The schema.help text for `models.providers.*.timeoutSeconds` documents the
key as the user-facing knob for "slow local or self-hosted model servers".
In practice the option is also the only configurable lever for the LLM
idle/first-token watchdog. However `resolveLlmIdleTimeoutMs` was still
running the explicit provider timeout through `clampImplicitTimeoutMs`,
clamping it back down to the implicit ~120s `DEFAULT_LLM_IDLE_TIMEOUT_MS`
ceiling for any non-cron, non-local provider.
Consequence (matches #77744 and #78361):
- User sets `models.providers.llamacpp.timeoutSeconds: 14400` (or 600 for
a slow Gemini/Opus turn with a large tool payload).
- Hot reload accepts the value, runtime resolves
`modelRequestTimeoutMs = 14_400_000`.
- Idle watchdog still trips at ~120s with
"LLM idle timeout (120s): no response from model", aborting an
otherwise-healthy upstream that is mid-prefill or buffering thinking
tokens.
Fix: when the caller passes an explicit `modelRequestTimeoutMs`
(sourced from `models.providers.<id>.timeoutSeconds` /
`model.requestTimeoutMs`), treat it as a deliberate ceiling for cloud
providers too. The run-timeout / agent-timeout bounds still apply via
`timeoutBounds`, so a shorter explicit run timeout always wins. The
implicit default watchdog still kicks in when the user has not set a
provider timeout, preserving the network-silence-as-hang guard for
default configs.
Updated the two corresponding test cases that asserted the old
clamp-on-cloud behavior; all 71 tests in `llm-idle-timeout.test.ts`
and the wider 430-test `src/agents/pi-embedded-runner/run/` lane pass.
Schema help text refreshed to call out that the same knob raises the
idle watchdog ceiling.
Refs: #77744, #78361
Summary:
- The branch adds per-agent `agents.list[].experimental.localModelLean` config and applies lean tool filtering through agent, session, and default-agent resolution.
- Reproducibility: not applicable. this is a feature/config PR rather than a current-main bug report. The chan ... or is supported by source review, focused tests in the branch, and the PR body's redacted live runtime log.
Automerge notes:
- PR branch already contained follow-up commit before automerge: feat(agents): support per-agent local model lean mode
- PR branch already contained follow-up commit before automerge: fix(clawsweeper): address review for automerge-openclaw-openclaw-8407…
Validation:
- ClawSweeper review passed for head 1f9a9554da.
- Required merge gates passed before the squash merge.
Prepared head SHA: 1f9a9554da
Review: https://github.com/openclaw/openclaw/pull/84073#issuecomment-4486397570
Co-authored-by: Bob <dutifulbob@gmail.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: osolmaz
Co-authored-by: osolmaz <2453968+osolmaz@users.noreply.github.com>
Fixes#83388.
- Honor per-agent `tools.codeMode` in config schema, runtime code-mode resolution, and model payload filtering.
- Preserve grouped OpenAI tool declarations when code-mode filtering keeps only `exec` and `wait`.
- Sync generated config/prompt baselines and carry a narrow media CI unblocker from current `main` fallout.
Co-authored-by: Kaspre <kaspre@gmail.com>
Summary:
- The PR narrows embedded PI session transcript write-lock scope, adds stale/max-hold config plumbing, and updates affected transcript, doctor, gateway, SDK, Codex mirroring, docs, and regression-test surfaces.
- Reproducibility: yes. Current main source still holds the embedded session write lock from early attempt set ... cksmith Testbox contention proof on unmodified main; I did not rerun the live repro in this read-only pass.
Automerge notes:
- PR branch already contained follow-up commit before automerge: fix(agents): narrow context engine session lock
- PR branch already contained follow-up commit before automerge: fix session lock runner build types
- PR branch already contained follow-up commit before automerge: Release embedded session write lock before model I/O
- PR branch already contained follow-up commit before automerge: fix(clawsweeper): address review for automerge-openclaw-openclaw-8289…
Validation:
- ClawSweeper review passed for head 4c6dd7ed6e.
- Required merge gates passed before the squash merge.
Prepared head SHA: 4c6dd7ed6e
Review: https://github.com/openclaw/openclaw/pull/82891#issuecomment-4469282923
Co-authored-by: Alex Knight <15041791+amknight@users.noreply.github.com>
Co-authored-by: clawsweeper <274271284+clawsweeper[bot]@users.noreply.github.com>
* Default bootstrap truncation warnings to always
Make bootstrap truncation warnings surface on every affected run by default while preserving explicit off and once configuration.
* Refresh checks after proof formatting fix
* Refresh checks after live proof update
* docs: align bootstrap warning default reference
Update the public agent config reference to match the new default bootstrapPromptTruncationWarning mode and recommended example.
* fix(agents): scope provider SSRF trust by origin
* fix(provider): preserve explicit private-network deny
* docs(provider): document exact-origin SSRF trust
* test(provider): cover exact-origin SSRF edges
* docs(provider): align local model private-origin guidance
* refactor(ssrf): keep policy merging in infra
* test(ssrf): cover exact-origin trust through guard
* test(ssrf): block sibling private-origin redirects
* fix(provider): keep loopback trust origin-scoped
* fix(provider): block metadata origin trust
* fix(ssrf): keep metadata rebinding blocked
* fix(ssrf): block cloud metadata origins
* fix(ssrf): block ipv6 metadata origins
* fix(ssrf): block embedded metadata origins
* test(ssrf): cover embedded link-local metadata
* test(provider): cover custom anthropic proxy classification
* test(provider): widen transport policy mock
* test(plugin-sdk): assert metadata-IP allowedOrigins entries are rejected
Plugin authors can construct an SsrFPolicy that lists any well-formed
http(s) origin in allowedOrigins. The abuse-resistance lives one layer
deeper, in resolvePinnedHostnameWithPolicy's metadata/link-local block.
Add an SDK-level smoke test asserting that contract directly:
- AWS/Alibaba IMDS IPv4 literals, GCP metadata canonical hostname,
IPv6 ULA metadata literal, and non-metadata link-local IPv4 entries
build a policy via ssrfPolicyFromHttpBaseUrlAllowedOrigin and are
then rejected at resolvePinnedHostnameWithPolicy.
- DNS rebinding from a trusted private DNS origin to a metadata IP is
rejected even when the request hostname is origin-trusted.
This would fail if the SDK helper or resolveSsrFPolicyForUrl ever
short-circuited past the metadata block.
* chore(docs): regenerate baselines after upstream rebase
upstream/main moved between rebases; the merged source state for the
PR's `src/config/schema.help.ts` change and the upstream plugin-sdk
surface changes both produce different hashes than the committed
baselines, so `config:docs:check` and `plugin-sdk:api:check` would fail.
Regenerated via `pnpm config:docs:gen` + `pnpm plugin-sdk:api:gen` on
Crabbox; both baselines verified with their respective `--check`
generators.
* test(plugin-sdk): assert SSRF blocked error class
* fix(lint): satisfy exact-origin PR lint rules
* docs: clarify custom provider origin trust
* chore(docs): refresh plugin sdk api baseline
---------
Co-authored-by: Peter Steinberger <steipete@gmail.com>
Summary:\n- Add optional per-agent bootstrap profile overrides for contextInjection, bootstrapMaxChars, and bootstrapTotalMaxChars.\n- Resolve per-agent bootstrap profile settings before agents.defaults and thread the resolved session agent through embedded, compact, CLI, and /context diagnostic paths.\n- Update schema/help/docs/changelog plus focused runtime, schema, and /context regression coverage.\n\nVerification:\n- Local focused auto-reply tests and formatter checks passed.\n- Local pnpm check:changed passed before landing follow-ups.\n- Local Node 24 pnpm check:test-types passed after merging latest main into the PR branch.\n- GitHub PR state CLEAN at 0ff12062840f42daf2666c5fabb127c3f7631669.\n- ClawSweeper re-review completed successfully with no actionable repair finding.\n\nFixes #69966.
* fix(heartbeat): unblock beads cadence — parallel broadcast, agent-scoped busy check, full HEARTBEAT.md prompt, connect-timeout, doctor warning
* docs(changelog): note heartbeat cadence fixes
* fix(heartbeat): address review feedback
* fix(heartbeat): append HEARTBEAT.md directives to commitment-only task dispatch (review feedback)
* docs(changelog): extend heartbeat fix entry — commitment-only task dispatch path (review feedback)
* fix(heartbeat): clear connect timer on synchronous baseFn throw (review feedback)
When the provider stream function passed to streamWithIdleTimeout throws
synchronously during setup, the connect watchdog timer was left armed
and could fire onIdleTimeout later with a stale error, keeping the
process open past the real failure. Wrap the synchronous baseFn(...)
invocation in a try/catch that clears the connect timer before
rethrowing, and add a regression test that asserts onIdleTimeout is
not invoked after the synchronous throw.
* docs(changelog): note round-4 heartbeat fix (review feedback)
Bump the heartbeat fixes list from six to seven and document the
synchronous-baseFn-throw connect-timer cleanup added in the prior
commit.
* fix(heartbeat): honor omitted doctor target (review feedback)
* fix(heartbeat): merge doctor heartbeat defaults (review feedback)
Teach the heartbeat session-target doctor warning to enumerate the same agents as the runtime heartbeat runner and merge agents.defaults.heartbeat with per-agent overrides before checking pinned sessions.
Add regression coverage for default-only heartbeat.session pins and explicit agent heartbeat entries that inherit the default session.
Validation:
- pnpm test src/commands/doctor-heartbeat-session-target.test.ts
- pnpm tsgo:core
- pnpm tsgo:core:test
- pnpm config:schema:check
- pnpm exec oxlint src/commands/doctor-heartbeat-session-target.ts src/commands/doctor-heartbeat-session-target.test.ts
- pnpm exec oxfmt --check src/commands/doctor-heartbeat-session-target.ts src/commands/doctor-heartbeat-session-target.test.ts
- git diff --check
Beads: openclaw-8zp
* test(heartbeat): avoid redundant doctor assertion (review feedback)
The CI lint shard flags the non-null assertion in the heartbeat doctor regression test. Keep the same test setup while using an explicit guard so the test still narrows the fixture before mutating the heartbeat entry.
Validation:
- pnpm exec oxlint src/commands/doctor-heartbeat-session-target.test.ts
- pnpm test src/commands/doctor-heartbeat-session-target.test.ts
- pnpm tsgo:core:test
- git diff --check
Beads: openclaw-8zp
* docs(config): refresh baseline after heartbeat branch update
* fix(heartbeat): narrow doctor session warnings (review feedback)
Adds opt-in `gateway.tailscale.preserveFunnel`. When `tailscale.mode = "serve"`
and an externally configured Tailscale Funnel route already covers the gateway
port, OpenClaw checks `tailscale funnel status --json` before re-applying
`tailscale serve` and skips both Serve and the `resetOnExit` teardown for that
run, preserving operator-managed Funnel exposure across gateway restarts.
The Funnel-status parser handles every documented Tailscale target scheme
(http, https, https+insecure) via an RFC 3986 scheme strip, plus loopback
hostnames (127.0.0.1, localhost, ::1) and bare-port forms. AllowFunnel-disabled
hosts and other-port routes are ignored.
Closes#57241.
Summary:
- The PR adds a `before_agent_run` plugin hook with pass/block decisions, redacted blocked-turn persistence, diagnostics/docs/changelog updates, and focused runner, gateway, session, and plugin tests.
- Reproducibility: not applicable. as a feature PR rather than a current-main bug report. Current main lacks ` ... un`, while the PR head adds source coverage and copied live Gateway/WebChat log proof for the new behavior.
Automerge notes:
- PR branch already contained follow-up commit before automerge: fix: trim before agent hook PR scope
- PR branch already contained follow-up commit before automerge: fix: keep before-agent blocks redacted
- PR branch already contained follow-up commit before automerge: fix: keep runtime context out of model prompt
- PR branch already contained follow-up commit before automerge: docs: refresh config baseline after rebase
- PR branch already contained follow-up commit before automerge: fix: align blocked turn clients with redacted content
- PR branch already contained follow-up commit before automerge: fix: remove out-of-scope client block UI changes
Validation:
- ClawSweeper review passed for head 767e46fde8.
- Required merge gates passed before the squash merge.
Prepared head SHA: 767e46fde8
Review: https://github.com/openclaw/openclaw/pull/75035#issuecomment-4351843275
Co-authored-by: Jesse Merhi <jessejmerhi@gmail.com>
Co-authored-by: jesse-merhi <79823012+jesse-merhi@users.noreply.github.com>
Co-authored-by: clawsweeper <274271284+clawsweeper[bot]@users.noreply.github.com>
- Add PostCompactionLoopPersistedError.fromVerdict factory.
- Add unit tests for the error class + fromVerdict adapter.
- Disabled guard is now truly dormant (no state mutation when enabled=false).
- Tighten help text for postCompactionGuard.enabled.
Refs #77474
Pure module with unit tests; not yet wired into runner. The guard arms
after auto-compaction-retry and aborts when the same (tool, args, result)
triple repeats within the configured window.
Refs #77474
Peter Steinberger