bare except
in append_to_transcript, causing multimodal turns to be lost from the
session transcript.
Fix at the DB layer: _encode_content wraps lists/dicts as
'\\x00json:' + json.dumps(...) on write, _decode_content unwraps on
read. Plain strings are untouched, so existing FTS search, previews,
and JSONL compat are unaffected. Paired decode in get_messages,
get_messages_as_conversation, and search_messages context previews.
Regression test covers: list content round-trip, dict content
round-trip, string content stored unchanged, replace_messages with
multimodal content.
Also included: aligned fix#17522 for TUI image attachment with
paths containing spaces (see previous commit).
Remove frontend regex pre-check that truncated paths containing spaces,
quotes, or Windows drive letters. Backend _detect_file_drop correctly
handles these patterns. This fixes image attachment for common filenames
like "Screenshot 2026-04-29.png".
Add tests:
- test_input_detect_drop_path_with_spaces: attaches image with spaces in name
- test_input_detect_drop_path_with_spaces_and_remainder: remainder handling
Also restored missing in test_rollback_restore_resolves_number_and_file_path.
Scope: tui, vision, tests
Widens the cherry-picked fix from @jatingodnani (#17343) to the
gateway path. On main, user_config.agent.disabled_toolsets was only
honored by _get_platform_tools' name-level subtraction — it did not
catch tools pulled in implicitly by a composite toolset (browser
includes web_search, hermes-* platforms include most tools).
Changes:
- gateway/run.py: resolve disabled_toolsets alongside enabled_toolsets
and pass to AIAgent at both user-facing construction sites (normal
message loop + single-turn cron-like path). Hygiene/compression
agents (fixed enabled_toolsets=[memory]) are intentionally untouched.
- gateway/run.py: add (agent, disabled_toolsets) to
_CACHE_BUSTING_CONFIG_KEYS so editing the list in config.yaml
invalidates the cached AIAgent on the next message.
- cli.py: drop unused 'import platform' left over from PR #17343's
import churn; restore 'import sys' used throughout the file.
- model_tools.py: drop unused 'import os, sys' added by PR #17343;
fix comment reference from #15291 (unrelated OAuth issue) to #17309.
Co-authored-by: jatin godnani <godnanijatin@gmail.com>
Refactor tool resolution logic in model_tools.py to ensure that
disabled_toolsets are always subtracted at the end, preventing
composite toolsets (e.g. 'browser') from implicitly enabling tools
that should be hidden.
- Added 'disabled_toolsets' to DEFAULT_CONFIG in hermes_cli/config.py
- Updated HermesCLI in cli.py to load and propagate disabled toolsets to AIAgent
- Implemented robust two-phase resolution (additive then subtractive) in model_tools.py
Themes previously embedded layout-affecting values (baseSize, lineHeight,
density, letterSpacing) alongside visual identity properties, coupling
user ergonomic preferences to color theme selection.
This change establishes a clear separation of concerns:
- Themes own: palette, font family, border-radius, and font-coupled
letterSpacing (e.g. Inter's -0.005em tracking)
- Layout scale (baseSize, lineHeight, density) is standardized via
DEFAULT_TYPOGRAPHY and DEFAULT_LAYOUT — not overridden per theme
All themes now spread DEFAULT_TYPOGRAPHY and DEFAULT_LAYOUT as their
base, removing silent divergence and making future layout settings
(e.g. user-configurable density) trivially applicable across all themes
without per-theme special-casing.
All built-in themes now spread DEFAULT_TYPOGRAPHY, removing independent
baseSize overrides and converging on 15px. All themes also use
density: comfortable, removing the compact/spacious divergence that
caused item-count shifts on fixed-height pages (e.g. Skills).
Two additional per-theme overrides are also normalized:
- rose: lineHeight: "1.7" removed — was paired with density: spacious
for an airy feel; once density was normalised the elevated line-height
became an orphaned artefact causing nav item height drift.
- cyberpunk: letterSpacing changed from "0.02em" to "0" — extra tracking
on top of an already-wide monospace font caused text to wrap earlier
than in other themes.
Switching themes is now a purely cosmetic change — color palette,
font family, border-radius, and typographic style differ; font size,
spacing, line-height, and letter-spacing do not.
- Move the disabled-ack guard above the debounce so we don't stamp
_busy_ack_ts[session_key] when no ack was actually sent. Harmless
(never read when disabled) but cosmetically off.
- Document display.busy_ack_enabled in user-guide/messaging/index.md
and HERMES_GATEWAY_BUSY_ACK_ENABLED in reference/environment-variables.md.
- Add JezzaHehn to scripts/release.py AUTHOR_MAP for contributor credit.
Follow-up to #17491 (Jezza Hehn).
