Refactor the subagent completion handoff path into the generic agent steering queue, preserving legacy persisted handoff lease fields by normalizing them into steering lease fields on restore.
Also allowlists the split cron run-log SQLite boundary in the Kysely guardrail after rebasing onto current main.
Refs #88407.
* fix(tasks): reclaim ACP zombie runs blocking gateway restart (#88205)
hasBackingSession treated an ACP task as backed whenever its persisted
session-store entry existed, so a crashed mid-turn ACP run left a
status=running record that survived the crash and wedged gateway
restart/update forever.
Gate ACP backing on in-process live-turn liveness instead of entry
existence, behind the existing authoritative-process flag (generalized
from cron-only) so a standalone maintenance CLI with an empty live-turn
map stays conservative and never reclaims. The liveness signal lives in a
core-internal active-turns registry (mirroring cron active-jobs) so it
stays off the SDK-exported AcpSessionManager surface. It is marked once
before the backend loop and cleared when the task is marked terminal, so
a slow init or backend failover cleanup cannot let the sweep reclaim a
still-live turn.
* fix(tasks): preserve cron operator JSON diagnostic reason
Split the merged runtime_not_authoritative reason back into the existing cron_runtime_not_authoritative (shipped, consumed by openclaw tasks maintenance --json operator scripts) and a new acp_runtime_not_authoritative for the ACP branch. Strengthen the cron non-authoritative test to lock the reason string contract.
* fix(tasks): clear ACP turn liveness on retry failures
---------
Co-authored-by: Peter Steinberger <steipete@gmail.com>
Auto-reply now uses the existing per-model model params thinking value before falling back to the global thinkingDefault, matching gateway/shared model selection behavior.\n\nVerified with targeted auto-reply and agents Vitest coverage plus formatting and diff checks.\n\nThanks @tynamite for the fix.
Preserve OpenAI-compatible replay reasoning when the selected custom or self-hosted model already has reasoning metadata enabled.
The transcript policy now treats existing model metadata as the replay contract instead of requiring a new provider config knob, and the OpenAI-compatible serializer preserves reasoning_content for those routes while keeping stock OpenAI, Gemma 4, and known non-replayable OpenRouter safeguards.
Fixes#88068.
Replaces #88071.
Treat pathless POSIX shell builtins (`:`, `cd`, `false`, `pwd`, `true`) as internally safe only during shell allowlist evaluation. This avoids approval prompts for chains like `cd /tmp && git status` when the executable segment is already allowlisted, without adding a `tools.exec.safeBuiltins` config knob.
Environment-mutating builtins (`export`, `unset`), code-evaluating builtins (`eval`, `source`, `.`), unknown commands, and direct argv execution remain approval-gated unless separately allowlisted.
Proof: `pnpm test src/infra/exec-safe-builtins.test.ts src/agents/bash-tools.exec.security-floor.test.ts -- --reporter=verbose`; `pnpm changed:lanes --json`; `pnpm check:no-conflict-markers`; `git diff --check origin/main...HEAD`. CI related failures were resolved on the final SHA; remaining `checks-node-core-runtime-media-ui` failure is unrelated to this PR.
Fixes#46056.
Thanks @kinjitakabe.
Co-authored-by: kevinkang-ai <273844887+kevinkang-ai@users.noreply.github.com>
Fixes#88443.
Cooldown-only edits under auth.cooldowns now hot reload the active runtime config instead of scheduling a gateway restart. This avoids dropping active gateway work while preserving restart-required behavior for gateway.auth.* credential changes.
Verification:
- pnpm test src/gateway/config-reload.test.ts -- --reporter=verbose
- env -u OPENCLAW_TESTBOX pnpm check:changed
- .agents/skills/autoreview/scripts/autoreview --mode branch --base origin/main --prompt 'Review PR 88474 after rebase. Focus on whether auth.cooldowns hot reload correctly refreshes active runtime config without weakening gateway auth/token restart behavior. Treat no-op vs hot reload semantics as central.'
- GitHub CI, Real behavior proof, CodeQL, Dependency Guard, OpenGrep PR Diff, and Workflow Sanity passed on 51232ff66c.
Thanks @IWhatsskill.
Suppress BOOT.md/internal-runtime-context echoes in fallback boot sends.
Wrap boot prompts as internal runtime context, track the active boot prompt during boot runs, and sanitize message-tool visible payloads before dispatch so fallback models cannot deliver copied BOOT.md instructions or leak them through raw-params errors. Preserves media/presentation sends that still contain non-text payload content after sanitization.
Fixes#53732.
Co-authored-by: stainlu <stainlu@newtype-ai.org>
Allow media understanding providers to opt into synthetic non-secret auth for local or self-hosted no-auth audio/video execution.
This preserves configured env/profile/literal provider credentials first, keeps explicit profile failures hard-fail, and leaves unmarked remote providers fail-closed.
Fixes#74644.
Limit plugin metadata snapshots to the channel, provider, and startup surfaces that need them, while preserving unscoped fallback for incomplete index data and provider runtime resolution.
Refs #70533.
Refs #84628.
Co-authored-by: IWhatsskill <IWhatsskill@users.noreply.github.com>
Fixes#67423.
Resolve provider-entry apiKey fields that intentionally reference model auth profiles through centralized binding logic, so runtime auth and status labeling agree. Preserve env-first precedence, SecretRef handling, provider/baseUrl compatibility checks, and model auth-mode guards.
Verification:
- node scripts/run-vitest.mjs src/agents/model-auth.profiles.test.ts src/agents/model-auth-label.test.ts
- PATH=/tmp/openclaw-corepack-shim.XXXXXX:$PATH CI=true pnpm check:changed
- .agents/skills/autoreview/scripts/autoreview --mode branch --base origin/main
- GitHub CI run 26710260760 and related CodeQL/proof checks on f55dec154d
Co-authored-by: kinjitakabe <273844887+kinjitakabe@users.noreply.github.com>
Stabilizes Claude CLI reusable sessions when Claude token rotation causes transient token-shaped credential reads. Local Claude CLI OAuth and token credential encodings now share the same identity-only auth-epoch, while ref-backed token auth profiles ignore refreshed token material and plaintext token profiles remain epoch-sensitive on manual token replacement.
Fixes#74312.
Proof: focused local Vitest, autoreview, Testbox-through-Crabbox tbx_01ksyrcknbt743x32x6k1s95qw, and GitHub CI run 26709864094 all passed.
Co-authored-by: stainlu <stainlu@newtype-ai.org>
Deliver plugin-owned bound-thread replies even when the source room is configured for `message_tool` visible replies. Normal agent final text still stays private unless the agent calls `message(action=send)`.
Document the distinction in the group/channel docs and root routing policy, and keep ambient room-event plus unauthorized text-slash suppression covered by regression tests.
Fixes#87721.
Align diagnostic stuck-session recovery in-flight dedup with the runtime recovery key. The coordinator now dedups by logical session ref only, so a mid-flight generation bump cannot emit a phantom `session.recovery.requested` event that runtime recovery skips as already in flight.
Adds a regression test for the idle-queued stall path where a queued message bumps generation while recovery is pending.
Fixes#88010
## Summary
- Document scoped configured mention-pattern policy on the Groups page, including allow/deny mode semantics, supported conversation IDs, account-level precedence, and native-mention behavior.
- Add config UI help for `mentionPatterns.mode`, `allowIn`, and `denyIn` on Discord, Matrix, Slack, Telegram, and WhatsApp.
- Regenerate channel config/docs/plugin SDK metadata baselines for the new hint copy.
Refs #70864.
## Verification
- git diff --check
- pnpm format:docs:check
- pnpm docs:check-mdx
- pnpm docs:check-links
- pnpm config:channels:check
- pnpm config:docs:check
- pnpm plugin-sdk:api:check
- node scripts/run-vitest.mjs src/config/schema.hints.test.ts
- .agents/skills/autoreview/scripts/autoreview --mode local
## Real behavior proof
Behavior addressed: Documentation and config UI metadata for scoped configured mention-pattern policy.
Real environment tested: Local OpenClaw checkout on macOS.
Exact steps or command run after this patch: The verification commands listed above.
Evidence after fix: Docs formatting, MDX, link audit, generated config/channel/API baselines, and config hint tests passed; autoreview reported no accepted/actionable findings.
Observed result after fix: The Groups page now explains how to scope `messages.groupChat.mentionPatterns` with `channels.<channel>.mentionPatterns`, and config metadata exposes field help for the supported channels.
What was not tested: Live Discord, Matrix, Slack, Telegram, or WhatsApp inbound messages; this PR is documentation/config metadata only and follows the already-landed runtime behavior from #70864.
* fix(agents): route media task hints below the system-prompt cache boundary
Per-turn image/video/music generation task hints were injected into the
static prependSystemContext slot, landing above the cache boundary inside the
cacheable prefix. The hints are present only on user/manual turns and vary
with active media tasks, so the cacheable prefix shifted turn-to-turn and
defeated Anthropic/OpenAI prompt caching (#85203).
Split the per-turn media hints out of the prepend resolver into
resolveAttemptMediaTaskSystemPromptAddition and route them below the boundary
via the existing prependSystemPromptAddition helper, matching how subagent and
context-engine system-prompt additions are already routed. The static plugin
prependSystemContext / appendSystemContext hook fields are unchanged and
remain in the cacheable prefix. Applied at both consumers (embedded agent
runner and CLI runner).
* fix(agents): keep media task hints below the cache boundary for hook systemPrompt overrides
A before_prompt_build hook that returns a full systemPrompt override replaces
the base prompt with marker-free text. Per-turn media-generation task hints
were then front-prepended into that marker-free prompt, which providers cache
as a single block, so the cached prefix still shifted turn-to-turn on the
override path (#85203).
Wrap the base with ensureSystemPromptCacheBoundary at both media-routing sites
(embedded agent runner and CLI runner) so a marker-free override gets an
appended boundary and the hint routes into the uncached suffix. The helper is
idempotent, so marker-bearing prompts are unchanged. The shared
prependSystemPromptAddition wrapper and the static prependSystemContext /
appendSystemContext hook fields are untouched.
* fix(agents): keep marker-free idle prompts cacheable below the boundary
A marker-free hook systemPrompt override only had the cache boundary
ensured on turns with an active media task. On idle turns the later
appendModelIdentitySystemPrompt landed above the absent boundary, so the
idle cached system prefix diverged from active turns and prompt caching
broke across active/idle transitions. Ensure the boundary regardless of
media state in both the embedded and CLI runners, and extend the
regression to cover the model-identity append across active->idle.
* fix(agents): scope cache-boundary ensure to the model-identity append
Ensuring the boundary unconditionally on media-idle turns appended a
boundary marker to empty raw/gateway system prompts (turning "" into a
marker-only prompt) and to prompts with nothing below the boundary.
Instead ensure the boundary only when a model identity line is actually
appended to a non-empty prompt, in both the embedded and CLI runners.
This still keeps the identity below the boundary for marker-free hook
systemPrompt overrides (the #85203 idle-cache regression) while leaving
empty and identity-less prompts untouched.
* test: refresh stale type and lint expectations
* test: stabilize CI timeout checks
* test: satisfy channel entry lint
* fix(agents): skip cache boundary for blank prompts
* fix(channels): keep draft flush timer referenced
* test(agents): tolerate failed exec timeout setup
---------
Co-authored-by: Peter Steinberger <steipete@gmail.com>
