Problem:
When users execute `/think off`, they still receive `reasoning_content`
from models configured with `reasoning: true` (e.g., GLM-4.7, GLM-4.6,
Kimi K2.5, MiniMax-M2.1).
Expected: `/think off` should completely disable reasoning content.
Actual: Reasoning content is still returned.
Root Cause:
The directive handlers delete `sessionEntry.thinkingLevel` when user
executes `/think off`. This causes the thinking level to become undefined,
and the system falls back to `resolveThinkingDefault()`, which checks the
model catalog and returns "low" for reasoning-capable models, ignoring the
user's explicit intent.
Why We Must Persist "off" (Design Rationale):
1. **Model-dependent defaults**: Unlike other directives where "off" means
use a global default, `thinkingLevel` has model-dependent defaults:
- Reasoning-capable models (GLM-4.7, etc.) → default "low"
- Other models → default "off"
2. **Existing pattern**: The codebase already follows this pattern for
`elevatedLevel`, which persists "off" explicitly to override defaults
that may be "on". The comment explains:
"Persist 'off' explicitly so `/elevated off` actually overrides defaults."
3. **User intent**: When a user explicitly executes `/think off`, they want
to disable thinking regardless of the model's capabilities. Deleting the
field breaks this intent by falling back to the model's default.
Solution:
Persist "off" value instead of deleting the field in all internal directive handlers:
- `src/auto-reply/reply/directive-handling.impl.ts`: Directive-only messages
- `src/auto-reply/reply/directive-handling.persist.ts`: Inline directives
- `src/commands/agent.ts`: CLI command-line flags
Gateway API Backward Compatibility:
The original implementation incorrectly mapped `null` to "off" in
`sessions-patch.ts` for consistency with internal handlers. This was a
breaking change because:
- Previously, `null` cleared the override (deleted the field)
- API clients lost the ability to "clear to default" via `null`
- This contradicts standard JSON semantics where `null` means "no value"
Restored original null semantics in `src/gateway/sessions-patch.ts`:
- `null` → delete field, fall back to model default (clear override)
- `"off"` → persist explicit override
- Other values → normalize and persist
This ensures backward compatibility for API clients while fixing the `/think off`
issue in internal handlers.
Signed-off-by: Liu Yuan <namei.unix@gmail.com>
* Config: reload dotenv before env substitution on runtime loads
* Test: isolate config env var regression from host state env
* fix: keep dotenv vars resolvable on runtime config reloads (#12748) (thanks @rodrigouroz)
---------
Co-authored-by: Tak Hoffman <781889+Takhoffman@users.noreply.github.com>
* fix(web_search): Fix invalid model name sent to Perplexity
* chore: Only apply fix to direct Perplexity calls
* fix(web_search): normalize direct Perplexity model IDs
* fix: add changelog note for perplexity model normalization (#12795) (thanks @cdorsey)
* fix: align tests and fetch type for gate stability (#12795) (thanks @cdorsey)
* chore: keep #12795 scoped to web_search changes
---------
Co-authored-by: Sebastian <19554889+sebslight@users.noreply.github.com>
* TypeScript: add extensions to tsconfig and fix type errors
- Add extensions/**/* to tsconfig.json includes
- Export ProviderAuthResult, AnyAgentTool from plugin-sdk
- Fix optional chaining for messageActions across channels
- Add missing type imports (MSTeamsConfig, GroupPolicy, etc.)
- Add type annotations for provider auth handlers
- Fix undici/fetch type compatibility in zalo proxy
- Correct ChannelAccountSnapshot property usage
- Add type casts for tool registrations
- Extract usage view styles and types to separate files
* TypeScript: fix optional debug calls and handleAction guards
Telegram DM access-control incorrectly used chatId as the allowFrom match key.
For DMs, allowFrom entries are typically Telegram user ids (msg.from.id) and/or @usernames. Using chatId causes legitimate DMs to be rejected or silently dropped even when dmPolicy is open/allowlist.
This change matches allowFrom against the sender's user id when available, falling back to chatId only if msg.from.id is missing.
Tests: existing telegram DM/thread routing tests pass.
Closes#4515
Telegram Bot API limits setMyCommands to 100 commands per scope. When
users have many skills installed (~15+), the combined native + plugin +
custom commands can exceed this limit, causing a 400 error on every
gateway restart.
Truncate the command list to 100 (native commands first, then plugins,
then custom) and log a warning instead of failing the registration.
Fixes#11567
- Create shared PNG encoder module (src/media/png-encode.ts)
- Refactor qr-image.ts and live-image-probe.ts to use shared encoder
- Add safeParseJson to utils.ts and plugin-sdk exports
- Update msteams and pairing-store to use centralized safeParseJson
* refactor: consolidate duplicate utility functions
- Add escapeRegExp to src/utils.ts and remove 10 local duplicates
- Rename bash-tools clampNumber to clampWithDefault (different signature)
- Centralize formatError calls to use formatErrorMessage from infra/errors.ts
- Re-export formatErrorMessage from cli/cli-utils.ts to preserve API
* refactor: consolidate remaining escapeRegExp duplicates
* refactor: consolidate sleep, stripAnsi, and clamp duplicates
* Memory/QMD: symlink default model cache into custom XDG_CACHE_HOME
QmdMemoryManager overrides XDG_CACHE_HOME to isolate the qmd index
per-agent, but this also moves where qmd looks for its ML models
(~2.1GB). Since models are installed at the default location
(~/.cache/qmd/models/), every qmd invocation would attempt to
re-download them from HuggingFace and time out.
Fix: on initialization, symlink ~/.cache/qmd/models/ into the custom
XDG_CACHE_HOME path so the index stays isolated per-agent while the
shared models are reused. The symlink is only created when the default
models directory exists and the target path does not already exist.
Includes tests for the three key scenarios: symlink creation, existing
directory preservation, and graceful skip when no default models exist.
* Memory/QMD: skip model symlink warning on ENOENT
* test: stabilize warning-filter visibility assertion (#12114) (thanks @tyler6204)
* fix: add changelog entry for QMD cache reuse (#12114) (thanks @tyler6204)
* fix: handle plain context-overflow strings in compaction detection (#12114) (thanks @tyler6204)
Add xAI's Grok as a new web_search provider alongside Brave and Perplexity.
Uses the xAI /v1/responses API with tools: [{type: "web_search"}].
Configuration:
- tools.web.search.provider: "grok"
- tools.web.search.grok.apiKey or XAI_API_KEY env var
- tools.web.search.grok.model (default: grok-4-1-fast)
- tools.web.search.grok.inlineCitations (optional, embeds markdown links)
Returns AI-synthesized answers with citations similar to Perplexity.
* fix(cron): recover flat params when LLM omits job wrapper (#11310)
Non-frontier models (e.g. Grok) flatten job properties to the top level
alongside `action` instead of nesting them inside the `job` parameter.
The opaque schema (`Type.Object({}, { additionalProperties: true })`)
gives these models no structural hint, so they put name, schedule,
payload, etc. as siblings of action.
Add a flat-params recovery step in the cron add handler: when
`params.job` is missing or an empty object, scan for recognised job
property names on params and construct a synthetic job object before
passing to `normalizeCronJobCreate`. Recovery requires at least one
meaningful signal field (schedule, payload, message, or text) to avoid
false positives.
Added tests:
- Flat params with no job wrapper → recovered
- Empty job object + flat params → recovered
- Message shorthand at top level → inferred as agentTurn
- No meaningful fields → still throws 'job required'
- Non-empty job takes precedence over flat params
* fix(cron): floor nowMs to second boundary before croner lookback
Cron expressions operate at second granularity. When nowMs falls
mid-second (e.g. 12:00:00.500) and the pattern targets that exact
second (like '0 0 12 * * *'), a 1ms lookback still lands inside the
matching second. Croner interprets this as 'already past' and skips
to the next occurrence (e.g. the following day).
Fix: floor nowMs to the start of the current second before applying
the 1ms lookback. This ensures the reference always falls in the
*previous* second, so croner correctly identifies the current match.
Also compare the result against the floored nowSecondMs (not raw nowMs)
so that a match at the start of the current second is not rejected by
the >= guard when nowMs has sub-second offset.
Adds regression tests for 6-field cron patterns with specific seconds.
* fix: add changelog entries for cron fixes (#12124) (thanks @tyler6204)
* test: stabilize warning filter emit assertion (#12124) (thanks @tyler6204)
* feat: Implement Telegram video note support with tests and docs
* fixing lint
* feat: add doctor-state-integrity command, Telegram messaging, and PowerShell Docker setup scripts.
* Update src/telegram/send.video-note.test.ts
Co-authored-by: greptile-apps[bot] <165735046+greptile-apps[bot]@users.noreply.github.com>
* fix: Set video note follow-up text to undefined for empty input and adjust caption test expectation.
* test: add assertion for `sendMessage` with reply markup and HTML parse mode in `send.video-note` test.
* docs: add changelog entry for Telegram video notes
---------
Co-authored-by: Evgenii Utkin <thewulf7@gmail.com>
Co-authored-by: greptile-apps[bot] <165735046+greptile-apps[bot]@users.noreply.github.com>
Co-authored-by: CLAWDINATOR Bot <clawdinator[bot]@users.noreply.github.com>
* fix(skills): ignore Python venvs and caches in skills watcher
Add .venv, venv, __pycache__, .mypy_cache, .pytest_cache, build, and
.cache to the default ignored patterns for the skills watcher.
This prevents file descriptor exhaustion when a skill contains a Python
virtual environment with tens of thousands of files, which was causing
EBADF spawn errors on macOS.
Fixes#1056
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
* docs: add changelog entry for skills watcher ignores
* docs: fill changelog PR number
---------
Co-authored-by: Kyle Howells <freerunnering@gmail.com>
Co-authored-by: Claude Opus 4.5 <noreply@anthropic.com>
Co-authored-by: CLAWDINATOR Bot <clawdinator[bot]@users.noreply.github.com>
* Gateway: preserve Pi transcript parentId for injected messages
Thread: unknown
When: 2026-02-08 20:08 CST
Repo: https://github.com/openclaw/openclaw.git
Branch: codex/wip/2026-02-09/compact-post-compaction-parentid-fix
Problem
- Post-compaction turns sometimes lost the compaction summary + kept suffix in the *next* provider request.
- Root cause was session graph corruption: gateway appended "stopReason: injected" transcript lines via raw JSONL writes without `parentId`.
- Pi's `SessionManager.buildSessionContext()` walks the `parentId` chain from the current leaf; missing `parentId` can sever the active branch and hide the compaction entry.
Fix
- Use `SessionManager.appendMessage(...)` for injected assistant transcript writes so `parentId` is set to the current leaf.
- Route `chat.inject` through the same helper to avoid duplicating the broken raw JSONL append logic.
Why This Matters
- The compaction algorithm may be correct, but if the leaf chain is broken right after compaction, the provider payload cannot include the summary/suffix "shape" Pi expects.
Testing
- pnpm test src/agents/pi-embedded-helpers.post-compaction-shape.test.ts src/agents/pi-embedded-runner/run.overflow-compaction.post-context.test.ts
- pnpm build
Notes
- This is provider-shape agnostic: it fixes transcript structure so Anthropic/Gemini/etc all see the same post-compaction context.
Resume
- If post-compaction looks wrong again, inspect the session transcript for entries missing `parentId` immediately after `type: compaction`.
* Gateway: guardrail test for transcript parentId (chat.inject)
* Gateway: guardrail against raw transcript appends (chat.ts)
* Gateway: add local AGENTS.md note to preserve Pi transcript parentId chain
* Changelog: note gateway post-compaction amnesia fix
* Gateway: store injected transcript messages with valid stopReason
* Gateway: use valid stopReason in injected fallback
* test: normalize paths in OPENCLAW_HOME tests for cross-platform support
* test: normalize paths in Nix integration tests for cross-platform support
* test: remove unnecessary Windows skip from pi-embedded-runner test
* test: fix nix integration tests for path.resolve behavior
* fix(paths): structurally resolve home dir to prevent Windows path bugs
Extract resolveRawHomeDir as a private function and gate the public
resolveEffectiveHomeDir through a single path.resolve() exit point.
This makes it structurally impossible for unresolved paths (missing
drive letter on Windows) to escape the function, regardless of how
many return paths exist in the raw lookup logic.
Simplify resolveRequiredHomeDir to only resolve the process.cwd()
fallback, since resolveEffectiveHomeDir now returns resolved values.
Fix shortenMeta in tool-meta.ts: the colon-based split for file:line
patterns (e.g. file.txt:12) conflicts with Windows drive letters
(C:\...) because indexOf(":") matches the drive colon first.
shortenHomeInString already handles file:line patterns correctly via
split/join, so the colon split was both unnecessary and harmful.
Update test assertions across all affected files to use path.resolve()
in expected values and input strings so they match the now-correct
resolved output on both Unix and Windows.
Fixes#12119
* fix(changelog): add paths Windows fix entry (#12125)
---------
Co-authored-by: Sebastian <19554889+sebslight@users.noreply.github.com>
