Summary:
- This PR changes pinned-session `/status` guidance, model-selection docs, and status tests to recommend `/model default` instead of `/model <configured>` or `/reset` for clearing a session model pin.
- PR surface: Source 0, Tests 0, Docs +4. Total +4 across 7 files.
- Reproducibility: yes. from source inspection. Current main and v2026.6.6 emit the old `/reset` hint, while `/model default` clears persisted model overrides and `/reset` intentionally preserves user-selected overrides.
Automerge notes:
- PR branch already contained follow-up commit before automerge: docs: align model clear hint docs
- PR branch already contained follow-up commit before automerge: fix(status): correct pinned model clear hint
Validation:
- ClawSweeper review passed for head 1181624daa.
- Required merge gates passed before the squash merge.
Prepared head SHA: 1181624daa
Review: https://github.com/openclaw/openclaw/pull/93231#issuecomment-4706327717
Co-authored-by: masonxhuang <masonxhuang@tencent.com>
Co-authored-by: Mason Huang <masonxhuang@tencent.com>
Co-authored-by: clawsweeper <274271284+clawsweeper[bot]@users.noreply.github.com>
Co-authored-by: clawsweeper[bot] <274271284+clawsweeper[bot]@users.noreply.github.com>
Approved-by: hxy91819
Co-authored-by: hxy91819 <8814856+hxy91819@users.noreply.github.com>
The standalone loopback HTTP API only starts when
OPENCLAW_EAGER_BROWSER_CONTROL_SERVER=1 is set in the gateway
service environment. Without it, browser control works via CLI and
agent tools but nothing listens on the loopback control port.
Fixes#92841
* feat(moonshot): add Kimi K2.7 Code support
* test(moonshot): surface K2.7 live provider errors
* ci(live): accept Kimi key for Moonshot sweeps
* test(moonshot): verify K2.7 across API regions
Fix MiniMax-M3 Anthropic-compatible requests so OpenClaw no longer sends the disabled-thinking payload that makes M3 return empty content. M3 defaults now stay on MiniMax's omitted/adaptive thinking path, explicit `/think off` is still respected, and MiniMax-M2.x keeps the disabled-thinking default that prevents reasoning_content leaks.
Also wires the MiniMax thinking policy through bundled provider-policy loading so pre-runtime and configless embedded-agent paths resolve the same defaults.
Thanks @IamVNIE for the live MiniMax API repro and initial patch.
* feat(parallel): add free Parallel Search MCP as the zero-config default web_search provider
Registers two Parallel web_search providers in the parallel plugin:
- parallel-free: keyless, always the free hosted Search MCP (search.parallel.ai/mcp);
the zero-config default (autoDetectOrder 76) so web_search works with no key.
- parallel: the existing paid v1 REST API (requires PARALLEL_API_KEY).
Shared query/result normalization lives in parallel-search-normalize.ts (used by both
transports); a minimal Streamable-HTTP JSON-RPC client (parallel-mcp-search.runtime.ts)
backs the free path. UI brands the tool-call chip 'Parallel Web Search' on the free path
via a searchTransport marker; setup default mirrors runtime auto-detect.
* chore(parallel): register parallel-free in doctor legacy-web-search owners
parallel-free is a bundled web_search provider, so add it to the doctor's
exhaustive BUNDLED_LEGACY_WEB_SEARCH_OWNERS map (owned by the parallel plugin)
and the NON_MIGRATED set — it has no legacy tools.web.search.* shape, so this is
a no-op for migration, matching paid parallel/tavily. Keeps the registry
complete. (Spotted by diffing the earlier local WIP branch.)
* docs(parallel): restore concise frontmatter summary
* docs(parallel): clearer, professional copy; drop v1 REST jargon and UI-label claim
- Frame the two providers as Parallel Search (Free) vs paid Parallel Search;
remove internal 'v1 REST API' wording.
- Remove conversational/overstated phrasing ('out of the box for everyone').
- Remove the 'labeled Parallel Web Search in the UI' claim (only renders in the
Control UI, not the TUI). Scope the searchTransport code comment accordingly.
* revert(parallel): drop the "Parallel Web Search" tool-call branding
The label only rendered in the Control UI, never the TUI (a separate renderer
via src/agents/tool-display.ts). Extending it would put provider-specific
labeling into a shared/core display path, against the plugin-agnostic-core rule.
Reverts the Control-UI labelOverride wiring and removes the now-orphaned
searchTransport marker from the free provider's result. The result still carries
provider: "parallel-free".
* fix(parallel): cap free Search MCP session_id at its 100-char tools/list contract
The free parallel-free provider reused the paid ParallelSearchSchema, whose
session_id allows 1000 chars, but the live Search MCP tools/list schema caps
session_id at 100. Parameterize normalizeParallelSessionId(value, maxLength);
the free path passes 100 (paid keeps 1000) and advertises the tighter bound in
its own ParallelFreeSearchSchema. An over-limit caller id is dropped and a
fresh in-contract id is minted. Updates tests and docs accordingly.
- New extensions/parallel package modeled on extensions/exa
- Wires Parallel's POST /v1/search through the generic web_search contract,
exposing Parallel's recommended {objective, search_queries} shape (plus
optional count, session_id, client_model) so the model can supply both the
natural-language goal and 2-3 short keyword queries as Parallel docs advise
- client_model lets the model report its own slug so Parallel can tailor
optimizations for the consuming model's capabilities; partitions the cache
by client_model so different models do not silently share ranked excerpts
- Honors top-level tools.web.search.{maxResults,timeoutSeconds,cacheTtlMinutes}
via the shared SDK helpers (mergeScopedSearchConfig, withTrustedWebSearchEndpoint,
buildSearchCacheKey, read/writeCachedSearchPayload)
- Auto-detect order 75; auth via PARALLEL_API_KEY or
plugins.entries.parallel.config.webSearch.apiKey
- Optional baseUrl override for proxies (e.g. Cloudflare AI Gateway)
- Threads caller-supplied session_id through follow-up calls; strips
auto-generated session_id from the shared cache to avoid cross-task leaks
- Always sends advanced_settings.max_results so result volume matches the
OpenClaw web_search default (5) instead of Parallel's default (10)
- Identifies the plugin via User-Agent header built from package version
- Runtime accepts the generic `query` arg as a fallback so the operator
CLI (openclaw capability web.search) keeps working when Parallel is the
active provider: it is promoted into the lone `search_queries` entry.
`objective` stays optional and is never synthesized from a keyword
query (Parallel documents it as natural-language intent). Agent callers
using the native objective+search_queries shape take precedence; the
schema still advertises only the native keys
- Updates the agent tool-display extractor (src/agents/tool-display-common.ts)
to recognize Parallel's objective+search_queries shape so calls render with
query context in CLI progress and Codex activity metadata
- Adds /tools/parallel-search docs page, web.md provider listing, docs nav,
labeler entry, per-plugin registration contract test, and minimal core
touch-points (legacy migrate, registration cases, providers contract list,
runtime bundled list, vitest extension paths)
## Summary
- Adds native Google Chat approval cards for exec and plugin approval requests that originate from Google Chat spaces or threads.
- Uses opaque server-side action tokens for Google Chat `cardsV2` button callbacks and updates delivered approval messages after resolution or expiry.
- Preserves the shipped Google Chat typing-message default while keeping approval cards on the channel-local native path.
- Suppresses duplicate manual `/approve ...` follow-up delivery inside `extensions/googlechat/` when the native card path owns the approval prompt.
- Documents Google Chat native approval behavior and the `typingIndicator: "message"` default.
## Linked context
Which issue does this close?
Closes #
Which issues, PRs, or discussions are related?
Related Spec 24.8: Google Chat native approval cards.
Was this requested by a maintainer or owner?
Requested by maintainer in the Codex task thread.
## Real behavior proof (required for external PRs)
- Behavior addressed: Google Chat exec and plugin approvals render as native cards and resolve through Google Chat button clicks. The latest change verifies an exec approval card is not accompanied by a duplicate manual `/approve` instruction bubble.
- Real environment tested: OpenClaw dev profile with a real Google Chat DM to the OpenClaw app, local gateway behind a temporary Cloudflare quick tunnel, and Arc/Computer Use against the signed-in Google Chat session.
- Exact steps or command run after this patch: Rebuilt the gateway runtime, started the dev-profile gateway with the Google Chat webhook routed through the tunnel, sent a fresh exec request from Google Chat, verified only the native approval card appeared, clicked `Allow Once` in Google Chat, and checked the command output reply plus marker file.
- Evidence after fix (screenshot, recording, terminal capture, console output, redacted runtime log, linked artifact, or copied live output): Latest proof used nonce `GCHAT_NODOUBLE_LIVE_20260604070730`, approval id `949bc08c-9e57-47c0-b045-137603782292`, and proof directory `.mem/main/proofs/demo-89502-dev-gchat-exec-approval-no-double-send-channel-race/`. `raw/google-chat-gchat-nodouble-request-card-only-clean.png` shows the fresh user message followed by a single native `Exec Approval Required` card with `Allow Once`/`Deny` and no manual `/approve` follow-up bubble. `raw/google-chat-gchat-nodouble-resolved-clean.png` shows the card edited to `Exec Approval: Allowed once` and the final successful command reply. `raw/gchat-nodouble-live-filtered-log.txt` contains `googlechat approval resolved id=949bc08c-9e57-47c0-b045-137603782292 decision=allow-once`. `raw/marker-file-check.txt` records `/tmp/openclaw-gchat-no-double-GCHAT_NODOUBLE_LIVE_20260604070730` as created.
- Observed result after fix: The approval prompt posted as a native Google Chat card only. No duplicate manual approval-instruction bubble was sent. Clicking `Allow Once` resolved the approval through the gateway and OpenClaw replied with the successful exec output in the same Google Chat DM.
- What was not tested: A persistent production Google Chat app URL; live proof used a temporary Cloudflare tunnel for the local dev callback.
- Proof limitations or environment constraints: Video was not captured for the final resumed manual UI run; still screenshots, gateway/proxy logs, a marker-file artifact, and Showboat verification were captured.
- Before evidence (optional but encouraged): Before the final channel-local suppression path, Google Chat could show both the native approval card and a separate manual `/approve` instruction bubble.
## Tests and validation
Which commands did you run?
- `node scripts/build-all.mjs gatewayWatch`
- `node scripts/run-vitest.mjs extensions/googlechat/src/monitor-webhook.test.ts extensions/googlechat/src/monitor.test.ts extensions/googlechat/src/monitor.reply-delivery.test.ts extensions/googlechat/src/monitor-durable.test.ts extensions/googlechat/src/approval-card-actions.test.ts extensions/googlechat/src/approval-handler.runtime.test.ts extensions/googlechat/src/approval-native.test.ts extensions/googlechat/src/approval-card-click.test.ts extensions/googlechat/src/channel-config.test.ts extensions/googlechat/src/targets.test.ts`
- `git diff --check`
- `pnpm docs:list`
- `uvx showboat --workdir .mem/main/proofs/demo-89502-dev-gchat-exec-approval-no-double-send-channel-race verify .mem/main/proofs/demo-89502-dev-gchat-exec-approval-no-double-send-channel-race/raw/showboat-summary.md`
- Live dev-profile Google Chat proof described above.
What regression coverage was added or updated?
- Added Google Chat native approval capability, runtime delivery, card token, and card-click resolver tests.
- Added in-flight native card send suppression coverage so manual follow-up text is suppressed while native card delivery is pending.
- Added cleanup coverage so manual follow-ups are restored if native card send fails.
- Updated webhook ACK coverage for card-click events and default typing-indicator behavior coverage.
What failed before this fix, if known?
Google Chat could deliver the native approval card and still allow a model/message-tool manual `/approve` follow-up to appear as a second visible bubble.
If no test was added, why not?
Tests were added for the changed runtime and webhook behavior.
## Risk checklist
Did user-visible behavior change? (`Yes/No`)
Yes.
Did config, environment, or migration behavior change? (`Yes/No`)
No migration. The shipped Google Chat `typingIndicator: "message"` default is preserved.
Did security, auth, secrets, network, or tool execution behavior change? (`Yes/No`)
Yes.
What is the highest-risk area?
Approval authorization and callback token handling for native Google Chat card actions.
How is that risk mitigated?
Callbacks carry opaque action tokens only, token bindings check account, space, message, expiry, allowed decision, and in-flight state, and actor authorization reuses the existing Google Chat approver allowlist adapter based on stable `users/<id>` principals.
## Current review state
What is the next action?
Merge after current-head CI for `5923f2af46`.
What is still waiting on author, maintainer, CI, or external proof?
Current-head CI is green for `5923f2af46`; live dev-profile proof is complete.
Which bot or reviewer comments were addressed?
Addressed duplicate approval delivery by keeping the final suppression path inside `extensions/googlechat/`, preserving default typing-message behavior, and proving the current Google Chat surface sends only the native approval card.
Add an opt-in bash/zsh shell snapshot cache for host exec runs, consolidate shell helper ownership into src/agents/shell-utils.ts, document OPENCLAW_EXEC_SHELL_SNAPSHOT, and keep Windows config command execution on the bash resolver. Also removes a redundant Discord gateway close-code type branch that was blocking test type checks.
Warn operators when message_tool_only produces unusually substantive private final text without a delivered source reply. Keeps short/NO_REPLY silence quiet, avoids logging response bodies, and distinguishes unrelated side effects from source-reply delivery.
Summary:
- The branch documents friendly browser tab references across docs, the browser skill, CLI help, and tool schema descriptions, and adds tests for target reference resolution and tab alias behavior.
- PR surface: Source +24, Tests +328, Docs +9. Total +361 across 21 files.
- Reproducibility: yes. for the documentation mismatch by source inspection: current main supports friendly ta ... schema/help surfaces still emphasize raw CDP target ids. Runtime behavior itself is not a new failing path.
Automerge notes:
- PR branch already contained follow-up commit before automerge: refactor(browser): share tab reference CLI help
Validation:
- ClawSweeper review passed for head 118af80b0b.
- Required merge gates passed before the squash merge.
Prepared head SHA: 118af80b0b
Review: https://github.com/openclaw/openclaw/pull/88393#issuecomment-4583558133
Co-authored-by: FMLS <kfliuyang@gmail.com>
Co-authored-by: clawsweeper[bot] <274271284+clawsweeper[bot]@users.noreply.github.com>
Approved-by: hxy91819
Co-authored-by: hxy91819 <8814856+hxy91819@users.noreply.github.com>
Adds Xiaomi MiMo voicedesign TTS support by registering the v2.5 voicedesign model and omitting audio.voice for that model's prompt-driven voice design flow.
Also accepts generic TTS aliases modelId, speakerVoice, and speakerVoiceId for Xiaomi provider config and request overrides.
Fixes exec timeout classification so a process that exits after a missed timeout callback is still reported as timed out, using monotonic deadlines to avoid wall-clock skew.
Verification:
- node scripts/run-vitest.mjs extensions/xiaomi/speech-provider.test.ts
- node scripts/run-vitest.mjs src/process/supervisor/supervisor.test.ts
- node scripts/run-vitest.mjs src/agents/bash-tools.exec-foreground-failures.test.ts
- git diff --check
- autoreview --mode local
- live Xiaomi MiMo voicedesign call returned wav RIFF/WAVE output, 169004 bytes
- GitHub CI success on fb3018ef31: CI 26708919072, CodeQL Critical Quality 26708919082, CodeQL 26708919091, OpenGrep PR Diff 26708919089, Workflow Sanity 26708919083, Dependency Guard 26708918574, Real behavior proof 26708921767
Thanks @GimingRao.
Co-authored-by: Raoyu <2425198313@qq.com>
Co-authored-by: giming <53329020+GimingRao@users.noreply.github.com>
* feat(browser): add optional vision understanding to screenshot tool
* fix(browser): wrap vision output as external content, enforce maxBytes, forward auth profiles
* fix(browser): remove no-op scope/attachments config, drop profile pass-through lacking runtime support
* feat(media-understanding): add profile/preferredProfile to DescribeImageFileWithModelParams and forward to describeImage
* style(browser): add curly braces to satisfy eslint curly rule
* fix(browser): correct tools.browser.enabled help text to match actual behavior
* fix(browser): thread agentDir/workspaceDir from plugin tool context into browser vision
* refactor(browser): move vision config from tools.browser to browser.models
The browser plugin's vision configuration now lives on the top-level
`browser` config namespace (browser.models, browser.visionEnabled,
browser.visionPrompt, etc.) instead of `tools.browser`. This aligns
with the plugin's existing config location and avoids confusion between
tool-level and plugin-level settings.
- Remove tools.browser from ToolsSchema and ToolsConfig
- Add models/vision* fields to BrowserConfig and its zod schema
- Update getBrowserVisionConfig to read from cfg.browser
- Update schema help, labels, and quality test
- Update vision.test.ts to use new config shape
* docs(browser): add screenshot vision configuration section
Document the new browser.models config for automatic screenshot
description via vision models, enabling text-only main models to
reason about web page content.
* fix(browser): remove deliverable media markers from vision result, drop unused import
P1: Vision-success path no longer exposes the raw screenshot as
deliverable media (removes MEDIA: line and details.media.mediaUrl).
This prevents channel delivery from auto-sending sensitive page content
when the intended output is a text description.
P2: Remove unused ToolsMediaUnderstandingSchema import that would fail
noUnusedLocals typecheck.
* fix(browser): add command/args fields to browser models schema
The browser vision model schema uses .strict(), so CLI-type entries
with command/args were rejected by TypeScript. Add these fields to
align with MediaUnderstandingModelSchema.
* chore(browser): remove debug console.log statements
* fix(browser): harden screenshot vision result against MEDIA: directive injection and restore image sanitization on failure fallback
ClawSweeper #84247 review round 2:
P1 (security, high): neutralize line-start MEDIA: directives in vision descriptions
before wrapping with wrapExternalContent. The agent media extractor scans every
browser tool-result text block via splitMediaFromOutput which treats line-start
MEDIA: as a trusted local-media delivery directive, and browser is on the
trusted-media allowlist. Without neutralization, page or vision-provider output
containing 'MEDIA:/tmp/secret.png' could synthesize a channel-deliverable media
artifact from untrusted content. wrapExternalContent itself does not strip
line-start directives. Introduce neutralizeMediaDirectives in vision.ts that
prepends '[neutralized] ' to any line whose trimStart() begins with MEDIA:
(case-insensitive), defanging the parser anchor while keeping the original
text human-readable.
P2 (compatibility): pass resolveRuntimeImageSanitization() to imageResultFromFile
in the vision-failure catch fallback. The non-vision screenshot path already
forwards this option (d5cc0d53b7) so configured agents.defaults.imageMaxDimensionPx
takes effect. Without this fix, any provider timeout/error silently bypasses the
sanitization guard and returns a raw full-resolution screenshot.
Regression coverage:
- vision.test.ts: 6 unit cases for neutralizeMediaDirectives (no-op fast path,
mid-line MEDIA: untouched, line-start defanged, leading-whitespace defanged,
case-insensitive, multiple directives per blob).
- browser-tool.test.ts: 2 integration cases that drive the full screenshot
tool execute path:
- 'neutralizes MEDIA: directives in vision text and does not attach media'
asserts no line matches /^\s*MEDIA:/i in returned text, secret path text
is preserved verbatim, details.media is absent, and imageResultFromFile
is not called on the success path.
- 'preserves screenshot image sanitization on vision failure fallback'
mocks describeImageFileWithModel to reject and asserts the fallback
imageResultFromFile call receives imageSanitization: {maxDimensionPx:1600}
plus the 'browser screenshot vision failed' extraText.
* fix(browser): apply clawsweeper fallback media fix from PR #84247
* refactor: reuse media image understanding for browser screenshots
* refactor: use structured media delivery
* test: update music completion media instruction expectation
* fix: trim buffered reply directive padding
* test: refresh codex prompt snapshots for message media aliases
---------
Co-authored-by: scotthuang <scotthuang@tencent.com>
Co-authored-by: Peter Steinberger <steipete@gmail.com>
