fix(command): retry claude-cli transcript probe to close flush race by benjamin1492 · Pull Request #81048 · openclaw/openclaw

benjamin1492 · 2026-05-12T14:12:05Z

Summary

claudeCliSessionTranscriptHasContent in src/agents/command/attempt-execution.helpers.ts decides whether the runtime can resume a claude-cli session via --resume or has to start cold. The probe does a single-pass scan of the project JSONL for at least one assistant message. The problem is that claude-cli flushes the user-message header to the transcript before the assistant turn lands — there is a sub-100ms window where the JSONL exists on disk but lacks an assistant message. The current single-pass probe races that flush, returns false, the runtime decides "no transcript, cold start," and the prior turn's context is lost. The negative path was also fully silent, so gateway logs gave no signal that a transcript-missing reset was actually a flush race rather than a genuine missing-session.

Two small changes:

Scan with one short retry. The scan now records per-project fileExists and hasAssistant. If the first scan finds the JSONL exists but lacks an assistant message, the probe sleeps 150ms and re-scans once. The sleep is gated on fileExists so the genuinely-no-session path doesn't pay the latency.
Diagnostic warn on the negative path. Both negative outcomes (retry exhausted with file still empty; no matching JSONL in any project dir) emit cliBackendLog.warn with sessionId, homeDir, and the per-project diagnostic. This makes the failure mode visible instead of silent.

Closes

Closes #81042

Test plan

src/agents/command/attempt-execution.test.ts is the established home for these helpers; new tests extend that file:

JSONL with assistant message present from t=0 → returns true immediately, no retry sleep observed. Spies on globalThis.setTimeout and asserts no 150ms delay was scheduled.
JSONL exists but empty at t=0, then assistant message appears during the 150ms window → returns true. Mocks setTimeout(_, 150) so that the test appends the assistant message to the file before firing the continuation, exercising the rescan path.
JSONL exists, never gains an assistant message → returns false, warn fires once with "after 150ms retry" and includes the sessionId.
No matching JSONL in any project dir → returns false, warn fires with "no matching jsonl" and projectCount=<n>.

pnpm test src/agents/command/attempt-execution.test.ts
 Test Files  1 passed (1)
      Tests  38 passed (38)

pnpm tsgo and pnpm check:test-types both pass.

Affected scope

claude-cli backend only, on the resume path. First-turn / cold-start callers never hit this probe.
API backends are unaffected — they don't use a transcript-based reuse check.

Risk

Adds at most one 150ms wait on the negative-but-file-exists path; that path is already a prelude to transcript-missing recovery, so the latency budget is essentially free. The truly-no-session path (no .claude/projects/*/sessionId.jsonl anywhere) keeps its current latency.
cliBackendLog was previously imported only from inside src/agents/cli-runner/. The new import in src/agents/command/attempt-execution.helpers.ts is a one-line cross-module reference into ../cli-runner/log.js — log.ts has no transitive imports from command/ so there's no cycle. (Verified with pnpm tsgo and tests.)
The new diagnostic log lines fire only on the negative path; on healthy resumes there is no log churn.

Real Behavior Proof

Behavior or issue addressed: Closes #81042. claudeCliSessionTranscriptHasContent was returning false in the sub-100ms window where the claude-cli session JSONL existed on disk but had not yet been flushed with an assistant message. The runtime read this as transcript-missing and forced a cold session restart, losing all cross-turn context. The race fired most often immediately after a session-id rotation when the next inbound turn arrived before claude-cli completed its async transcript flush.

Real environment tested: Live OpenClaw 2026.5.7 install on Ubuntu 24.04 (Linux 6.17.0-23-generic x86_64), Node v22.22.0, claude-cli backend (Claude Max subscription), Telegram bot channel. Same scan-with-retry refactor as proposed in this PR was first applied as a runtime patch to the bundled dist/attempt-execution.helpers-CENmB56f.js so the comparison was made against actual runtime behavior, not staged simulations.

Exact steps or command run after this patch:

Apply the scan-with-retry refactor directly to the bundled claudeCliSessionTranscriptHasContent (matches this PR's diff: split scan into a helper returning { hasAssistant, fileExists, ... } per project, retry once after 150ms when fileExists && !hasAssistant, log a cliBackendLog.warn diagnostic on the negative path).
systemctl --user restart openclaw-gateway
Resume normal Telegram chat use with the agent for ~75 minutes spanning ~30 turns including session rotations triggered by config changes and natural idle timeouts.
After the patch window, query the gateway journal for any missing-transcript or restart events:

journalctl --user -u openclaw-gateway --since "2026-05-12 09:11" \
  | grep -E "(cli-backend|claude-live|live session)" \
  | grep -E "reset|close|restart|rotate|reason="

Evidence after fix: Redacted live runtime journal output captured directly from journalctl --user -u openclaw-gateway on the actual install.

Pre-patch window (08:00–09:11 EDT) — three missing-transcript resets in 38 minutes of normal use, each cascading into a session restart. Post-hoc inspection of the JSONL files for these session ids confirmed they did exist on disk; the probe was racing the flush:

2026-05-12T08:13:06.295-04:00 [agent/cli-backend] cli session reset: provider=claude-cli reason=missing-transcript
2026-05-12T08:13:06.327-04:00 [agent/cli-backend] claude live session close: provider=claude-cli model=claude-opus-4-7 reason=restart
2026-05-12T08:45:08.710-04:00 [agent/cli-backend] cli session reset: provider=claude-cli reason=missing-transcript
2026-05-12T08:46:26.340-04:00 [agent/cli-backend] claude live session close: provider=claude-cli model=claude-opus-4-7 reason=restart
2026-05-12T08:51:32.073-04:00 [agent/cli-backend] cli session reset: provider=claude-cli reason=missing-transcript
2026-05-12T08:51:32.099-04:00 [agent/cli-backend] claude live session close: provider=claude-cli model=claude-opus-4-7 reason=restart

User-visible Telegram conversation excerpt at the moment of the 08:45:08 reset, showing the cross-turn amnesia symptom from the user's perspective:

[8:39 AM] Leonard: Watcher is armed — I'll page you as soon as the sentinel appears.
[8:44 AM] User:    still working on it?
[8:45 AM] Leonard: Nothing's actively in flight on my side… What's "it" referring to?

Post-patch window (09:11+ EDT) — full output of the same journal query:

$ journalctl --user -u openclaw-gateway --since "2026-05-12 09:11" \
    | grep -E "(cli-backend|claude-live|live session)" \
    | grep -E "reset|close|restart|rotate|reason="
(no matches)

Live Telegram screenshot from the affected install, captured pre-patch on 2026-05-12. The user asks "still working on it?" 5 minutes after the agent's earlier reply describing in-flight work; the agent denies anything is in flight and asks "What's 'it' referring to?". The user's next message quote-backs the agent's own prior message (with a hand-drawn arrow connecting the two). Two unrelated project names in the parenthetical have been redacted; the bug demonstration is not affected. This is the user-visible symptom of the runtime events shown in the journal log above.

Observed result after fix: Zero missing-transcript resets in the post-patch window. Cross-turn context survives the resume path through transcript flushes that previously raced. The 150ms retry is gated on fileExists, so the no-session path takes the same time as before — verified by spot-checking gateway timing logs across cold-start invocations.

What was not tested: API-backend providers were not tested because they don't use claudeCliSessionTranscriptHasContent. The race was reproduced on Linux only; the same flush behavior in claude-cli should apply on macOS, but it was not exercised here. The companion fingerprint bug #81041 remained patched throughout these runs to keep session lifetimes long enough for the transcript probe to be exercised.

clawsweeper · 2026-05-12T14:15:42Z

Codex review: needs maintainer review before merge. Reviewed May 29, 2026, 1:21 PM ET / 17:21 UTC.

Summary
This PR changes Claude CLI resume validation to probe a deterministic cwd/session JSONL path with a short grace rescan, threads cwd/workspace context into the probe, and captures the last Claude JSONL session_id.

PR surface: Source +40, Tests +415. Total +455 across 11 files.

Reproducibility: yes. The source path is clear from the current transcript probe and the PR discussion includes live missing-transcript logs; I did not rerun the live Claude CLI/Telegram scenario in this read-only review.

Review metrics: none identified.

Merge readiness
Overall: 🐚 platinum hermit
Proof: 🐚 platinum hermit
Patch quality: 🐚 platinum hermit
Result: ready for maintainer review.

Overall follows the weaker of proof and patch quality, so missing proof can cap an otherwise strong patch.

Rank-up moves:

none.

Risk before merge

[P1] The deterministic path is now fail-closed for session reuse; if the resolver or cwd choice disagrees with Claude CLI's actual project path, existing valid bindings could be cleared and users would get cold starts.
[P1] The adjacent write-side session-state PR remains open, so maintainers should choose merge order and resolver assumptions deliberately before landing both changes.

Maintainer options:

Land with session-state risk acknowledged (recommended)
Accept the fail-closed session-state risk once required checks are green because the branch has targeted coverage and live proof for the core read-side failure.
Ask for a latest-head live soak
Before merge, require a short maintainer or contributor log proof on the current head if reviewers want runtime evidence for the final cwd/resolver commits too.
Sequence with the write-side PR
Hold this PR until maintainers decide whether the companion write-side work should land first or be reviewed as one coordinated session-state rollout.

Next step before merge

[P2] No narrow automated repair remains; maintainer review should decide merge timing and session-state rollout with the adjacent write-side PR.

Security
Cleared: No concrete security or supply-chain concern found; the diff changes local transcript probing, parser behavior, tests, and an e2e diagnostic helper without new dependencies, workflow permissions, downloads, or secret handling.

Review details

Best possible solution:

Land this read-side Claude CLI transcript probe fix after maintainer review and green checks, while coordinating with #84234 for the write-side recovery path.

Do we have a high-confidence way to reproduce the issue?

Yes. The source path is clear from the current transcript probe and the PR discussion includes live missing-transcript logs; I did not rerun the live Claude CLI/Telegram scenario in this read-only review.

Is this the best way to solve the issue?

Yes. The deterministic cwd-based transcript path plus last-wins session id capture is a narrow read-side repair; the separate write-side binding/orphan handling should remain coordinated in the companion PR.

AGENTS.md: found and applied where relevant.

Codex review notes: model gpt-5.5, reasoning high; reviewed against 5620229f9f92.

Label changes

Label changes:

add P1: The PR targets user-visible Claude CLI context loss in active chat sessions, which is a broken agent workflow affecting real users.
add merge-risk: 🚨 session-state: The diff changes how saved Claude CLI session ids are validated and when bindings are cleared, so a resolver mismatch could drop native session continuity.
add proof: sufficient: Contributor real behavior proof is sufficient. The PR body and comments include live runtime logs, a Telegram-visible symptom excerpt, and later v5 journalctl output showing zero missing-transcript resets after the updated patch soak.
add rating: 🐚 platinum hermit: Overall readiness is 🐚 platinum hermit; proof is 🐚 platinum hermit and patch quality is 🐚 platinum hermit.
add status: 👀 ready for maintainer look: ClawSweeper has no concrete contributor-facing blocker left for this PR. Sufficient (logs): The PR body and comments include live runtime logs, a Telegram-visible symptom excerpt, and later v5 journalctl output showing zero missing-transcript resets after the updated patch soak.

Label justifications:

P1: The PR targets user-visible Claude CLI context loss in active chat sessions, which is a broken agent workflow affecting real users.
merge-risk: 🚨 session-state: The diff changes how saved Claude CLI session ids are validated and when bindings are cleared, so a resolver mismatch could drop native session continuity.
rating: 🐚 platinum hermit: Overall readiness is 🐚 platinum hermit; proof is 🐚 platinum hermit and patch quality is 🐚 platinum hermit.
status: 👀 ready for maintainer look: ClawSweeper has no concrete contributor-facing blocker left for this PR. Sufficient (logs): The PR body and comments include live runtime logs, a Telegram-visible symptom excerpt, and later v5 journalctl output showing zero missing-transcript resets after the updated patch soak.
proof: sufficient: Contributor real behavior proof is sufficient. The PR body and comments include live runtime logs, a Telegram-visible symptom excerpt, and later v5 journalctl output showing zero missing-transcript resets after the updated patch soak.

Evidence reviewed

PR surface:

Source +40, Tests +415. Total +455 across 11 files.

View PR surface stats

Area	Files	Added	Removed	Net
Source	6	112	72	+40
Tests	5	429	14	+415
Docs	0	0	0	0
Config	0	0	0	0
Generated	0	0	0	0
Other	0	0	0	0
Total	11	541	86	+455

Acceptance criteria:

[P1] node scripts/run-vitest.mjs src/agents/command/attempt-execution.test.ts src/agents/command/attempt-execution.cli.test.ts src/agents/cli-runner/prepare.test.ts src/agents/cli-output.test.ts.
[P1] pnpm tsgo:core.
[P1] pnpm check:test-types.

What I checked:

Repository policy applied: Read the full root AGENTS.md and the scoped src/agents/AGENTS.md; session-state compatibility, proof requirements, and agent-test guidance shaped the review. (AGENTS.md:1, 5620229f9f92)
Current main behavior: Current main uses a single scan over Claude project dirs for the saved session id and returns false when no assistant message is found, which matches the reported false missing-transcript surface. (src/agents/command/attempt-execution.helpers.ts:80, 3654ea32a972)
Deterministic probe on PR head: The PR head derives the expected Claude JSONL path from session id plus workspace/cwd, scans it, waits 250 ms on a miss, rescans, and logs a structured v4 miss before failing closed. (src/agents/command/attempt-execution.helpers.ts:85, e3df84ab09b6)
cwd threading on PR head: The command path now passes the CLI process cwd into the transcript probe, and the prepare path passes its resolved cwd, avoiding the prior workspace-only mismatch. (src/agents/command/attempt-execution.ts:508, e3df84ab09b6)
Session id parser fix on PR head: parseCliJsonl now keeps the last backend session id, matching the PR discussion's v5 root cause around ephemeral Claude session ids preceding the canonical id. (src/agents/cli-output.ts:753, e3df84ab09b6)
Regression coverage: The branch adds tests for the grace-window rescan, missing/header-only warnings, canonical resolver use, cwd threading, and last-wins Claude session id parsing. (src/agents/command/attempt-execution.test.ts:573, e3df84ab09b6)

Likely related people:

steipete: The available current-main blame in this shallow checkout attributes the existing transcript probe and parseCliJsonl implementation to commit 3654ea3; deeper history is not available locally. (role: recent area contributor; confidence: medium; commits: 3654ea32a972; files: src/agents/command/attempt-execution.helpers.ts, src/agents/cli-output.ts)
obviyus: The latest force-pushed commits on this PR share the Claude project-dir resolver and thread the CLI cwd into the probe, directly addressing the prior review concern. (role: recent branch maintainer; confidence: medium; commits: 7409717ac30e, e3df84ab09b6; files: src/agents/command/claude-cli-project-dir.ts, src/agents/command/attempt-execution.ts, src/commands/doctor-claude-cli.ts)

What the crustacean ranks mean

🦀 challenger crab: rare, exceptional readiness with strong proof, clean implementation, and convincing validation.
🦞 diamond lobster: very strong readiness with only minor maintainer review expected.
🐚 platinum hermit: good normal PR, likely mergeable with ordinary maintainer review.
🦐 gold shrimp: useful signal, but proof or patch confidence is still limited.
🦪 silver shellfish: thin signal; proof, validation, or implementation needs work.
🧂 unranked krab: not merge-ready because proof is missing/unusable or there are serious correctness or safety concerns.
🌊 off-meta tidepool: rating does not apply to this item.

Shiny media proof means a screenshot, video, or linked artifact directly shows the changed behavior. Runtime, network, CSP, and security claims still need visible diagnostics.

How this review workflow works

ClawSweeper keeps one durable marker-backed review comment per issue or PR.
Re-runs edit this comment so the latest verdict, findings, and automation markers stay together instead of adding duplicate bot comments.
A fresh review can be triggered by eligible @clawsweeper re-review comments, exact-item GitHub events, scheduled/background review runs, or manual workflow dispatch.
PR/issue authors and users with repository write access can comment @clawsweeper re-review or @clawsweeper re-run on an open PR or issue to request a fresh review only.
Maintainers can also comment @clawsweeper review to request a fresh review only.
Fresh-review commands do not start repair, autofix, rebase, CI repair, or automerge.
Maintainer-only repair and merge flows require explicit commands such as @clawsweeper autofix, @clawsweeper automerge, @clawsweeper fix ci, or @clawsweeper address review.
Maintainers can comment @clawsweeper explain to ask for more context, or @clawsweeper stop to stop active automation.

benjamin1492 · 2026-05-13T16:14:18Z

Update: probe widened to 5 attempts / 3250ms after live evidence v1 + v2 were both insufficient

The single-retry shape from the PR body (one 150ms wait, gated on fileExists) closed the within-file flush race in synthetic tests but kept missing the second window in the field — the file-creation race, where the JSONL itself has not yet been created when the runtime probes after a session-id rotation.

Evolution after PR submission

v2 (intermediate, not pushed to this branch). Three attempts at 0/200/400ms (600ms total wait), retry on both branches (file-missing AND file-exists-no-asst), and a branch-specific warn message so the diagnostic tells us which window we lost. Applied as a runtime patch only; never landed upstream because field telemetry from the next day showed even 600ms was short.
v3 (this push, 15b6626de0). Five attempts at 0/250/500/1000/1500ms (3250ms total worst-case wait), same dual-branch retry shape. Sized to the slowest file-creation lag we observed in production.

Field evidence driving the change (2026-05-13 EDT)

Three live transcript probe negative after 600ms (no matching jsonl) events in a single morning, all on the file-creation branch — meaning v2's 600ms window expired before claude-cli wrote the JSONL at all:

Time (EDT)	Session ID prefix	Outcome
08:09	(logged)	`cli session reset reason=missing-transcript` → cold start
11:37	`d6b7c900…3ef136`	`cli session reset reason=missing-transcript` → cold start
11:54:41	`f1b5e368…29fb39`	User-visible amnesia: a multi-screen pfBlockerNG fix delivered at 11:54 was not in context when the user followed up at 11:56 in the same Telegram session

All three hit the no-matching-jsonl branch, not the within-file branch. v1 didn't retry at all on that branch (it was gated on fileExists), and v2's 0/200/400ms back-off was still too short to catch the lag. v3 mirrors the dual-branch retry from v2 but extends the ladder so a slow flush has room to land before the runtime declares transcript-missing.

What changed in this commit

src/agents/command/attempt-execution.helpers.ts:

CLAUDE_CLI_TRANSCRIPT_FLUSH_RETRY_MS = 150 → CLAUDE_CLI_TRANSCRIPT_FLUSH_DELAYS_MS = [0, 250, 500, 1000, 1500] (with a derived _TOTAL_MS = 3250).
The claudeCliSessionTranscriptHasContent body becomes a back-off loop. The fast path is unchanged: if the first scan finds an assistant message, it returns immediately without touching setTimeout.
The if (first.fileExists) gate is gone — the loop retries uniformly on both branches, because production data shows the file-creation window is the one we were losing.
Two warn variants on exhaustion: after 3250ms (file exists, no asst) (carries projects=[…]) and after 3250ms (no matching jsonl) (carries projectCount=N). Operators can tell from one log line which window blew the budget — important if a future field event needs us to widen the ladder again.

src/agents/command/attempt-execution.test.ts:

Existing fast-path test extended to assert no back-off sleep fires when the assistant message is already present.
The previous "lands during 150ms window" test is replaced by two: one for the within-file branch (assistant message lands during the 500ms slot) and one for the file-creation branch (the JSONL itself materializes during the 1000ms slot). Both verify the loop traversed the right number of back-offs (backOffsObserved).
The two negative-path warn tests now mock all four back-off delays ([250, 500, 1000, 1500]) and assert the v3 messages (after 3250ms (file exists, no asst) and after 3250ms (no matching jsonl) respectively).

39/39 in attempt-execution.test.ts pass; pnpm tsgo:core and pnpm check:test-types both clean.

Risk note

Worst-case probe latency on the genuinely-no-session path is now 3250ms instead of 0ms. That path always feeds into transcript-missing recovery anyway, so the latency budget is still essentially free vs. losing context — but reviewers should weigh whether the "negative path is a prelude to recovery" assumption from the original PR still holds at this magnitude. If a tighter cap is preferred, a 3-step [0, 500, 1500] (2000ms) ladder still beats v2 by ~3.3× and would have caught at least the 11:37 and 11:54 events from today's log if the timing distribution we've sampled is representative.

PR is left at its current status; this is a follow-up commit only.

benjamin1492 · 2026-05-13T17:01:11Z

v3 still raced.

Live amnesia event on this branch's running gateway at 12:37:58 EDT today
(2026-05-13). The 3250 ms back-off ladder ran end-to-end and never saw the
JSONL appear — same no matching jsonl branch v2 was hitting at 600 ms,
just at a longer timeout:

2026-05-13T16:37:58.700Z  WARN  claude-cli transcript probe negative after 3250ms (no matching jsonl)
                                  sessionId=da349eb3-5357-4657-bbbf-2a5582ec8c3a projectCount=6
2026-05-13T16:37:58.708Z  INFO  cli session reset: provider=claude-cli reason=missing-transcript
2026-05-13T16:37:58.830Z  INFO  cli exec: ... reuse=invalidated:missing-transcript

That's a clean repro of the failure mode v3's narrative warned about, and
the n=1 datapoint after one full ladder budget. Continuing to widen the
ladder isn't viable: claude-cli decides on its own clock when to flush the
JSONL, the orchestrator probe is open-loop relative to that clock, and the
race is unbounded — picking a 6 s ladder just moves the next field failure
to whatever the next outlier flush takes. The probe needs a deterministic
reference, not a longer back-off.

v4: orchestrator owns the path

The orchestrator already mints --session-id <uuid> for new Claude CLI
sessions (claude-live-session.ts:154), and claude is spawned with
cwd: workspaceDir (claude-live-session.ts:701, execute.ts:553). Those
two facts together fully determine the on-disk JSONL location:

<homeDir>/.claude/projects/<encoded(workspaceDir)>/<sessionId>.jsonl

The encoding rule was live-verified against claude-cli 2.1.140: every non-
alphanumeric character in the absolute cwd is rewritten to - (so
/home/faris/.openclaw/workspace → -home-faris--openclaw-workspace,
/tmp/foo_bar.baz → -tmp-foo-bar-baz). v4 lifts that rule into
claudeCliSessionTranscriptPath() and replaces the previous scan-every-
project-dir loop with a single scanJsonlFile(expectedPath). Because the
path is now derived from data the orchestrator already controls, the only
remaining race window is local FS flush latency on the prior turn's exit;
a 250 ms one-shot grace is plenty for that and any miss past it is a real
signal worth surfacing.

On miss, v4 emits a structured claude-cli transcript probe v4 miss (sessionId-deterministic path, grace 250ms): sessionId=… expectedPath=… fileExists=… warn — distinct prefix from v1/v2/v3 so log readers can see
at a glance which strategy is live and which window failed. Downstream
behavior is unchanged: a miss still produces cli session reset reason=missing-transcript and the turn resumes fresh, but the warn line
ahead of it now points at the exact file that should have been there.

Why this is the last call site change

If v4 ever misses, it isn't "we picked the wrong filename" or "we probed
too early" — both of those are pinned. A v4 miss means either (a) the prior
turn's claude-cli legitimately failed to flush the JSONL (real crash / FS
failure) or (b) upstream Claude Code changed the cwd-encoding rule. Those
need investigating individually instead of being masked by a wider ladder,
which is what "fail loud" buys us.

Tests

src/agents/command/attempt-execution.test.ts adds a
claudeCliSessionTranscriptPath describe block (null on malformed sessionId
/ empty workspace, path encoding for both the openclaw workspace and an
underscore/dot synthetic case) and rewrites the
claudeCliSessionTranscriptHasContent block end-to-end:

happy path: assistant already flushed → no grace sleep fires
FS-latency: first scan miss, JSONL lands during the 250 ms grace, second
scan succeeds (asserts exactly one grace-window sleep)
miss with fileExists=false (JSONL never appeared) — asserts the v4 warn
prefix, sessionId, grace, encoded workspace dir
miss with fileExists=true (header-only JSONL stayed headerless) —
asserts fileExists=true lands in the warn payload
malformed sessionId / undefined workspaceDir still resolve to false

prepare.test.ts and attempt-execution.cli.test.ts were updated for the
new workspaceDir argument and the deterministic path the pre-seeded
fixtures now write to. Local: pnpm test src/agents/command/ src/agents/cli-runner/ src/agents/cli-session.test.ts → 167/167 passed,
pnpm tsgo:core && pnpm check:test-types clean.

Re-review progress:

State: Complete
Detail: The targeted re-review finished, the durable review comment was updated, and the synced verdict was routed.
Run: https://github.com/openclaw/clawsweeper/actions/runs/25888263468
Updated: 2026-05-14T22:13:09.528Z

benjamin1492 · 2026-05-14T22:03:02Z

v5 — `parseCliJsonl` last-wins (commit `e628d42`)

Pushed a follow-up that closes the last context-loss gap we've been able to reproduce locally. Even with v4's deterministic-path probe, we observed cli session reset reason=missing-transcript on --resume turns where claude-cli rotated its session_id mid-stream.

Root cause. parseCliJsonl was first-wins on session_id:

if (!sessionId) {
  sessionId = pickCliSessionId(parsed, backend);
}

parseCliJson (line 294) and createCliJsonlStreamingParser (line 395) had already moved to last-wins (pickCliSessionId(parsed, backend) ?? sessionId), so parseCliJsonl was the lone first-wins capture path. claude-cli 2.1.141 emits ephemeral session_ids from SessionStart hooks before the canonical resumed session_id surfaces in the init event and all subsequent records (including the terminal result). First-wins bound cliSessionBinding.sessionId to the ephemeral id, whose JSONL never lands on the path the v4 probe checks → probe miss → fresh session → continuous context loss.

Fix. Flip parseCliJsonl to last-wins, matching the two sibling parsers. Regression test feeds [ephemeral init, canonical init, result] and asserts the captured sessionId is canonical.

Local stability. Patches 1–5 (systemPromptHash removal, v4 deterministic probe, v4-hotfix workspaceDir threading, v5 last-wins) have been live on my Telegram-facing gateway since 12:25:53 EDT 2026-05-14 (~5h soak at the time of this comment). Zero missing-transcript or transcript probe v4 miss events in the log post-restart; main-session model prefix is steady at claude-cli/claude-opus-4-7 across consecutive turns. Two of my downstream users running 2026.5.7 (one of whom has been reporting amnesia symptoms) are next in line for this rev.

Also dropped a redundant as string cast in attempt-execution.test.ts:641 (oxlint warning surfaced after rebase).

Pinging maintainers — this is the third revision of the PR and the gap-finding has converged. Happy to split the v5 commit into its own PR if that's preferred for review surface.

benjamin1492 · 2026-05-15T12:05:57Z

Overnight soak update — v5 holding clean

Quick follow-up with fresh data on top of yesterday's comment.

Soak window: v5 deployed at 12:25:53 EDT 2026-05-14, gateway restarted at 01:11:32 EDT 2026-05-15 (unrelated overnight cycle, patches re-applied automatically), now at 08:05 EDT 2026-05-15 — ~20h total, ~7h on the post-restart leg.

Result: zero cli session reset reason=missing-transcript events, zero transcript probe v[0-9]+ miss events across the entire window. The last reset of any kind was the pre-v5 12:04 EDT 2026-05-14 event that motivated this revision.

$ journalctl --user -u openclaw-gateway --since "2026-05-14 12:25:53" \
    | grep -cE "cli session reset.*missing-transcript"
0

Main session has stayed bound to a single sessionId across heartbeats, scheduled compaction cycles, and the 04:00 daily rotation. Context continuity from the user's perspective: intact.

Lesson learned on this PR

In hindsight I shipped the initial PR (commit ebbb2627, the retry-with-back-off probe) too early — it addressed the surface symptom but missed two underlying issues that needed four follow-up commits to fully close:

The probe was checking the wrong path family because workspaceDir wasn't threaded through every call site (v4-hotfix, commit dc429bef precursor work).
The deterministic-path probe alone wasn't enough because parseCliJsonl was first-wins on session_id and was binding to ephemeral SessionStart-hook ids before the canonical resumed id surfaced (v5, commit e628d42a).

Going forward I'll hold a similar PR until I have at least 24h of clean soak before opening, rather than treating "passes locally + reproducer no longer fires" as ready. Apologies for the review-cycle churn.

Happy to split v5 into its own PR for a cleaner review surface if maintainers prefer — or land as-is.

…s (3250ms)

…egy (v4)

… (v5) parseCliJsonl was first-wins: it captured the first session_id encountered and never updated. claude-cli emits ephemeral session_ids from SessionStart hooks before the canonical resumed session_id surfaces in the init event and the terminal result event. The orchestrator therefore bound to an ephemeral id whose JSONL never lands on the deterministic path probed by v4, triggering cli session reset reason=missing-transcript on every turn that observed a rotation. Flip parseCliJsonl to last-wins to match the existing parseCliJson and createCliJsonlStreamingParser semantics. Adds a regression test that feeds an ephemeral id followed by a canonical id and asserts the canonical id is captured.

oxlint flagged the `as string` cast on warnSpy.mock.calls[0]?.[0] as unnecessary — the value is already typed as string for the assertion path.

obviyus · 2026-05-29T17:26:50Z

Landed via squash onto main.

Scoped tests: git diff --check; pnpm exec oxfmt --check --threads=1 src/agents/command/claude-cli-project-dir.ts src/commands/doctor-claude-cli.ts src/commands/doctor-claude-cli.test.ts src/agents/command/attempt-execution.helpers.ts src/agents/command/attempt-execution.test.ts src/agents/command/attempt-execution.cli.test.ts src/agents/command/attempt-execution.ts src/agents/cli-output.ts src/agents/cli-output.test.ts src/agents/cli-runner/prepare.ts src/agents/cli-runner/prepare.test.ts; node scripts/run-oxlint.mjs src/agents/command/claude-cli-project-dir.ts src/commands/doctor-claude-cli.ts src/commands/doctor-claude-cli.test.ts src/agents/command/attempt-execution.helpers.ts src/agents/command/attempt-execution.test.ts src/agents/command/attempt-execution.cli.test.ts src/agents/command/attempt-execution.ts src/agents/cli-output.ts src/agents/cli-output.test.ts src/agents/cli-runner/prepare.ts src/agents/cli-runner/prepare.test.ts; CI=1 pnpm tsgo:core && CI=1 pnpm tsgo:core:test; CI=1 OPENCLAW_VITEST_MAX_WORKERS=1 node scripts/run-vitest.mjs run src/agents/cli-output.test.ts src/agents/cli-runner/prepare.test.ts src/agents/command/attempt-execution.test.ts src/agents/command/attempt-execution.cli.test.ts src/commands/doctor-claude-cli.test.ts; CI=1 OPENCLAW_VITEST_MAX_WORKERS=1 node scripts/run-vitest.mjs run src/agents/embedded-agent-runner/run/attempt.cwd-split.test.ts
Autoreview: .agents/skills/autoreview/scripts/autoreview --mode branch --base origin/main clean after fixes.
Changelog: not edited; release-note context is in the PR body and squash commit message per repo policy.
CI notes: check-additional-boundaries-bcd failed on pre-existing extensions/minimax/oauth.ts raw fetch() hits already present on origin/main; checks-node-core-fast failed attempt.cwd-split.test.ts once in CI but passed locally on this exact head.
Land commit: bf260ae
Merge commit: de45530

Thanks @benjamin1492!

@giodl73-repo

* fix(exec): bind node auto-review commands Co-authored-by: Vincent Koc <25068+vincentkoc@users.noreply.github.com> Co-authored-by: jesse-merhi <79823012+jesse-merhi@users.noreply.github.com> * fix(exec): honor node runtime policy for auto-review Co-authored-by: Vincent Koc <25068+vincentkoc@users.noreply.github.com> Co-authored-by: jesse-merhi <79823012+jesse-merhi@users.noreply.github.com> * fix(exec): harden auto-review prompt boundaries Co-authored-by: Vincent Koc <25068+vincentkoc@users.noreply.github.com> Co-authored-by: jesse-merhi <79823012+jesse-merhi@users.noreply.github.com> * fix(exec): align release validation surfaces Co-authored-by: Vincent Koc <25068+vincentkoc@users.noreply.github.com> Co-authored-by: jesse-merhi <79823012+jesse-merhi@users.noreply.github.com> * fix(exec): align release validation checks Co-authored-by: Vincent Koc <25068+vincentkoc@users.noreply.github.com> Co-authored-by: jesse-merhi <79823012+jesse-merhi@users.noreply.github.com> * test(e2e): repair release docker smoke fixtures Co-authored-by: Vincent Koc <25068+vincentkoc@users.noreply.github.com> Co-authored-by: jesse-merhi <79823012+jesse-merhi@users.noreply.github.com> * fix(exec): resolve auto approvals as runtime Co-authored-by: Vincent Koc <25068+vincentkoc@users.noreply.github.com> Co-authored-by: jesse-merhi <79823012+jesse-merhi@users.noreply.github.com> * ci: relax native OpenAI live proof timing Co-authored-by: Vincent Koc <25068+vincentkoc@users.noreply.github.com> Co-authored-by: jesse-merhi <79823012+jesse-merhi@users.noreply.github.com> * fix(exec): include mode in doctor policy warnings Co-authored-by: Vincent Koc <25068+vincentkoc@users.noreply.github.com> Co-authored-by: jesse-merhi <79823012+jesse-merhi@users.noreply.github.com> * test(release): repair live matrix expectations Co-authored-by: Vincent Koc <25068+vincentkoc@users.noreply.github.com> Co-authored-by: jesse-merhi <79823012+jesse-merhi@users.noreply.github.com> * fix(tts): centralize directive number parsing * fix(provider): bound Vydra and Comfy media downloads * fix(discord): validate error code integers * fix(discord): reject unsafe rate limit headers * ci(release): make plugin publish retries idempotent * perf(agent): lazy load embedded agent cli path * fix(whatsapp): validate inbound timestamps * refactor: share agent harness loader helpers * fix(agents): cap unsafe retry-after delays * perf(agent): defer session resolver for scoped gateway turns * fix(msteams): ignore unsafe retry-after delays * refactor: share store writer queue * fix(slack): reject unsafe inbound timestamps * fix(discord): reject unsafe retry-after delays * fix(qa-matrix): cap fault proxy bodies * fix(discord): bound delivery retry delays * refactor: share cron state parsing * Delete changelog directory * fix(zalouser): reject unsafe inbound timestamps * fix(cli): avoid underscored gateway test export * fix(scripts): cap clawtributor avatar probes * fix(telegram): centralize safe thread id parsing * fix(googlechat): drop invalid inbound timestamps * fix(doctor): label auth health by agent (openclaw#85924) Merged via squash. Prepared head SHA: 8c179fc Co-authored-by: giodl73-repo <235387111+giodl73-repo@users.noreply.github.com> Co-authored-by: giodl73-repo <235387111+giodl73-repo@users.noreply.github.com> Reviewed-by: @giodl73-repo * fix(qqbot): validate token expiry lifetimes * fix(openai): validate codex oauth token lifetimes * refactor: share node pairing surface helpers * fix(anthropic): validate oauth token lifetimes * fix(scripts): cap memory FD repro RPC bodies * fix(github-copilot): validate device code lifetimes * fix(msteams): validate oauth token lifetimes * refactor: share cli help argv scan * fix(github-copilot): validate oauth expiry values * fix(scripts): cap realtime smoke responses * fix(chutes): validate oauth token lifetimes * fix(auto-reply): reuse cli sessions for room events * fix(auto-reply): keep room event cli sessions transient * fix(agent-core): reject invalid session timestamps * fix(scripts): cap Claude usage response reads * refactor: centralize skills subsystem * refactor: move skill lifecycle code into skills subsystem * fix: bound skill index cache invalidation * fix: preserve skill snapshot freshness * fix: preserve preloaded skill snapshot entries * refactor: move session skill loader into skills subsystem * fix: preserve empty skill filter short circuit * fix: align empty default skill filter behavior * fix: align skills branch with upstream tar verbose test * fix: drop stale system prompt override imports * refactor: centralize skills runtime paths * refactor: remove stale agents skills barrel * refactor: use direct skills imports * refactor: organize skills subsystem layout * fix: lint centralized skills subsystem * refactor: split skills index follow-up * refactor: centralize skills subsystem * fix: unblock skills centralization checks * fix: route moved skills tests through unit-fast * refactor: centralize skills runtime tests * refactor: share web secret target selection * refactor: centralize safe expiry parsing * fix(exec): normalize unsafe timeout values * fix: persist Copilot SDK session bindings Persist GitHub Copilot SDK session ids in the plugin-state SQLite store so separate OpenClaw process turns can resume the same Copilot-side session when the compatibility fingerprint still matches. The fingerprint covers provider/model/cwd, resolved agent id, resolved Copilot home, and auth identity. Plugin-state lookup/register/delete failures are non-fatal, stale rows are invalidated, and reset delete failures use an in-process tombstone so reset does not accidentally reuse a durable binding. Also routes the QQBot token POST through the plugin SDK SSRF guard with capture disabled for the secret-bearing request, preserving the current token lifetime validation from main. Verification: focused Copilot and QQBot Vitest suites, raw channel fetch guard, autoreview clean, Blacksmith Testbox pnpm check:changed tbx_01kst9fwjmsfzwaxqatszcbf40, live local Copilot two-turn smoke with the same SDK session id persisted in SQLite. Refs openclaw#88064 * fix(exec): cap node run timeouts * perf(agent): skip plugin validation for gateway dispatch * fix(scripts): cap firecrawl compare HTML reads * fix(xai): normalize unsafe oauth lifetimes * refactor: share e2e text file helpers * fix(google): normalize unsafe oauth expiry * fix(openai): normalize codex device lifetimes * refactor: reuse e2e text tail helper * test(xai): type device-code note mock * fix(minimax): reject unsafe oauth expiry * fix(ci): cap dependency guard error bodies * fix(google-meet): normalize oauth expiry * fix(command): stabilize claude-cli transcript resume (openclaw#81048) Fix claude-cli transcript resume so session-id rotation and transcript flush timing do not drop valid resume state. - Capture the latest claude-cli session_id from JSONL output. - Resolve Claude project transcript paths through the shared canonical project-dir resolver. - Probe transcript content from the actual CLI process cwd. - Thanks @benjamin1492! * refactor: share codex e2e install helpers * fix(feishu): bound streaming token expiry * fix(openshell): cap command timeout config * refactor: centralize timer-safe timeout bounds * refactor: share e2e websocket open helper * fix(minimax): guard oauth token fetches (openclaw#88088) * fix(feishu): normalize app registration poll timers * fix(google): reject unsafe vertex adc lifetimes * fix(scripts): cap npm packument reads * fix(auth): reject unsafe wham reset windows * refactor: share qa report arg parsing * fix(retry): cap unsafe retry delays * fix(sandbox): bound novnc observer token ttl * feat(workboard): add agent coordination tools Summary: - Add Workboard agent coordination tools for list/read/claim/heartbeat/release/comment/proof/unblock flows. - Store artifacts, claims, diagnostics, and notifications in the Workboard SQLite-backed plugin state; surface the new metadata through Gateway, Control UI, docs, and plugin manifest contracts. - Add scoped claim authorization, token redaction, stale diagnostic cleanup, atomic proof artifact writes, and generated i18n metadata. Verification: - pnpm test ui/src/i18n/test/translate.test.ts extensions/browser/src/cli/browser-cli-actions-input/register.element.test.ts extensions/workboard/src/store.test.ts extensions/workboard/src/gateway.test.ts extensions/workboard/src/tools.test.ts ui/src/ui/controllers/workboard.test.ts ui/src/ui/views/workboard.test.ts - pnpm ui:i18n:check - env -u OPENCLAW_TESTBOX pnpm check:changed - autoreview --mode local: clean - PR CI passed; Windows checkout failure rerun passed on attempt 2 * perf(gateway): reuse session maintenance config during turns * fix(node-host): cap timeout wrapper delays * fix(talk): cap fast context timeout delay * fix(e2e): harden kitchen sink probe body caps * refactor: share bounded response reader * fix(providers): cap model request timeout delays * fix(oauth): cap request abort timeout delays * test: speed up slow assertions * test: stabilize slow assertion timings * test: shard channel import guardrails * perf(sessions): patch single-entry store writes * refactor: share script bounded response helper * fix(codex): cap responses request timeout delays * fix(scripts): cap gh-read json bodies * fix(lmstudio): cap model fetch timeout delays * feat(ios): default to hosted push relay (openclaw#88096) Merged via squash. Prepared head SHA: 75f939a Co-authored-by: ngutman <1540134+ngutman@users.noreply.github.com> Co-authored-by: ngutman <1540134+ngutman@users.noreply.github.com> Reviewed-by: @ngutman * fix(minimax): cap tts timeout delays * build(plugins): externalize copilot runtime * refactor: share codex app server start context * test(file-transfer): remove stale tar fixture awaits * fix(runtime): centralize safe timer timeout resolution * refactor: share ui chat send wrapper * docs(plugins): clarify external plugin installs * fix: close native hook relay replacement race * fix(qa-lab): cap credential broker request timeouts * refactor: share e2e incremental line reader * test(ci): fix main test expectations (openclaw#88122) * fix(copilot): cap oauth request timeouts * fix(oauth): cap tls preflight timeout * build(plugins): externalize tokenjuice * docs(plugins): add external package readmes * perf: reuse gateway session and plugin metadata paths * fix(exec): bind node auto-review to prepared plans * fix(auth): cap GitHub Copilot OAuth timeouts * docs(skills): expand Discrawl archive workflow * fix(discord): cap request timeout signals * fix(agents): preserve rotated compaction session identity Fix `sessions.json` persistence after compaction transcript rotation. When the agent runtime rotates from the pre-compaction session transcript to the post-compaction transcript, post-run consumers now receive the effective OpenClaw session id and session file. Backend CLI session ids remain backend metadata and no longer overwrite the top-level OpenClaw session identity. Refs openclaw#88040. Thanks @1052326311. Verification: - `node scripts/run-vitest.mjs src/agents/agent-command.compaction-rotation.test.ts src/agents/agent-command.live-model-switch.test.ts src/agents/command/session-store.test.ts` - Autoreview clean - GitHub CI green on PR head `c3d3c77ddf675bbba0b9ba6681b030a2f69a898c` * fix: keep compaction timeout snapshots continuable * feat(ios): add talk tab realtime playback (openclaw#88105) Merged via squash. Prepared head SHA: f41112a Co-authored-by: ngutman <1540134+ngutman@users.noreply.github.com> Co-authored-by: ngutman <1540134+ngutman@users.noreply.github.com> Reviewed-by: @ngutman * fix(signal): cap container timeout timers * fix(agents): forward ACP spawn attachments Forward initial image/file attachments when spawning ACP subagents through the existing sessions_spawn attachment opt-in. Remove the PR-only acpEnabled config split so ACP uses the same attachment gate as other runtimes. Also fix the PR branch CI fallout: type the browser element CLI request mock and use Vitest env stubs in the Azure speech test to satisfy the changed-path security scan. Verification: - GitHub CI passed on f6ca26b. - Autoreview clean. - Crabbox AWS live OpenAI proof passed: cbx_a576d49493fe / run_081dcc6c6a1b. Thanks @zhangguiping-xydt. * refactor: share e2e bounded response reader * docs(browser): add Notte cloud browser to direct WebSocket CDP providers Notte exposes a CDP-compatible WebSocket gateway at wss://us-prod.notte.cc/sessions/connect?token=<NOTTE_API_KEY> that auto-creates a session on connect — the same shape OpenClaw's existing "Direct WebSocket CDP providers" section was generically framed for (per openclaw#31085). Real behaviour proof (against wss://us-prod.notte.cc/sessions/connect): $ openclaw browser --browser-profile notte open https://example.com opened: https://example.com/ tab: t4 id: 7FE04AC44931A6E1C799DE4ABF0DC807 A screenshot captured against the same session is a 1254x1111 PNG of the rendered example.com page. Playwright connectOverCDP flow against the same URL (today): connectOverCDP 695ms context.newCDPSession(page) 169ms session.send('Target.getTargetInfo') → targetId 87ms page.goto('https://example.com') 631ms total 1.8s AI-assisted (Claude Opus 4.7). codex review --base origin/main returned clean. See PR description for the full pre-flight checklist. Co-authored-by: Claude Opus 4.7 <noreply@anthropic.com> * fix(zalo): cap api request timeouts * fix: stabilize codex supervisor session listing * fix(qa-matrix): cap substrate request timeouts * fix(xiaomi): cap tts request timeouts * refactor: share e2e mock http helpers * docs(skills): require grouped release changelogs * fix(zai): cap endpoint probe timeouts * fix(mattermost): cap dm retry timeouts * perf: reuse provider handles and strict tool schemas * feat: add core session goals (openclaw#87469) * feat: add core session goals * feat: polish session goals in tui * fix: resolve goal tool session stores * fix: keep get goal read-only * fix: migrate legacy goal session slots * fix: persist goal token accounting * fix: validate goal session rows * refactor: remove unshipped goal legacy handling * fix: handle goal commands in local tui * fix: satisfy goal tool display checks * fix: reset goal budget on overdue resume * feat: surface session goals across control surfaces * test: update gateway protocol test import * test: align goal fixture types with protocol * fix: scope selected global transcript usage fallback * fix: scope selected global web subscriptions * fix: preserve selected global agent during chat dispatch * fix: scope chat inject to selected global agents * test: fix timeout mock return types * fix(crestodian): cap probe timeouts * fix: keep live OpenClaw session locks during cleanup (openclaw#88129) Keep session lock cleanup from removing live OpenClaw-owned locks solely because they are old. Cleanup now reports age-only stale locks without deleting them, while still removing dead, orphaned, recycled, malformed-old, and non-OpenClaw-owned locks. Update doctor docs and regression coverage for the cleanup/repair contract. Refs openclaw#87779 * fix(agents): cap model scan timeouts * refactor: share script budget number parsing * fix(provider): cap operation timeouts * fix(usage): cap provider usage fetch timeouts * fix: bound default heartbeat run timeout (openclaw#88133) Fixes openclaw#87438. Bound unset heartbeat run timeouts so background heartbeat turns no longer inherit the built-in 48-hour interactive agent default. Timeout precedence is explicit heartbeat timeout, explicit global agent timeout, then heartbeat cadence capped at 600 seconds. Verification: - git diff --check - Testbox tbx_01kstna69zvznn4fq7zrqr04a1: corepack pnpm test src/infra/heartbeat-runner.model-override.test.ts -- --reporter=verbose passed 13 tests - Direct node --import tsx runtime probe verified 300s, 600s, 60s, and 45s timeout precedence cases - Autoreview clean Known CI state: - PR CI run 26661465248 has failures matching latest main CI run 26661386468 at a7820b2; failures are outside this six-file heartbeat/docs diff. * fix(signal): cap client request timeouts * fix(feishu): cap async helper timeouts * refactor: share script bounded response reader * fix: move compaction planning off the event loop Move compaction planning work to a bounded worker-thread path so large transcript planning no longer monopolizes the agent event loop. Extract pure planning helpers, sanitize worker inputs before structured clone, package the worker entrypoint, and keep synchronous fallback only for worker-unavailable cases. Fixes openclaw#86358. * fix(browser): cap control fetch timeouts * fix(ci): repair main checks * fix(browser): cap node runtime timeouts * fix(codex-supervisor): centralize session limit parsing * fix(discord): cap monitor helper timeouts * perf: reuse gateway runtime metadata * fix(acp): cap turn timeout timers * refactor: share media temp save wrapper * fix(tts): cap speech provider timeouts * fix(media): cap generation provider timeouts * fix ci mainline checks (openclaw#88137) * fix(infra): cap request body timeouts * ci: stabilize main checks * feat: add skills index * perf: avoid unnecessary skills index maps * refactor: share skill command exposure policy * perf: centralize skill status lookup * refactor: reuse shared skills prompt formatter * perf: reuse resolved skills allowlist * perf: speed up skills filtering * perf: prepare bundled skill allowlist once * perf: use set for bundled skill allowlist * test: preserve real skills status exports * test: share skills entry fixtures * test: remove duplicate skill fixture wrappers * test: complete skills status mock surface * fix(gateway-client): cap stop wait timeout * perf: prefer package-local bundled plugin artifacts * fix(openai): cap codex oauth preflight timeout * fix(supervisor): narrow stored session limit parsing * refactor: share diagnostics timeline span helpers * fix(ci): repair main checks * fix(ci): break skills loading cycle * test: fix main CI regressions * fix(apns): cap relay timeout * fix(infra): cap jsonl socket timeouts * fix(infra): cap shell env timeouts * test: stabilize remaining CI flakes * fix(apns): cap direct timeout paths * Add plugin manifest contract for SecretRef provider integrations (openclaw#82326) * secret-provider-integrations Signed-off-by: sallyom <somalley@redhat.com> * feat(secrets): configure plugin provider presets * secrets: use plugin-managed provider refs Signed-off-by: sallyom <somalley@redhat.com> * fix secretref auth profile service env * test secret provider integration e2e * fix secretref plugin config service env * fix secret provider preset schema alignment * stabilize secret provider service proof * validate secret provider plugin integrations * harden secret provider resolver paths * scope secret provider config validation * stabilize openai secret provider proof * fix secret provider metadata proof * stabilize config baseline proof * fix secret provider e2e lint --------- Signed-off-by: sallyom <somalley@redhat.com> Co-authored-by: joshavant <830519+joshavant@users.noreply.github.com> * fix(proxy): cap connect tunnel timeouts * fix: route media completions through requester agent (openclaw#88141) * fix(scripts): cap issue labeler response bodies * refactor: share media understanding post params * fix(infra): cap transport readiness timeouts * ci: reduce main workflow critical path * test(gateway): stabilize live helper shard * refactor: share native approval route gates Share native approval route gate helpers across mainstream channel approval runtimes and keep PR openclaw#87770 green on current main. * fix(channels): centralize stall watchdog timer bounds * perf: resolve native esm plugin sdk imports * test: stabilize infra state shard * fix(nostr): cap profile import relay timers * test(infra): stabilize main CI tests * test(infra): preserve script wrapper fixture * fix(web): cap guarded fetch timeout seconds * fix(zalouser): cap probe timeout timer * refactor: add shared sqlite state database Adds the shared SQLite state database base, moves plugin keyed state into it with doctor migration coverage, and keeps generated Kysely guardrails aligned. Proof: focused SQLite/plugin-state tests, db:kysely:check, lint:kysely, architecture/dependency guards, autoreview, and PR CI all clean. * fix(codex): recover app-server completion stalls Fix Codex app-server completion-stall recovery so replay-safe stdio completion-idle failures retry once, while progress/terminal turn-watch timeouts only surface timeout payloads. Also preserve post-tool completion guards for scoped native response deltas and stabilize the oversized CONNECT timeout regression test picked up from latest main. Co-authored-by: Kelaw - Keshav's Agent <keshavbotagent@gmail.com> * fix(ci): repair main normalization checks * fix(zalouser): cap qr login timeouts * fix(dev): cap Discord smoke response bodies * fix(agents): centralize terminal run outcome precedence (openclaw#88136) * fix(agents): centralize terminal run outcome precedence * docs(agents): explain terminal outcome precedence * docs(agents): note terminal outcome helper * fix(agents): preserve pending hard timeout over late completion * test(agents): align global session scoping expectation * Revert "test(agents): align global session scoping expectation" This reverts commit 9b4a0c3. * test(infra): stabilize CONNECT timeout cap test * fix(agents): prioritize hard timeout terminal evidence * fix(gateway): preserve pending hard timeout snapshots * ci: skip bundled dts in artifact build * fix(memory): cap qmd process timeouts * fix(ci): repair main lint gates * test(infra): avoid max fake-timer jumps (openclaw#88155) * fix(whatsapp): cap credential flush timeout * ci: satisfy build profile lint * refactor: share live transport scenario helpers * fix(telegram): cap polling lease wait timer * fix(release): avoid gh api for candidate reads * fix(release): harden candidate run status polling * fix(feishu): reopen retryable bot menu replay * fix(release): avoid gh api in beta smoke * fix(release): build beta smoke REST curl command * test(realtime): stabilize websocket timeout test * test: stabilize realtime websocket timeout * fix(telegram): centralize positive timer bounds * fix(providers): cap local service timers * refactor: share provider oauth runtime helpers * fix(openrouter): cap music stream timeout * fix(release): harden release ci summary lookup * fix(fal): cap video queue deadline * test(ci): stabilize tool search gateway timeout helper * fix(reply): hide ACP tool traces from Telegram Telegram's surface renders tool-call traces poorly compared to Discord's. Add a per-channel visibility isolation list (currently just `telegram`) so the dispatch-acp delivery coordinator drops tool/status payloads to those channels and rewrites error payloads to a sanitized message that points to local logs instead of leaking the trace. - New ACP_VISIBILITY_ISOLATED_CHANNELS set + helper prepareAcpPayloadForChannelVisibility - Coordinator picks the effective target channel (originating or direct) and skips delivery when the payload is a tool / status / error trace - 89 lines of test coverage in dispatch-acp.test.ts for the new path --------- Signed-off-by: sallyom <somalley@redhat.com> Co-authored-by: joshavant <830519+joshavant@users.noreply.github.com> Co-authored-by: Vincent Koc <25068+vincentkoc@users.noreply.github.com> Co-authored-by: jesse-merhi <79823012+jesse-merhi@users.noreply.github.com> Co-authored-by: Peter Steinberger <steipete@gmail.com> Co-authored-by: Vincent Koc <vincentkoc@ieee.org> Co-authored-by: Shadow <shadow@openclaw.ai> Co-authored-by: Gio Della-Libera <giodl73@gmail.com> Co-authored-by: giodl73-repo <235387111+giodl73-repo@users.noreply.github.com> Co-authored-by: Ayaan Zaidi <hi@obviy.us> Co-authored-by: Shakker <shakkerdroid@gmail.com> Co-authored-by: Peter Steinberger <peter@steipete.me> Co-authored-by: benjamin1492 <35176637+benjamin1492@users.noreply.github.com> Co-authored-by: Nimrod Gutman <nimrod.gutman@gmail.com> Co-authored-by: ngutman <1540134+ngutman@users.noreply.github.com> Co-authored-by: Dallin Romney <dallinromney@gmail.com> Co-authored-by: xin zhuang <65798732+1052326311@users.noreply.github.com> Co-authored-by: zhang-guiping <zhang.guiping@xydigit.com> Co-authored-by: Lucas Giordano <giordano3102lucas@gmail.com> Co-authored-by: Claude Opus 4.7 <noreply@anthropic.com> Co-authored-by: Sally O'Malley <somalley@redhat.com> Co-authored-by: Kevin Lin <kevin@dendron.so> Co-authored-by: keshavbotagent <keshavbotagent@gmail.com>

@benjamin1492

Fix claude-cli transcript resume so session-id rotation and transcript flush timing do not drop valid resume state. - Capture the latest claude-cli session_id from JSONL output. - Resolve Claude project transcript paths through the shared canonical project-dir resolver. - Probe transcript content from the actual CLI process cwd. - Thanks @benjamin1492!

…026.5.28) (#759) This PR contains the following updates: | Package | Update | Change | |---|---|---| | [ghcr.io/openclaw/openclaw](https://openclaw.ai) ([source](https://github.com/openclaw/openclaw)) | patch | `2026.5.27` → `2026.5.28` | --- ### Release Notes <details> <summary>openclaw/openclaw (ghcr.io/openclaw/openclaw)</summary> ### [`v2026.5.28`](https://github.com/openclaw/openclaw/blob/HEAD/CHANGELOG.md#2026528) [Compare Source](openclaw/openclaw@v2026.5.27...v2026.5.28) ##### Highlights - Agent and Codex runtime recovery is steadier: subagents keep cwd/workspace separation, hook context stays prompt-local, session locks release on timeout abort while live OpenClaw locks survive cleanup, stale restart continuations are avoided, and Codex app-server/helper failures no longer tear down shared runtime state. ([#87218](openclaw/openclaw#87218), [#86875](openclaw/openclaw#86875), [#87409](openclaw/openclaw#87409), [#87399](openclaw/openclaw#87399), [#87375](openclaw/openclaw#87375), [#88129](openclaw/openclaw#88129)) - Channel delivery and session identity got safer across outbound plugin hooks, Matrix room ids, iMessage reactions/approvals, Slack final replies, Discord recovered tool warnings, runtime-config message actions, WhatsApp profile auth roots, Telegram polling, and Microsoft Teams service URL trust checks. ([#73706](openclaw/openclaw#73706), [#75670](openclaw/openclaw#75670), [#87366](openclaw/openclaw#87366), [#87451](openclaw/openclaw#87451), [#87334](openclaw/openclaw#87334), [#84535](openclaw/openclaw#84535), [#82492](openclaw/openclaw#82492), [#83304](openclaw/openclaw#83304), [#87160](openclaw/openclaw#87160)) - Mobile and chat surfaces got a broader refresh: the iOS Pro UI, hosted push relay default, realtime Talk tab playback, Gateway chat transport, onboarding, Talk permissions, WebChat reconnect delivery, and session picker behavior now preserve more state across reconnects and empty searches. ([#87367](openclaw/openclaw#87367), [#87531](openclaw/openclaw#87531), [#87682](openclaw/openclaw#87682), [#88096](openclaw/openclaw#88096), [#88105](openclaw/openclaw#88105)) Thanks [@ngutman](https://github.com/ngutman) and [@BunsDev](https://github.com/BunsDev). - Browser, channel, and automation inputs are stricter: Browser tool timeouts, viewport/tab indices, Gateway ports, cron retry handling, Discord component ids, schema array refs, Telegram callback pages, and channel progress callbacks now reject malformed values earlier and preserve the intended delivery context. ([#82887](openclaw/openclaw#82887)) - Provider, media, and document coverage expands with Claude Opus 4.8, Fal Krea image schemas, NVIDIA featured models, MiniMax streaming music responses, encrypted PDF extraction, voice model catalogs, GitHub Copilot agent runtime support, and a Codex Supervisor plugin path for delegated Codex workflows. ([#87845](openclaw/openclaw#87845), [#87890](openclaw/openclaw#87890), [#80775](openclaw/openclaw#80775), [#84764](openclaw/openclaw#84764), [#87751](openclaw/openclaw#87751), [#87794](openclaw/openclaw#87794)) - CLI, auth, doctor, and provider paths fail faster and recover more clearly: malformed numeric/version options are rejected, workspace dotenv provider credentials are ignored, heartbeat defaults, OAuth/token lifetimes, and local service startup requests are bounded, agent auth health labels are clearer, legacy `api_key` auth profiles migrate to canonical form, and restart guidance is actionable. ([#87398](openclaw/openclaw#87398), [#86281](openclaw/openclaw#86281), [#87361](openclaw/openclaw#87361), [#88133](openclaw/openclaw#88133), [#83655](openclaw/openclaw#83655), [#87559](openclaw/openclaw#87559), [#88088](openclaw/openclaw#88088), [#85924](openclaw/openclaw#85924)) Thanks [@vincentkoc](https://github.com/vincentkoc) and [@giodl73-repo](https://github.com/giodl73-repo). - Plugin and Gateway hot paths do less repeated work while preserving cache correctness for install records, config JSON parsing, tool search catalogs, session stores, manifest model rows, auto-enabled plugin config, browser tokens, viewer assets, and release-split external plugin packages. ([#86699](openclaw/openclaw#86699)) - Release, QA, and E2E validation now bound more log, artifact, harness, and cross-OS waits so failing lanes produce proof instead of hanging or false-greening. ##### Changes - Status: show active subagent details in status output. - Diffs: split the default language pack and expand default Diffs language coverage while keeping the host floor aligned. ([#87370](openclaw/openclaw#87370), [#87372](openclaw/openclaw#87372)) Thanks [@RomneyDa](https://github.com/RomneyDa). - ClawHub: add plugin display names plus skill verification and trust surfaces. ([#87354](openclaw/openclaw#87354), [#86699](openclaw/openclaw#86699)) Thanks [@thewilloftheshadow](https://github.com/thewilloftheshadow) and [@Patrick-Erichsen](https://github.com/Patrick-Erichsen). - iOS: refresh the dev app with Pro Command, Chat, Agents, Settings, hosted push relay defaults, and realtime Talk playback wired to gateway sessions, diagnostics, chat, and realtime Talk. ([#87367](openclaw/openclaw#87367), [#88096](openclaw/openclaw#88096), [#88105](openclaw/openclaw#88105)) Thanks [@Solvely-Colin](https://github.com/Solvely-Colin) and [@ngutman](https://github.com/ngutman). - Docs: clarify Codex computer-use setup, paste-token stdin auth setup, macOS gateway sleep troubleshooting, native Codex hook relay recovery, container model auth, install deployment cards, device-token admin gating, CLI setup flow compatibility, Notte cloud browser CDP setup, and backport targets. ([#87313](openclaw/openclaw#87313), [#63050](openclaw/openclaw#63050), [#87685](openclaw/openclaw#87685)) Thanks [@bdjben](https://github.com/bdjben), [@liaoandi](https://github.com/liaoandi), and [@thewilloftheshadow](https://github.com/thewilloftheshadow). - PDF/tools: use ClawPDF for PDF extraction, support encrypted PDF extraction, and surface MCP structured content in agent tool results. ([#87670](openclaw/openclaw#87670), [#87751](openclaw/openclaw#87751)) - Providers: add Claude Opus 4.8 support, Fal Krea image model schemas, NVIDIA featured model catalogs, MiniMax streaming music responses, and provider-backed voice model catalogs. ([#87845](openclaw/openclaw#87845), [#87890](openclaw/openclaw#87890), [#80775](openclaw/openclaw#80775), [#84764](openclaw/openclaw#84764), [#87794](openclaw/openclaw#87794)) Thanks [@eleqtrizit](https://github.com/eleqtrizit) and [@vincentkoc](https://github.com/vincentkoc). - Codex/GitHub: add the GitHub Copilot agent runtime and the Codex Supervisor plugin package. - Plugins: externalize GitHub Copilot and Tokenjuice as official install-on-demand plugins with npm and ClawHub publish metadata. - Workboard: add agent coordination tools for tracking and handing off active agent work. - Discord: show commentary in progress drafts so live Discord runs expose useful in-progress context. ([#85200](openclaw/openclaw#85200)) - Plugin SDK: add a reply payload sending hook for plugins that need to deliver channel-owned replies and flatten package types for SDK declarations. ([#82823](openclaw/openclaw#82823), [#87165](openclaw/openclaw#87165)) Thanks [@piersonr](https://github.com/piersonr) and [@RomneyDa](https://github.com/RomneyDa). - Policy: add policy comparison, ingress-channel conformance, and sandbox-posture conformance checks. ([#85572](openclaw/openclaw#85572), [#85744](openclaw/openclaw#85744), [#86768](openclaw/openclaw#86768)) ##### Fixes - Agents: fall back to local config pruning when the optional `agents delete` Gateway probe cannot authenticate, so offline installs can still delete agents without removing shared workspaces. - Tighten phone-control mutation authorization \[AI]. ([#87150](openclaw/openclaw#87150)) Thanks [@pgondhi987](https://github.com/pgondhi987). - Clarify directive persistence authorization policy \[AI]. ([#86369](openclaw/openclaw#86369)) Thanks [@pgondhi987](https://github.com/pgondhi987). - Agents/Codex: keep spawned agent cwd/workspace state separated, forward ACP spawn attachments, keep hook context prompt-local, release session locks on timeout abort and runtime teardown without deleting live OpenClaw-owned locks during cleanup, avoid session event queue self-wait, clean up exec abort listeners, stream assistant deltas incrementally, recover raw missing-thread compaction failures, preserve rotated compaction session identity, keep compaction-timeout snapshots continuable, preserve shared app-server state across startup or helper failures, keep native hook relay alive across restarts and prune stale bridge files, close native hook relay replacement races, keep Claude live tool progress visible for watchdog recovery, suppress abandoned requester completion handoff, route workspace memory through tools, resolve Codex runtime models first, report quarantined dynamic tools, format `skills` command output, bind node auto-review to prepared plans, retry Claude CLI transcript probes, and bound compaction/steering retries. ([#87218](openclaw/openclaw#87218), [#86875](openclaw/openclaw#86875), [#86123](openclaw/openclaw#86123), [#88129](openclaw/openclaw#88129), [#87399](openclaw/openclaw#87399), [#87375](openclaw/openclaw#87375), [#72574](openclaw/openclaw#72574), [#87383](openclaw/openclaw#87383), [#87400](openclaw/openclaw#87400), [#83022](openclaw/openclaw#83022), [#87671](openclaw/openclaw#87671), [#87738](openclaw/openclaw#87738), [#87747](openclaw/openclaw#87747), [#87706](openclaw/openclaw#87706), [#87546](openclaw/openclaw#87546), [#87541](openclaw/openclaw#87541), [#81048](openclaw/openclaw#81048)) Thanks [@mbelinky](https://github.com/mbelinky), [@Alix-007](https://github.com/Alix-007), [@luoyanglang](https://github.com/luoyanglang), [@yetval](https://github.com/yetval), [@sjf](https://github.com/sjf), [@joshavant](https://github.com/joshavant), [@benjamin1492](https://github.com/benjamin1492), [@c19354837](https://github.com/c19354837), [@fuller-stack-dev](https://github.com/fuller-stack-dev), [@pfrederiksen](https://github.com/pfrederiksen), and [@dodge1218](https://github.com/dodge1218). - Codex Supervisor: keep real-home app-server MCP session listing on the loaded state path, bound stored history scans, and close WebSocket probes cleanly. - Channels: thread canonical session keys into outbound hooks, preserve Matrix room-id case, keep fallback tool warnings mention-inert, retain delivered Slack final replies during late cleanup, continue iMessage polling after denied reactions, suppress duplicate native exec approvals, resolve Gateway message actions against the active runtime config, preserve Telegram SecretRef prompt config and polling keepalives, preserve WhatsApp profile auth roots, QR display, document filenames, and plugin hook config, suppress Discord recovered tool warnings, preserve the Discord voice outbound helper, cap Discord/Signal/Zalo channel request and container timeouts, and block untrusted Teams service URLs while keeping TeamsSDK patterns aligned. ([#73706](openclaw/openclaw#73706), [#75670](openclaw/openclaw#75670), [#87366](openclaw/openclaw#87366), [#87451](openclaw/openclaw#87451), [#87465](openclaw/openclaw#87465), [#87334](openclaw/openclaw#87334), [#84535](openclaw/openclaw#84535), [#76262](openclaw/openclaw#76262), [#83304](openclaw/openclaw#83304), [#82492](openclaw/openclaw#82492), [#87581](openclaw/openclaw#87581), [#77114](openclaw/openclaw#77114), [#86426](openclaw/openclaw#86426), [#85529](openclaw/openclaw#85529), [#87160](openclaw/openclaw#87160)) Thanks [@zeroaltitude](https://github.com/zeroaltitude), [@lukeboyett](https://github.com/lukeboyett), [@jarvis-mns1](https://github.com/jarvis-mns1), [@xiaotian](https://github.com/xiaotian), [@funmerlin](https://github.com/funmerlin), [@joshavant](https://github.com/joshavant), [@eleqtrizit](https://github.com/eleqtrizit), [@heyitsaamir](https://github.com/heyitsaamir), [@amittell](https://github.com/amittell), [@lidge-jun](https://github.com/lidge-jun), [@liorb-mountapps](https://github.com/liorb-mountapps), [@masatohoshino](https://github.com/masatohoshino), [@bladin](https://github.com/bladin), and [@giodl73-repo](https://github.com/giodl73-repo). - CLI/auth/doctor/providers: reject malformed numeric/timeout/subcommand-version inputs, ignore workspace dotenv provider credentials, wait for respawn child shutdown, bound heartbeat defaults plus Codex, GitHub Copilot, OpenAI, Anthropic, Google, Feishu, LM Studio, MiniMax, Xiaomi TTS, and local-provider OAuth/token/model requests, harden Codex auth probes, label auth health by agent, preserve explicit agentRuntime pins during Codex model migration, warm provider auth off the main thread, honor Codex response timeouts, stop migrating current Claude Haiku 4.5 profiles to Sonnet, bound local service startup, resolve GPT-5.5 without cached catalog, migrate legacy memory auto-provider config, rewrite non-canonical `api_key` auth profiles, and make doctor restart follow-ups actionable. ([#87398](openclaw/openclaw#87398), [#86281](openclaw/openclaw#86281), [#87361](openclaw/openclaw#87361), [#88133](openclaw/openclaw#88133), [#83655](openclaw/openclaw#83655), [#87559](openclaw/openclaw#87559), [#87719](openclaw/openclaw#87719), [#88088](openclaw/openclaw#88088), [#85924](openclaw/openclaw#85924), [#84362](openclaw/openclaw#84362)) Thanks [@Patrick-Erichsen](https://github.com/Patrick-Erichsen), [@samzong](https://github.com/samzong), [@giodl73-repo](https://github.com/giodl73-repo), [@alkor2000](https://github.com/alkor2000), [@mmaps](https://github.com/mmaps), [@nxmxbbd](https://github.com/nxmxbbd), and [@vincentkoc](https://github.com/vincentkoc). - Gateway/security/session state: expire browser tokens after auth rotation, scope assistant idempotency dedupe, drain probe client closes, avoid stale restart continuation reuse, preserve retry-after fallbacks and stale rate-limit cooldown probes, bound webchat image and artifact transcript scans, include seconds in inbound metadata timestamps, clear completed session active runs, clear stale chat stream buffers, and evict current plugin-state namespaces at row caps. ([#87810](openclaw/openclaw#87810), [#87833](openclaw/openclaw#87833), [#75089](openclaw/openclaw#75089)) Thanks [@joshavant](https://github.com/joshavant) and [@litang9](https://github.com/litang9). - Config/parsing/network: reject partial numeric parsing, parse provider/Discord retry headers and dates strictly, honor IPv6 and bare IPv6 `no_proxy` entries, preserve empty plugin allowlists, canonicalize secret target array indexes, and reject malformed media content lengths, inspected TCP ports, marketplace content lengths, cron epochs, sandbox stat fields, unsafe duration values, empty config path segments, noncanonical schema array refs, unsafe Telegram callback pages, and invalid Teams attachment-fetch DNS targets. ([#87883](openclaw/openclaw#87883)) Thanks [@zhangguiping-xydt](https://github.com/zhangguiping-xydt). - Browser/input hardening: reject invalid tab indexes, excessive viewport resizes, explicit zero CDP ports, malformed geolocation options, unsafe screenshot or permission-grant timeouts, loose response-body limits, invalid cookie expiries, and non-finite Browser tool delays/timeouts. - Cron/automation: retry recurring jobs after transient model rate limits before waiting for the next scheduled slot, and preflight model fallbacks before skipping scheduled work. ([#82887](openclaw/openclaw#82887)) Thanks [@chen-zhang-cs-code](https://github.com/chen-zhang-cs-code). - Auto-reply/directives: respect provider and relayed channel metadata during directive persistence so channel-originated decisions keep their intended context. ([#87683](openclaw/openclaw#87683)) - WhatsApp: resolve the auth directory from the active profile so profile-scoped WhatsApp installs do not drift to the wrong credential root. ([#82492](openclaw/openclaw#82492)) Thanks [@lidge-jun](https://github.com/lidge-jun). - Gateway/session state: clear completed session active runs, avoid cold-loading providers for MCP inventory, cache single-session child indexes, cap handshake timers, and bound preauth, auth-guard, media, transcript, readiness, and port options. - Channels/replies: preserve channel-owned progress callbacks when verbose output is off, keep group-room progress suppression intact, prefer external session delivery context, escape Discord component id delimiters, force final TUI chat repaints, show Slack reasoning previews, and normalize Discord/Matrix/Mattermost channel numeric options. ([#87476](openclaw/openclaw#87476), [#87423](openclaw/openclaw#87423)) - Agents/tool args: harden smart-quoted argument repair for edit arrays and exact escaped arguments so model-produced tool calls recover without corrupting valid input. ([#86611](openclaw/openclaw#86611)) Thanks [@ferminquant](https://github.com/ferminquant). - Providers/agents: preserve seeded Anthropic signatures, preserve signed thinking payloads, concatenate signature-delta chunks, preserve DeepSeek `reasoning_content` replay across tier suffixes, apply OpenRouter strict9 ids to Mistral routes, promote Ollama plain-text tool calls, load NVIDIA featured model catalogs, stream MiniMax music generation responses, and recover empty preflight compaction. ([#87593](openclaw/openclaw#87593), [#87493](openclaw/openclaw#87493), [#80775](openclaw/openclaw#80775), [#84764](openclaw/openclaw#84764)) Thanks [@Pluviobyte](https://github.com/Pluviobyte) and [@eleqtrizit](https://github.com/eleqtrizit). - Media/images: skip CLI image cache refs when resolving generated images, allow trusted generated HTML attachments, and bound generated video downloads so stale refs and slow providers fail cleanly. ([#87523](openclaw/openclaw#87523), [#87982](openclaw/openclaw#87982)) - File transfer: handle late tar stdin pipe errors after archive validation or unpacking has already settled. - Performance: trust install-record caches between reloads, prefer native JSON parsing, reuse unchanged tool-search catalogs, reuse gateway session and plugin metadata paths, skip unchanged store serialization, patch single-entry session writes, add precomputed session patch writers, reduce store clone allocations, cache manifest model catalog rows and auto-enabled plugin config, avoid full session snapshots for entry reads, defer configured Slack full startup, prefer bundled plugin dist entries, and slim current metadata identity caches. ([#87760](openclaw/openclaw#87760)) - Docker/release/QA: package runtime workspace templates, stream cross-OS served artifacts, preserve sparse Crabbox run artifacts, isolate npm plugin installs per package, reject incompatible package plugin API installs, drop the leftover root Sharp dependency from package manifests after the Rastermill migration, bound OpenClaw instance logs, plugin gauntlet relay logs, MCP channel buffers, kitchen-sink scans, agent-turn assertions, QA-Lab credential broker calls, QA Matrix substrate requests, and release scenario logs, and keep release/google live guards current. ([#87647](openclaw/openclaw#87647), [#87477](openclaw/openclaw#87477)) Thanks [@rohitjavvadi](https://github.com/rohitjavvadi) and [@vincentkoc](https://github.com/vincentkoc). - Release/CI: bound manual git fetches, ClawHub verifier responses, ClawHub owner metadata, dependency-guard error bodies, Parallels limits, startup/test/memory budget parsing, and diffs viewer build warnings so release lanes fail with useful proof instead of hanging. ([#87839](openclaw/openclaw#87839)) </details> --- ### Configuration 📅 **Schedule**: Branch creation - At any time (no schedule defined), Automerge - At any time (no schedule defined). 🚦 **Automerge**: Disabled by config. Please merge this manually once you are satisfied. ♻ **Rebasing**: Whenever PR becomes conflicted, or you tick the rebase/retry checkbox. 🔕 **Ignore**: Close this PR and you won't be reminded about these updates again. --- - [ ] If you want to rebase/retry this PR, check this box --- This PR has been generated by [Renovate Bot](https://github.com/renovatebot/renovate).  Reviewed-on: https://git.erwanleboucher.dev/eleboucher/homelab/pulls/759

@benjamin1492

Fix claude-cli transcript resume so session-id rotation and transcript flush timing do not drop valid resume state. - Capture the latest claude-cli session_id from JSONL output. - Resolve Claude project transcript paths through the shared canonical project-dir resolver. - Probe transcript content from the actual CLI process cwd. - Thanks @benjamin1492!

@benjamin1492

Fix claude-cli transcript resume so session-id rotation and transcript flush timing do not drop valid resume state. - Capture the latest claude-cli session_id from JSONL output. - Resolve Claude project transcript paths through the shared canonical project-dir resolver. - Probe transcript content from the actual CLI process cwd. - Thanks @benjamin1492!

openclaw-barnacle Bot added agents Agent runtime and tooling size: M triage: needs-real-behavior-proof Candidate: external PR needs after-fix proof from a real setup. labels May 12, 2026

openclaw-barnacle Bot added proof: supplied External PR includes structured after-fix real behavior proof. and removed triage: needs-real-behavior-proof Candidate: external PR needs after-fix proof from a real setup. labels May 12, 2026

clawsweeper Bot added the proof: sufficient ClawSweeper judged the real behavior proof convincing. label May 12, 2026

openclaw-barnacle Bot removed the proof: sufficient ClawSweeper judged the real behavior proof convincing. label May 12, 2026

benjamin1492 marked this pull request as ready for review May 12, 2026 15:08

clawsweeper Bot added the proof: sufficient ClawSweeper judged the real behavior proof convincing. label May 12, 2026

openclaw-barnacle Bot removed the proof: sufficient ClawSweeper judged the real behavior proof convincing. label May 13, 2026

This was referenced May 14, 2026

[Bug]: OpenClaw session transcript stops flushing mid-session on claude-cli runtime (Claude CLI side keeps updating) #78273

Open

fix(agents/cli-runner): gate cliSessionBinding persist on transcript flush #81821

Closed

This was referenced May 15, 2026

claude-cli Telegram DM context loss from missing-transcript race before transcript flush #77974

Closed

[Bug]: claudeCliSessionTranscriptHasContent races claude-cli's transcript flush, returning false negatives that force cold session starts #81042

Closed

adele-with-a-b mentioned this pull request May 19, 2026

fix(cli-runner): write-side flush gate + orphan-tool-use invalidator #84234

Merged

4 tasks

clawsweeper Bot mentioned this pull request May 23, 2026

claude-cli runner emits false missing-transcript resets from stale session-id binding #85757

Closed

obviyus self-assigned this May 29, 2026

obviyus force-pushed the fix/cli-transcript-probe-flush-race branch from 952f92a to e3df84a Compare May 29, 2026 17:13

openclaw-barnacle Bot added commands Command implementations size: L and removed size: M labels May 29, 2026

benjamin1492 added 4 commits May 29, 2026 22:45

fix(command): retry claude-cli transcript probe to close flush race

00ef1a2

fix(command): widen claude-cli transcript probe back-off to 5 attempt…

d9ce589

…s (3250ms)

fix(command): pivot claude-cli transcript probe to --session-id strat…

498a475

…egy (v4)

benjamin1492 and others added 3 commits May 29, 2026 22:45

test: drop redundant string cast in v4 transcript warn assertion

080b464

oxlint flagged the `as string` cast on warnSpy.mock.calls[0]?.[0] as unnecessary — the value is already typed as string for the assertion path.

fix(command): share claude transcript project resolver

383e50a

fix(command): probe claude transcript from cli cwd

bf260ae

obviyus force-pushed the fix/cli-transcript-probe-flush-race branch from e3df84a to bf260ae Compare May 29, 2026 17:20

obviyus merged commit de45530 into openclaw:main May 29, 2026
120 of 126 checks passed

github-actions Bot mentioned this pull request May 29, 2026

📡 Upstream Digest — 2026-05-29 18:27 UTC curtismercier/openclaw-mods#972

Open

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

fix(command): retry claude-cli transcript probe to close flush race#81048

fix(command): retry claude-cli transcript probe to close flush race#81048
obviyus merged 7 commits into
openclaw:mainfrom
benjamin1492:fix/cli-transcript-probe-flush-race

benjamin1492 commented May 12, 2026 •

edited

Loading

Uh oh!

clawsweeper Bot commented May 12, 2026 •

edited

Loading

Uh oh!

benjamin1492 commented May 13, 2026

Uh oh!

benjamin1492 commented May 13, 2026 •

edited by clawsweeper Bot

Loading

Uh oh!

benjamin1492 commented May 14, 2026

Uh oh!

benjamin1492 commented May 15, 2026

Uh oh!

Uh oh!

obviyus commented May 29, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Uh oh!

Conversation

benjamin1492 commented May 12, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Closes

Test plan

Affected scope

Risk

Real Behavior Proof

Uh oh!

clawsweeper Bot commented May 12, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

benjamin1492 commented May 13, 2026

Update: probe widened to 5 attempts / 3250ms after live evidence v1 + v2 were both insufficient

Uh oh!

benjamin1492 commented May 13, 2026 • edited by clawsweeper Bot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

v4: orchestrator owns the path

Why this is the last call site change

Tests

Uh oh!

benjamin1492 commented May 14, 2026

v5 — parseCliJsonl last-wins (commit e628d42)

Uh oh!

benjamin1492 commented May 15, 2026

Overnight soak update — v5 holding clean

Lesson learned on this PR

Uh oh!

Uh oh!

obviyus commented May 29, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

benjamin1492 commented May 12, 2026 •

edited

Loading

clawsweeper Bot commented May 12, 2026 •

edited

Loading

benjamin1492 commented May 13, 2026 •

edited by clawsweeper Bot

Loading

v5 — `parseCliJsonl` last-wins (commit `e628d42`)