Previous handler dumped the raw skills.manage response into a pager, which
was unreadable and hid the pagination metadata. Also silently accepted
non-numeric page args.
Now:
- validates page arg (rejects NaN / <1 with a usage message)
- shows "fetching community skills (scans 6 sources, may take ~15s)…" up
front so the 10-30s hub fetch isn't a silent hang
- renders items as {name · trust, description (truncated 160 chars)} rows
in the existing Panel component
- footer shows "page X of Y · N skills total · /skills browse N+1 for more"
when the server returned pagination metadata
Skills hub's remote fetch latency is a separate upstream issue
(browse_skills hits 6 sources sequentially) — client-side we just stop
misrepresenting it.
Based on #12152 by @LVT382009.
Two fixes to run_agent.py:
1. _ephemeral_max_output_tokens consumption in chat_completions path:
The error-recovery ephemeral override was only consumed in the
anthropic_messages branch of _build_api_kwargs. All chat_completions
providers (OpenRouter, NVIDIA NIM, Qwen, Alibaba, custom, etc.)
silently ignored it. Now consumed at highest priority, matching the
anthropic pattern.
2. NVIDIA NIM max_tokens default (16384):
NVIDIA NIM falls back to a very low internal default when max_tokens
is omitted, causing models like GLM-4.7 to truncate immediately
(thinking tokens exhaust the budget before the response starts).
3. Progressive length-continuation boost:
When finish_reason='length' triggers a continuation retry, the output
budget now grows progressively (2x base on retry 1, 3x on retry 2,
capped at 32768) via _ephemeral_max_output_tokens. Previously the
retry loop just re-sent the same token limit on all 3 attempts.
Based on #11984 by @maxchernin. Fixes#8259.
Some providers (MiniMax M2.7 via NVIDIA NIM) resend the full function
name in every streaming chunk instead of only the first. The old
accumulator used += which concatenated them into 'read_fileread_file'.
Changed to simple assignment (=), matching the OpenAI Node SDK, LiteLLM,
and Vercel AI SDK patterns. Function names are atomic identifiers
delivered complete — no provider splits them across chunks, so
concatenation was never correct semantics.
Models that emit reasoning inline as <think>/<reasoning>/<thinking>/<thought>/
<REASONING_SCRATCHPAD> tags in the content field (rather than a separate API
reasoning channel) had the raw tags + inner content shown twice: once as body
text with literal <think> markers, and again in the thinking panel when the
reasoning field was populated.
Port v1's tag set to lib/reasoning.ts with a splitReasoning(text) helper that
returns { reasoning, text }. Applied in three spots:
- scheduleStreaming: strips tags from the live streaming view so the user
never sees <think> mid-turn.
- flushStreamingSegment: when a tool interrupts assistant output mid-turn,
the saved segment is the stripped text; extracted reasoning promotes to
reasoningText if the API channel hasn't already populated it.
- recordMessageComplete: final message text is split, extracted reasoning
merges with any existing reasoning (API channel wins on conflicts so we
don't double-count when both are present).
Before: textInput explicitly ignored Ctrl+C so the app-level handler took
over — with no knowledge of the TextInput's own selection — and fell through
to clearIn() whenever input had text. Selecting part of the composer and
pressing Ctrl+C silently nuked everything you typed.
Now: Ctrl+C with an active in-input selection writes the selected substring
to the clipboard via OSC 52 and clears the selection. The original semantics
(Ctrl+C with no selection → app-level interrupt/clear/die chain) are
preserved by still returning early in that case.
Pass 3 of `_prune_old_tool_results` previously shrunk long `function.arguments`
blobs by slicing the raw JSON string at byte 200 and appending the literal
text `...[truncated]`. That routinely produced payloads like::
{"path": "/foo.md", "content": "# Long markdown
...[truncated]
— an unterminated string with no closing brace. Strict providers (observed
on MiniMax) reject this as `invalid function arguments json string` with a
non-retryable 400. Because the broken call survives in the session history,
every subsequent turn re-sends the same malformed payload and gets the same
400, locking the session into a re-send loop until the call falls out of
the window.
Fix: parse the arguments first, shrink long string leaves inside the parsed
structure, and re-serialise. Non-string values (paths, ints, booleans, lists)
pass through intact. Arguments that are not valid JSON to begin with (rare,
some backends use non-JSON tool args) are returned unchanged rather than
replaced with something neither we nor the provider can parse.
Observed in the wild: a `write_file` with ~800 chars of markdown `content`
triggered this on a real session against MiniMax-M2.7; every turn after
compression got rejected until the session was manually reset.
Tests:
- 7 direct tests of `_truncate_tool_call_args_json` covering valid-JSON
output, non-JSON pass-through, nested structures, non-string leaves,
scalar JSON, and Unicode preservation
- 1 end-to-end test through `_prune_old_tool_results` Pass 3 that
reproduces the exact failure payload shape from the incident
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
renderLink was discarding the URL entirely — it rendered the label as amber
underlined text and dropped the href. Result: Cmd+Click / Ctrl+Click did
nothing in any terminal, including Ghostty.
Now both markdown links `[label](url)` and bare `https://…` URLs are wrapped
in @hermes/ink's Link component, which emits OSC 8 (\\x1b]8;;url\\x07label\\x1b]8;;\\x07)
when supportsHyperlinks() returns true. ADDITIONAL_HYPERLINK_TERMINALS already
includes ghostty, iTerm2, kitty, alacritty, Hyper.
Autolinks that look like bare emails (foo@bar.com) now prepend mailto: in the
href so they open the mail client correctly.
Also adds a typed declaration for Link in hermes-ink.d.ts.
Large inline scripts (e.g. Python code_execution bodies) rendered as a single
unbounded <Text> block, pushing the Allow/Deny options below the visible
viewport. Users had to scroll the terminal to vote.
Preview now shows the first 10 lines with truncate-end wrap per line and a
dim "… +N more lines" indicator. Full text remains in the transcript above.
* perf(docker): layer-cache npm/Playwright and skip redundant web rebuild
Copy package manifests before source so npm install + Playwright only
re-run when lockfiles change. Use COPY --chown instead of chown -R,
set HERMES_WEB_DIST to skip runtime web rebuild, and drop the
USER root / chmod dance since entrypoint.sh is already executable in git.
* Update Dockerfile
The Dockerfile installs root-level npm dependencies (for Playwright) and the
whatsapp-bridge bundle, but never builds the web/ Vite project. As a result,
'hermes dashboard' starts FastAPI on :9119 but serves a broken SPA because
hermes_cli/web_dist/ is empty and requests to /assets/index-<hash>.js 404.
Add a build step inside web/ so the Vite output is baked into the image.
Reproduce (before):
docker build -t hermes-repro -f Dockerfile .
docker run --rm -p 9119:9119 hermes-repro hermes dashboard
curl -sI http://localhost:9119/assets/ | head -1 # -> 404
After: /assets/ returns the built asset path.
* fix(kimi): force fixed temperature on kimi-k2.* models (k2.5, thinking, turbo)
The prior override only matched the literal model name "kimi-for-coding",
but Moonshot's coding endpoint is hit with real model IDs such as
`kimi-k2.5`, `kimi-k2-turbo-preview`, `kimi-k2-thinking`, etc. Those
requests bypassed the override and kept the caller's temperature, so
Moonshot returns HTTP 400 "invalid temperature: only 0.6 is allowed for
this model" (or 1.0 for thinking variants).
Match the whole kimi-k2.* family:
* kimi-k2-thinking / kimi-k2-thinking-turbo -> 1.0 (thinking mode)
* all other kimi-k2.* -> 0.6 (non-thinking / instant mode)
Also accept an optional vendor prefix (e.g. `moonshotai/kimi-k2.5`) so
aggregator routings are covered.
* refactor(kimi): whitelist-match kimi coding models instead of prefix
Addresses review feedback on PR #12144.
- Replace `startswith("kimi-k2")` with explicit frozensets sourced from
Moonshot's kimi-for-coding model list. The prefix match would have also
clamped `kimi-k2-instruct` / `kimi-k2-instruct-0905`, which are the
separate non-coding K2 family with variable temperature (recommended 0.6
but not enforced — see huggingface.co/moonshotai/Kimi-K2-Instruct).
- Confirmed via platform.kimi.ai docs that all five coding models
(k2.5, k2-turbo-preview, k2-0905-preview, k2-thinking, k2-thinking-turbo)
share the fixed-temperature lock, so the preview-model mapping is no
longer an assumption.
- Drop the fragile `"thinking" in bare` substring test for a set lookup.
- Log a debug line on each override so operators can see when Hermes
silently rewrites temperature.
- Update class docstring. Extend the negative test to parametrize over
kimi-k2-instruct, Kimi-K2-Instruct-0905, and a hypothetical future
kimi-k2-experimental name — all must keep the caller's temperature.
- /retry: use session['history'] instead of non-existent
agent.conversation_history; truncate history at last user message
to match CLI retry_last() behavior; add history_lock safety
- /plan: pass user instruction (arg) to build_plan_path instead of
session_key; add runtime_note so agent knows where to save the plan
- ANSI tool results: render full text via <Ansi wrap=truncate-end>
instead of slicing raw ANSI through compactPreview (which cuts
mid-escape-sequence producing garbled output)
- Move _PENDING_INPUT_COMMANDS frozenset to module level
- Use get_skill_commands() (cached) instead of scan_skill_commands()
(rescans disk) in slash.exec skill interception
- Add 3 retry tests: happy path with history truncation verification,
empty history error, multipart content extraction
- Update test mock target from scan_skill_commands to get_skill_commands
Additional TUI fixes discovered in the same audit:
1. /plan slash command was silently lost — process_command() queues the
plan skill invocation onto _pending_input which nobody reads in the
slash worker subprocess. Now intercepted in slash.exec and routed
through command.dispatch with a new 'send' dispatch type.
Same interception added for /retry, /queue, /steer as safety nets
(these already have correct TUI-local handlers in core.ts, but the
server-side guard prevents regressions if the local handler is
bypassed).
2. Tool results were stripping ANSI escape codes — the messageLine
component used stripAnsi() + plain <Text> for tool role messages,
losing all color/styling from terminal, search_files, etc. Now
uses <Ansi> component (already imported) when ANSI is detected.
3. Terminal tab title now shows model + busy status via useTerminalTitle
hook from @hermes/ink (was never used). Users can identify Hermes
tabs and see at a glance whether the agent is busy or ready.
4. Added 'send' variant to CommandDispatchResponse type + asCommandDispatch
parser + createSlashHandler handler for commands that need to inject
a message into the conversation (plan, queue fallback, steer fallback).
Two TUI fixes:
1. Hyperlinks are now clickable (Cmd+Click / Ctrl+Click) in terminals
that support OSC 8. The markdown renderer was rendering links as
plain colored text — now wraps them in the existing <Link> component
from @hermes/ink which emits OSC 8 escape sequences.
2. Skill slash commands (e.g. /hermes-agent-dev) now work in the TUI.
The slash.exec handler was delegating to the _SlashWorker subprocess
which calls cli.process_command(). For skills, process_command()
queues the invocation message onto _pending_input — a Queue that
nobody reads in the worker subprocess. The skill message was lost.
Now slash.exec detects skill commands early and rejects them so
the TUI falls through to command.dispatch, which correctly builds
and returns the skill payload for the client to send().
The web dashboard (Vite/React frontend) is now built as a separate Nix
derivation and baked into the Hermes package. The build output is
installed to a standard location and exposed via the `HERMES_WEB_DIST`
environment variable, allowing the dashboard command to use pre-built
assets when available (e.g., in packaged releases) instead of rebuilding
on every invocation.
Adds a minimal hand-rolled highlighter for ts/js/jsx/tsx, py, sh/bash, go, rust,
json, yaml, sql. Recognizes whole-line comments, single/double/backtick strings,
numbers, and per-language keyword sets. Unknown langs fall through to the current
plain rendering; the existing diff-specific colorization is preserved.
Closes the §8 "Markdown syntax highlighting is missing (only diff gets colored)"
finding from the TUI v2 audit without pulling in a highlighter library.
/skills install, inspect, search, browse, list now call the typed skills.manage RPC
and render results via panel/page. Previously they fell through to slash.exec which
invokes v1's curses code path — that hangs or crashes inside the Ink worker per the
§2 parity-audit finding.
Also drop Enter-as-install from the Skills Hub action stage since the Hub lists
locally installed skills; primary action is inspect-and-close. x still triggers a
manual reinstall for power users.
Intercept bare /skills locally and flip overlay.skillsHub, so the
overlay opens instantly without waiting on slash.exec. /skills <args>
still forwards to slash.exec and paginates any output. Tests cover
both branches.
New SkillsHub mirrors ModelPicker's category → item → actions flow with
paginated 12-line lists, 1-9/0 quick-pick, Esc-back navigation, and
lazy skills.manage inspect/install calls. Mount it from appOverlays
when overlay.skillsHub is true.
Extend OverlayState with a skillsHub flag, fold it into $isBlocked, and
teach Ctrl+C to close the overlay so later PRs can render the component
behind this slot.
- turnController gates scheduleStreaming / reasoning recorders on
streaming + showReasoning so disabling them keeps the buffer silent
until message.complete flushes
- createGatewayEventHandler only surfaces inline_diff previews when
inlineDiffs is on
- StatusRule takes a showCost prop and renders `· $X.XXXX` with the
same toFixed(4) formatting as /usage when usage.cost_usd is present
- Usage grows cost_usd?: number to match the gateway payload
- Existing handler tests flip showReasoning on in beforeEach so
reasoning-flow assertions keep their meaning
Extends ConfigDisplayConfig and UiState so the four new display flags
flow from `config.get {key:"full"}` into the nanostore. applyDisplay is
exported to keep the fan-out testable without an Ink harness.
Defaults mirror v1 parity: streaming + inline_diffs default true
(opt-out via `=== false`), show_cost + show_reasoning default false
(opt-in via plain truthy check).
* Add setuptools build dep for legacy alibabacloud packages and updated
stale npm-deps hash
* Add HERMES_NODE env var to pin Node.js version
The TUI requires Node.js 20+ for regex `/v` flag support (used by
string-width). Instead of relying on PATH lookup, explicitly set
HERMES_NODE to the bundled Node 22 in the Nix wrapper, and add a
fallback check in the Python code to use HERMES_NODE if available.
Also upgrade container provisioning to Node 22 via NodeSource (Ubuntu
24.04 ships Node 18 which is EOL) and add a Nix check to verify the
wrapper and Node version at build time.
* feat(steer): /steer <prompt> injects a mid-run note after the next tool call
Adds a new slash command that sits between /queue (turn boundary) and
interrupt. /steer <text> stashes the message on the running agent and
the agent loop appends it to the LAST tool result's content once the
current tool batch finishes. The model sees it as part of the tool
output on its next iteration.
No interrupt is fired, no new user turn is inserted, and no prompt
cache invalidation happens beyond the normal per-turn tool-result
churn. Message-role alternation is preserved — we only modify an
existing role:"tool" message's content.
Wiring
------
- hermes_cli/commands.py: register /steer + add to ACTIVE_SESSION_BYPASS_COMMANDS.
- run_agent.py: add _pending_steer state, AIAgent.steer(), _drain_pending_steer(),
_apply_pending_steer_to_tool_results(); drain at end of both parallel and
sequential tool executors; clear on interrupt; return leftover as
result['pending_steer'] if the agent exits before another tool batch.
- cli.py: /steer handler — route to agent.steer() when running, fall back to
the regular queue otherwise; deliver result['pending_steer'] as next turn.
- gateway/run.py: running-agent intercept calls running_agent.steer(); idle-agent
path strips the prefix and forwards as a regular user message.
- tui_gateway/server.py: new session.steer JSON-RPC method.
- ui-tui: SessionSteerResponse type + local /steer slash command that calls
session.steer when ui.busy, otherwise enqueues for the next turn.
Fallbacks
---------
- Agent exits mid-steer → surfaces in run_conversation result as pending_steer
so CLI/gateway deliver it as the next user turn instead of silently dropping it.
- All tools skipped after interrupt → re-stashes pending_steer for the caller.
- No active agent → /steer reduces to sending the text as a normal message.
Tests
-----
- tests/run_agent/test_steer.py — accept/reject, concatenation, drain,
last-tool-result injection, multimodal list content, thread safety,
cleared-on-interrupt, registry membership, bypass-set membership.
- tests/gateway/test_steer_command.py — running agent, pending sentinel,
missing steer() method, rejected payload, empty payload.
- tests/gateway/test_command_bypass_active_session.py — /steer bypasses
the Level-1 base adapter guard.
- tests/test_tui_gateway_server.py — session.steer RPC paths.
72/72 targeted tests pass under scripts/run_tests.sh.
* feat(steer): register /steer in Discord's native slash tree
Discord's app_commands tree is a curated subset of slash commands (not
derived from COMMAND_REGISTRY like Telegram/Slack). /steer already
works there as plain text (routes through handle_message → base
adapter bypass → runner), but registering it here adds Discord's
native autocomplete + argument hint UI so users can discover and
type it like any other first-class command.
- Note that /browser connect is CLI-only and won't work in gateways (WebUI, Telegram, Discord).
- Update the Chrome launch command to use a dedicated --user-data-dir, so port 9222 actually comes up even when Chrome is already running with the user's regular profile.
- Add --no-first-run --no-default-browser-check to skip the fresh-profile wizard.
- Explain why the dedicated user-data-dir matters.
Community tip via Karamjit Singh.
Co-authored-by: teknium1 <teknium@noreply.github.com>
base.py's _keep_typing refresh loop calls send_typing every ~2s while
the agent is processing. If signal-cli returns NETWORK_FAILURE for the
recipient (offline, unroutable, group membership lost), the unmitigated
path was a WARNING log every 2 seconds for as long as the agent stayed
busy — a user report showed 1048 warnings in 41 minutes for one
offline contact, plus the matching volume of pointless RPC traffic to
signal-cli.
- _rpc() accepts log_failures=False so callers can route repeated
expected failures (typing) to DEBUG while keeping send/receive at
WARNING.
- send_typing() tracks consecutive failures per chat. First failure
still logs WARNING so transport issues remain visible; subsequent
failures log at DEBUG. After three consecutive failures we skip the
RPC during an exponential cooldown (16s, 32s, 60s cap) so we stop
hammering signal-cli for a recipient it can't deliver to. A
successful sendTyping resets the counters.
- _stop_typing_indicator() clears the backoff state so the next agent
turn starts fresh.
E2E simulation against the reported 41-minute window: RPCs drop from
1230 to 45 (-96%), log lines from 1048 WARNINGs to 1 WARNING + 44
DEBUGs.
Credits kshitijk4poor (#12056) for the _rpc log_failures kwarg idea;
the broader restructure in that PR (nested per-chat loop inside
send_typing) is avoided here in favour of stateful backoff that
preserves base.py's existing _keep_typing architecture.
Stacking both features on the same event produces duplicate, delayed
notifications — delivery is async and continues firing after the process
exits, so matches on end-of-run markers (SUMMARY, DONE, PASS) arrive
after the agent has already polled/waited and moved on.
Updates both the terminal tool JSON schema description and the
terminal_tool() function docstring to make the split explicit:
- watch_patterns: mid-process signals only (errors, readiness markers,
intermediate steps you want to react to before the process exits)
- notify_on_complete: end-of-run completion signal
No behavioural change.
Twelve tests under TestCJKSearchFallback guarding:
- CJK detection across Chinese/Japanese/Korean/Hiragana/Katakana ranges
(including the full Hangul syllables block \uac00-\ud7af, to catch
the shorter-range typo from one of the duplicate PRs)
- Substring match for multi-char Chinese, Japanese, Korean queries
- Filter preservation (source_filter, exclude_sources, role_filter)
in the LIKE path — guards against the SQL-builder bug from another
duplicate PR where filter clauses landed after LIMIT/OFFSET
- Snippet centered on the matched term (instr-based substr window),
not the leading 200 chars of content
- English fast-path untouched
- Empty/no-match cases
- Mixed CJK+English queries
Also:
- hermes_state.py: LIKE-fallback snippet is now
`substr(content, max(1, instr(content, ?) - 40), 120)`, centered on
the match instead of the whole-content default. Credit goes to
@iamagenius00 for the snippet idea in PR #11517.
- scripts/release.py: add @iamagenius00 to AUTHOR_MAP so future
release attribution resolves cleanly.
Refs #11511, #11516, #11517, #11541.
Co-authored-by: iamagenius00 <iamagenius00@users.noreply.github.com>
FTS5 default tokenizer splits CJK text character-by-character, causing
multi-character queries like '记忆断裂' to return 0 results.
This fix adds a LIKE fallback: when FTS5 returns no results and the
query contains CJK characters, retry with WHERE content LIKE '%query%'.
Preserves FTS5 performance for English queries.
Fixes#11511
Follow-up to PR #11971. Documents the new code_execution.mode config
key and what each mode actually does.
- user-guide/configuration.md: add mode: project to the yaml example,
explain project vs strict and call out that security invariants are
identical across modes.
- user-guide/features/code-execution.md: new 'Execution Mode' section
with a comparison table and usage guidance; update the 'temporary
directory' note so it reflects that script.py runs in the session
CWD in project mode (staging dir stays on PYTHONPATH for imports);
drop stale 'sandboxed' framing from the intro and skill-passthrough
paragraph.
- getting-started/learning-path.md: update the one-line Code Execution
summary to match (no longer 'sandboxed environments' — the default
runs in the session's real working directory).
No code changes.
