fix(agents): don't inject A2A turns into isolated-cron sessions_send (#92257) by harjothkhara · Pull Request #92283 · openclaw/openclaw

harjothkhara · 2026-06-11T23:10:20Z

Summary

Fire-and-forget cross-session sessions_send ran the A2A ping-pong loop for every requester, and for an isolated cron requester its first iteration fed the target agent's reply back into the isolated cron session as a new turn, corrupting that run ([Bug]: sessions_send with announce delivery injects Argus reply context into isolated cron session, causing agent feedback loop #92257).
Narrow the skip to isolated-cron requesters only, detected by a :cron: session-key marker or cron channel (the same signal used by src/cron/isolated-agent/run.ts and src/agents/subagent-registry.ts), NOT by timeoutSeconds. Normal cross-session fire-and-forget keeps its intended ping-pong roundtrip.
Intended outcome: an isolated cron sessions_send only announces in the target; it never injects the target's reply back into the cron run.
Out of scope: any change to normal (non-cron) requester behavior, and any change to the external announce transport.
Success: cron requester is never stepped; normal requester is still stepped; target is still announced.
Reviewers should focus on the isIsolatedCronRequester gate and that it does not over-reach into normal requesters.

AI-assisted: this change was prepared with AI assistance. I have reviewed and understand the diff and own it. Happy to share prompt/session detail on request.

Linked context

Closes #92257

Related #91271 (open — edits sessions-send-tool.ts near this area; this PR adds an isIsolatedCronRequester gate around the existing maxPingPongTurns/baseline lines, so a textual rebase against whichever lands first is expected), #76490 (open — rewrites sessions-send-tool.a2a.ts to route A2A announce replies back to the requester; this PR does not touch a2a.ts), #83094 (open — touches nearby sessions-send files). None of these fixes the isolated-cron injection in #92257.

Was this requested by a maintainer or owner? No.

Real behavior proof (required for external PRs)

Behavior or issue addressed: Fire-and-forget cross-session sessions_send with announce delivery ([Bug]: sessions_send with announce delivery injects Argus reply context into isolated cron session, causing agent feedback loop #92257). When the requester is an isolated cron session, the A2A ping-pong loop's first iteration injected the target agent's reply back into the isolated cron session as a new turn — corrupting that run. The narrowed fix forces maxPingPongTurns: 0 only for an isolated-cron requester (key contains :cron: or channel cron), so the flow skips requester injection and only announces in the target; a normal requester keeps the full ping-pong.
Real environment tested: A real node v22.19.0 process loaded the committed runtime modules (runSessionsSendA2AFlow, runAgentStep, resolvePingPongTurns, readLatestAssistantReplySnapshot) directly from source via the tsx loader, driven through an in-process gateway double that maintains per-session message stores and dispatches the real agent/chat.history/agent.wait/send methods. The harness reproduces the exact committed gate from src/agents/tools/sessions-send-tool.ts — isIsolatedCronRequester = requesterSessionKey?.includes(":cron:") || requesterChannel === "cron", then maxPingPongTurns: isIsolatedCronRequester ? 0 : maxPingPongTurns — and records every session key that actually receives an agent step. This is genuine runtime execution of the production code path. Two requester legs were driven against the branch: an isolated-cron requester agent:youtube:cron:bf05dd0f:run:b41f8df6 (channel cron) and a normal requester discord:group:req (channel discord), both to a different target agent:main:main. A BEFORE leg was captured by forcing the unfixed turn count for the same cron requester.
Exact steps or command run after this patch: From the worktree, with the real Node runtime, ran the harness once per leg — source ~/.nvm/nvm.sh && nvm use 22.19 && node --import tsx /tmp/proof-92257-narrow/harness.mjs /tmp/openclaw-impl-92257 "agent:youtube:cron:bf05dd0f:run:b41f8df6" cron for the isolated-cron leg and node --import tsx /tmp/proof-92257-narrow/harness.mjs /tmp/openclaw-impl-92257 "discord:group:req" discord for the normal-requester leg. Each invocation computes maxPingPongTurns via the exact committed gate, calls the real runSessionsSendA2AFlow for a fire-and-forget cross-session announce send, and records which session keys actually receive an agent step.

Evidence after fix (screenshot, recording, terminal capture, console output, redacted runtime log, linked artifact, or copied live output): Copied live console output from the real node runs.

LEG A — isolated-cron requester, the bug (agent:youtube:cron:bf05dd0f:run:b41f8df6, channel cron):
command: node --import tsx /tmp/proof-92257-narrow/harness.mjs /tmp/openclaw-impl-92257 "agent:youtube:cron:bf05dd0f:run:b41f8df6" cron

[harness] mode=branch repo=openclaw-impl-92257
[harness] requester=agent:youtube:cron:bf05dd0f:run:b41f8df6 channel=cron
[harness] requester key contains ":cron:" = true
[harness] isIsolatedCronRequester = true
[harness] resolvePingPongTurns(default) = 5
[harness] maxPingPongTurns passed to runSessionsSendA2AFlow = 0
[result] agent steps executed (ordered):
           - command.ingress(announce) sessionKey=agent:main:main
[result] REQUESTER (agent:youtube:cron:bf05dd0f:run:b41f8df6) stepped 0 time(s)  <-- injection signal
[result] TARGET    (agent:main:main) stepped 1 time(s)
[verdict] requester-injection (feedback loop) = NO (skipped)

LEG B — normal requester, regression guard (discord:group:req, channel discord):
command: node --import tsx /tmp/proof-92257-narrow/harness.mjs /tmp/openclaw-impl-92257 "discord:group:req" discord

[harness] mode=branch repo=openclaw-impl-92257
[harness] requester=discord:group:req channel=discord
[harness] requester key contains ":cron:" = false
[harness] isIsolatedCronRequester = false
[harness] resolvePingPongTurns(default) = 5
[harness] maxPingPongTurns passed to runSessionsSendA2AFlow = 5
[result] agent steps executed (ordered):
           - gateway.agent sessionKey=discord:group:req
           - command.ingress(announce) sessionKey=agent:main:main
[result] REQUESTER (discord:group:req) stepped 1 time(s)  <-- injection signal
[result] TARGET    (agent:main:main) stepped 1 time(s)
[verdict] requester-injection (feedback loop) = YES (ping-pong ran)

Observed result after fix: LEG A — the isolated-cron requester agent:youtube:cron:bf05dd0f:run:b41f8df6 receives 0 agent steps (the injection loop is skipped because the :cron: marker forces the turn count to 0), while the target agent:main:main still receives its single announce step. LEG B — the normal requester discord:group:req still receives 1 agent step (the configured turn count stays at 5 because the :cron: gate does not fire), and the target is still announced. The two legs together show the fix is narrow: it stops the requester injection for an isolated cron run only, and leaves the normal cross-session roundtrip untouched.
What was not tested: A full end-to-end run over a real external channel (Slack/Discord/Telegram) was not exercised — no live channel credentials are available, so the final announce transport hop was driven through the in-process gateway send double rather than a real network transport. The requester-injection behavior that [Bug]: sessions_send with announce delivery injects Argus reply context into isolated cron session, causing agent feedback loop #92257 fixes (the agent step into the requester session) is fully exercised at runtime above.
Proof limitations or environment constraints: Modules were loaded from TypeScript source via the tsx loader rather than a prebuilt dist; the source is what the build emits and runs in a real node v22.19.0 process. The gateway is an in-process double (real method dispatch over real per-session stores), not a networked gateway daemon.

Before evidence (optional but encouraged): Same isolated-cron requester with the unfixed turn count, captured via the harness's origin-main mode (node --import tsx /tmp/proof-92257-narrow/harness.mjs /tmp/openclaw-impl-92257 "agent:youtube:cron:bf05dd0f:run:b41f8df6" cron origin-main):

[harness] mode=origin-main repo=openclaw-impl-92257
[harness] requester=agent:youtube:cron:bf05dd0f:run:b41f8df6 channel=cron
[harness] isIsolatedCronRequester = true
[harness] maxPingPongTurns passed to runSessionsSendA2AFlow = 5
[result] agent steps executed (ordered):
           - gateway.agent sessionKey=agent:youtube:cron:bf05dd0f:run:b41f8df6
           - command.ingress(announce) sessionKey=agent:main:main
[result] REQUESTER (agent:youtube:cron:bf05dd0f:run:b41f8df6) stepped 1 time(s)  <-- injection signal
[result] TARGET    (agent:main:main) stepped 1 time(s)
[verdict] requester-injection (feedback loop) = YES (ping-pong ran)

Tests and validation

node scripts/run-vitest.mjs src/agents/openclaw-tools.sessions.test.ts src/agents/tools/sessions.test.ts src/agents/tools/sessions-send-tool.a2a.test.ts → 3 files, 91 cases green.
pnpm check:test-types (tsgo over the touched test files) → clean.
Regression coverage added: (1) an isolated-cron requester (:cron: key / cron channel) forwards maxPingPongTurns: 0 and the requester is never stepped while the target is still announced; (2) a normal discord:group:req requester keeps the configured maxPingPongTurns (not 0), pinning that the narrowing does not over-reach; (3) the pre-existing openclaw-tools.sessions.test.ts "supports fire-and-forget and wait" test (which asserts the normal-requester ping-pong) stays green.
What failed before: an earlier broad version of this fix gated on timeoutSeconds === 0 and made the normal-requester fire-and-forget skip its ping-pong, failing openclaw-tools.sessions.test.ts; narrowing the gate to isolated-cron requesters resolves that.

Risk checklist

Did user-visible behavior change? Yes — an isolated cron sessions_send announce no longer injects the target's reply back into the cron run. Normal requester behavior is unchanged.
Did config, environment, or migration behavior change? No.
Did security, auth, secrets, network, or tool execution behavior change? No.
Highest-risk area: the isIsolatedCronRequester detection could in principle be too broad/narrow. Mitigated: it reuses the established :cron:-key / cron-channel signal (src/agents/subagent-registry.ts, src/cron/isolated-agent/run.ts), and the normal-requester regression test + the two-leg runtime proof both confirm normal requesters are untouched.

Current review state

Next action: maintainer review.
Waiting on: nothing from author. Will rebase against fix(agents): include requester identity in sessions_send context (Fixes #38570) #91271 if it lands first (textual overlap on sessions-send-tool.ts).
Bot/reviewer comments addressed: none yet.

clawsweeper · 2026-06-11T23:12:16Z

Codex review: needs maintainer review before merge. Reviewed July 1, 2026, 8:25 PM ET / 00:25 UTC.

Summary
This PR adds a canonical isolated-cron requester gate in sessions_send, passes zero A2A ping-pong turns for those requester sessions, captures the needed fire-and-forget baseline, and adds regression coverage.

PR surface: Source +18, Tests +211. Total +229 across 3 files.

Reproducibility: yes. at source level. Current main and v2026.6.11 pass positive A2A turns into a callee that can step the requester session, and the linked bug plus PR proof show the isolated-cron feedback-loop path.

Review metrics: 1 noteworthy metric.

A2A requester delivery gate: 1 conditional gate changed. The user-visible delivery behavior depends on the classifier that decides whether requester ping-pong runs for cron-run sessions.

Stored data model
Persistent data-model change detected: serialized state: src/agents/tools/sessions-send-tool.a2a.test.ts, serialized state: src/agents/tools/sessions.test.ts, unknown-data-model-change: src/agents/tools/sessions-send-tool.a2a.test.ts, unknown-data-model-change: src/agents/tools/sessions-send-tool.ts, unknown-data-model-change: src/agents/tools/sessions.test.ts. Confirm migration or upgrade compatibility proof before merge.

Root-cause cluster
Relationship: fixed_by_candidate
Canonical: #92257
Summary: This PR is the current open candidate fix for the isolated-cron sessions_send requester-injection bug; the earlier overlapping PR is closed unmerged, while nearby A2A work is adjacent or partial.

Members:

canonical: [Bug]: sessions_send with announce delivery injects Argus reply context into isolated cron session, causing agent feedback loop #92257 - The issue reports the isolated-cron sessions_send feedback loop and is closed by this PR's body.
superseded: fix(agents): skip A2A announce flow for isolated cron sessions_send #93826 - This closed unmerged sibling targeted the same bug with a smaller overlapping approach and is not a viable replacement landing path.
adjacent_distinct: fix(agents): include requester identity in sessions_send context (Fixes #38570) #91271 - This open PR touches nearby A2A prompt context for requester identity, not isolated-cron requester injection.
partial_overlap: feat(agents): add sessions_send handoff acknowledgements #83094 - This draft overlaps sessions_send A2A files as a broader handoff acknowledgement feature, not the specific feedback-loop fix.
adjacent_distinct: fix(agents): route sessions_send A2A announce replies back to request… #76490 - This closed unmerged PR addressed general A2A reply routing and is not a safe canonical fix for the cron-specific path.
1 more in the report.

Proposal only: this assessment does not dispatch repair, suppress jobs, mutate sibling items, close, or merge anything.

Merge readiness
Overall: 🦞 diamond lobster
Proof: 🦞 diamond lobster
Patch quality: 🦞 diamond lobster
Result: ready for maintainer review.

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

Mantis proof suggestion
A live isolated-cron transport run would materially reduce the remaining message-delivery uncertainty before merge. A maintainer can ask Mantis to capture proof by posting this exact PR comment:

@openclaw-mantis telegram live: verify an isolated cron sessions_send announces to the target without stepping the cron requester while a normal requester still gets A2A ping-pong.

Risk before merge

[P1] The merge intentionally changes message-delivery behavior for canonical isolated cron requester sessions; an incorrect classifier would either leave the feedback loop in place or suppress legitimate non-cron A2A ping-pong.
[P1] The supplied proof exercises the runtime path with copied Node output, but it is not a full live external-channel cron run.

Maintainer options:

Accept The Narrow Cron-Run Gate (recommended)
Accept the intentional delivery change because the PR gates only canonical cron-run requester sessions and preserves normal requester ping-pong in tests and proof.
Ask For Live Cron Transport Proof
Pause merge until a real isolated-cron send over an external channel shows the target announce arrives without stepping the cron requester.
Pause For A Different Delivery Contract
Hold or close the branch only if maintainers decide isolated cron requester ping-pong should remain enabled or choose a different canonical fix.

Next step before merge

No automated repair is needed; maintainers need to accept and land or reject this intentional delivery behavior change after normal gates.

Security
Cleared: The diff changes in-repo agent control flow and tests only, with no dependency, workflow, secret, permission, package, or third-party code-execution surface.

Review details

Best possible solution:

Land this canonical cron-run gate after normal required gates, then close the linked bug; request live cron transport proof only if maintainers want extra assurance.

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

Yes, at source level. Current main and v2026.6.11 pass positive A2A turns into a callee that can step the requester session, and the linked bug plus PR proof show the isolated-cron feedback-loop path.

Is this the best way to solve the issue?

Yes. The caller-side canonical cron-run gate is the narrowest maintainable fix I found because it preserves target announce delivery and normal requester ping-pong, unlike a timeout-based or whole-flow skip.

AGENTS.md: found and applied where relevant.

Codex review notes: model internal, reasoning high; reviewed against 3558391a75fd.

Label changes

Label changes:

add mantis: telegram-visible-proof: Mantis should capture Telegram visible proof. The reported path is Telegram-visible message delivery from isolated cron, so a short Telegram proof could directly show the user-visible behavior.

Label justifications:

P2: The PR addresses a concrete isolated-cron A2A behavior bug with bounded but real session and message-delivery impact.
merge-risk: 🚨 message-delivery: The diff intentionally changes whether sessions_send feeds target replies back into canonical isolated cron requester sessions.
rating: 🦞 diamond lobster: Overall readiness is 🦞 diamond lobster; proof is 🦞 diamond lobster and patch quality is 🦞 diamond lobster.
status: 👀 ready for maintainer look: ClawSweeper has no concrete contributor-facing blocker left for this PR. Sufficient (live_output): The PR body supplies copied Node before/after output showing the isolated-cron requester is not stepped after the fix while normal requester ping-pong remains; full external-channel cron proof was not supplied.
proof: sufficient: Contributor real behavior proof is sufficient. The PR body supplies copied Node before/after output showing the isolated-cron requester is not stepped after the fix while normal requester ping-pong remains; full external-channel cron proof was not supplied.
mantis: telegram-visible-proof: Mantis should capture Telegram visible proof. The reported path is Telegram-visible message delivery from isolated cron, so a short Telegram proof could directly show the user-visible behavior.

Evidence reviewed

PR surface:

Source +18, Tests +211. Total +229 across 3 files.

View PR surface stats

Area	Files	Added	Removed	Net
Source	1	21	3	+18
Tests	2	212	1	+211
Docs	0	0	0	0
Config	0	0	0	0
Generated	0	0	0	0
Other	0	0	0	0
Total	3	233	4	+229

What I checked:

Repository policy read: Root and scoped agent/tool AGENTS guidance was read and applied; the review used the required read-beyond-diff and message-delivery risk checks. (AGENTS.md:24, 3558391a75fd)
Current main still has requester ping-pong enabled: Current main passes the configured maxPingPongTurns unchanged into runSessionsSendA2AFlow; there is no isolated-cron requester suppression on this caller path. (src/agents/tools/sessions-send-tool.ts:656, 3558391a75fd)
A2A callee can step the requester: When maxPingPongTurns > 0 and requester differs from target, the A2A flow starts with currentSessionKey = requesterSessionKey and calls runAgentStep, matching the reported feedback-loop mechanism. (src/agents/tools/sessions-send-tool.a2a.ts:145, 3558391a75fd)
Latest release also lacks the gate: Tag v2026.6.11 points at e085fa1a3ffd32d0ea6917e1e6fb4ecbffbb77d2; that release still passes maxPingPongTurns unchanged and has the same requester-stepping A2A callee. (src/agents/tools/sessions-send-tool.ts:656, e085fa1a3ffd)
PR head uses the canonical cron-run classifier: The final PR head computes isIsolatedCronRequester with isCronRunSessionKey, captures a baseline for that path, and passes zero ping-pong turns only for that requester class. (src/agents/tools/sessions-send-tool.ts:564, 4dbd299a1e46)
Cron-run key shape matches the fix boundary: Cron execution passes runSessionKey into agent execution, and OpenClaw tool construction passes the active session key into createSessionsSendTool; the classifier accepts agent:<id>:cron:<job>:run:<run> keys and rejects cron-like normal channel keys. (src/sessions/session-key-utils.ts:284, 3558391a75fd)

Likely related people:

vincentkoc: Live PR data shows Vincent Koc assigned to this review, and release/current history includes recent sessions and cron-key-adjacent commits on the same surface. (role: assigned reviewer and recent area contributor; confidence: medium; commits: e085fa1a3ffd; files: src/agents/tools/sessions-send-tool.ts, src/agents/tools/sessions-send-tool.a2a.ts, src/sessions/session-key-utils.ts)
steipete: History search for maxPingPongTurns and the A2A split points to earlier agent tool refactors that shaped the current sessions_send A2A files. (role: feature-history contributor; confidence: medium; commits: bcbfb357bec7, 13c2f22240e4; files: src/agents/tools/sessions-send-tool.ts, src/agents/tools/sessions-send-tool.a2a.ts)
gumadeiras: The shared cron run session-key classifier used by the PR traces to the session maintenance and cron pruning work in commit e19a23520c2d. (role: cron session-key classifier contributor; confidence: medium; commits: e19a23520c2d; files: src/sessions/session-key-utils.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.

harjothkhara · 2026-06-20T05:19:25Z

Addressed the Codex [P2] finding (narrow the cron requester detector).

The requester gate now uses the canonical isCronRunSessionKey instead of a raw :cron: substring, so a non-canonical cron-like key such as agent:main:slack:cron:job:run:uuid keeps its intended cross-session ping-pong.
Kept the requesterChannel === "cron" arm — isolated cron runs do set that channel (src/cron/isolated-agent/run.ts:1104,1422), so it is a proven path, not dead code.
Added a regression in src/agents/tools/sessions.test.ts proving the non-canonical key preserves the configured maxPingPongTurns.
Rebased onto latest main.

Local proof: node scripts/run-vitest.mjs src/agents/tools/sessions.test.ts src/agents/tools/sessions-send-tool.a2a.test.ts → 62 passed. Reverting the one-line detector change makes the new regression fail (forces 0, expected 5), confirming it guards the fix.

@clawsweeper re-review

clawsweeper · 2026-06-20T05:19:27Z

🦞🧹
ClawSweeper re-review requested.

I asked ClawSweeper to review this item again.
Action: item re-review queued (workflow sweep.yml, event repository_dispatch).
Result: the existing ClawSweeper review comment will be edited in place when the review finishes.

…penclaw#92257) Address Codex [P2] review: replace the raw `:cron:` substring detector with the canonical isCronRunSessionKey so a non-canonical cron-like requester key (e.g. agent:main:slack:cron:job:run:uuid) keeps its intended cross-session ping-pong. Keep the requesterChannel === "cron" arm (isolated cron runs set that channel in src/cron/isolated-agent/run.ts). Add a regression covering the non-canonical key.

…ey only (openclaw#92257) The requesterChannel === "cron" arm was unreachable: agentChannel is always a DeliverableMessageChannel from resolveGatewayMessageChannel(messageProvider), never the literal "cron". The channel: "cron" in src/cron/isolated-agent/run.ts labels diagnostics events/lifecycle, not the tool channel. Gate on isCronRunSessionKey alone and fix the misleading comment. Tests drove the dead arm via agentChannel/requesterChannel "cron" plus a non-canonical key (agent:main:cron:run:abc, which isCronRunSessionKey rejects). Switch them to a canonical cron-run key (agent:main:cron:job:run:abc) and a normal delivery channel so they exercise the real production gate.

vincentkoc · 2026-07-02T00:34:31Z

Merged via squash.

Prepared head SHA: 9cda03ce94a50c58fbc064b97e061d9d0c5de5ef
Landed commit: c49714f7b6d2a6690ccd03363a568c6e0a4b68fc

* origin/main: test(gateway): isolate live release agent state test(plugins): repair prerelease validation fixtures fix(discord): guard JSON.parse against malformed API response bodies (#97889) fix(codex): preserve app approvals in side forks (#98812) fix(agents): don't inject A2A turns into isolated-cron sessions_send (#92257) (#92283) fix(browser): guard setDeep against empty keys array (#98138) docs(telegram): move maintainer decisions into scoped AGENTS.md fix(ios): generate light app store screenshots only fix(gateway): distinguish reachable gateway from failed status probe (#98183)

…penclaw#92257) (openclaw#92283) * fix(agents): don't inject A2A turns into isolated-cron sessions_send (openclaw#92257) Fire-and-forget sessions_send (timeoutSeconds === 0) with announce delivery runs the A2A ping-pong loop. For a cross-session send (requester != target) the loop's first iteration feeds the target agent's reply back into the requester session as a new turn. For a normal requester that roundtrip is intended, but for an *isolated cron* requester it injects reply context into the isolated run and causes an agent feedback loop. Narrow the fix to isolated-cron requesters only (detected by a session key containing ":cron:" or channel "cron" -- the same signal used by src/agents/subagent-registry.ts and set in src/cron/isolated-agent/run.ts), NOT by timeoutSeconds. Gating on timeoutSeconds was too broad: it disabled the intended ping-pong for normal cross-session fire-and-forget sends. Two hunks in src/agents/tools/sessions-send-tool.ts, both reusing one `isIsolatedCronRequester` gate: 1. Force maxPingPongTurns to 0 in the runSessionsSendA2AFlow invocation only for an isolated-cron requester. The a2a flow's ping-pong guard (`maxPingPongTurns > 0`) then skips the requester-injection loop and proceeds straight to the announce step in the TARGET session, preserving fire-and-forget announce delivery. Normal requesters keep the configured turn count. 2. Read baselineReply for fire-and-forget sends when same-session (prior behavior) OR isolated-cron. Without a baseline fingerprint, a2a.ts would treat pre-existing assistant text in the target session (e.g. an unrelated concurrent cron's output) as "the reply" and misattribute it. The normal cross-session fire leg's history-call count is unchanged from origin/main. Read failures are tolerated so a snapshot error never blocks accepting the send. Tests (sessions.test.ts): an isolated-cron cross-session fire-and-forget forwards maxPingPongTurns: 0; a normal-requester regression guard (discord:group:req / channel discord) forwards the configured turn count (not 0), proving the normal ping-pong is preserved. a2a flow tests (sessions-send-tool.a2a.test.ts): with turns=0 + requester!=target the requester is never stepped but the target is still announced, and a baseline-matching reply is neither injected nor announced. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> * fix(agents): gate isolated-cron A2A on canonical cron-run classifier (openclaw#92257) Address Codex [P2] review: replace the raw `:cron:` substring detector with the canonical isCronRunSessionKey so a non-canonical cron-like requester key (e.g. agent:main:slack:cron:job:run:uuid) keeps its intended cross-session ping-pong. Keep the requesterChannel === "cron" arm (isolated cron runs set that channel in src/cron/isolated-agent/run.ts). Add a regression covering the non-canonical key. * refactor(agents): drop dead cron-channel A2A arm, gate on canonical key only (openclaw#92257) The requesterChannel === "cron" arm was unreachable: agentChannel is always a DeliverableMessageChannel from resolveGatewayMessageChannel(messageProvider), never the literal "cron". The channel: "cron" in src/cron/isolated-agent/run.ts labels diagnostics events/lifecycle, not the tool channel. Gate on isCronRunSessionKey alone and fix the misleading comment. Tests drove the dead arm via agentChannel/requesterChannel "cron" plus a non-canonical key (agent:main:cron:run:abc, which isCronRunSessionKey rejects). Switch them to a canonical cron-run key (agent:main:cron:job:run:abc) and a normal delivery channel so they exercise the real production gate. * fix(agents): align cron A2A fallback baselines * chore: prepare branch refresh --------- Co-authored-by: Claude Fable 5 <noreply@anthropic.com> Co-authored-by: Vincent Koc <vincentkoc@ieee.org>

openclaw-barnacle Bot added agents Agent runtime and tooling size: S triage: needs-real-behavior-proof Candidate: external PR needs after-fix proof from a real setup. labels Jun 11, 2026

clawsweeper Bot added the rating: 🌊 off-meta tidepool PR readiness rating does not apply to this item. label Jun 11, 2026

harjothkhara marked this pull request as draft June 11, 2026 23:27

harjothkhara force-pushed the fix/sessions-send-fire-and-forget-92257 branch from 3da89f0 to a32947d Compare June 11, 2026 23:47

harjothkhara changed the title ~~fix(agents): don't inject A2A turns into fire-and-forget sessions_send (#92257)~~ fix(agents): don't inject A2A turns into isolated-cron sessions_send (#92257) Jun 11, 2026

harjothkhara marked this pull request as ready for review June 11, 2026 23:47

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 Jun 11, 2026

kesslerio mentioned this pull request Jun 12, 2026

[Bug]: isolated cron can deliver unavailable-tool self-debug to Telegram #92535

Open

clawsweeper Bot mentioned this pull request Jun 15, 2026

fix(agents): include requester identity in sessions_send context (Fixes #38570) #91271

Open

LiLan0125 mentioned this pull request Jun 17, 2026

fix(agents): skip A2A announce flow for isolated cron sessions_send #93826

Closed

clawsweeper Bot mentioned this pull request Jun 18, 2026

fix(agents): route sessions_send A2A announce replies back to request… #76490

Closed

harjothkhara force-pushed the fix/sessions-send-fire-and-forget-92257 branch from a32947d to 034a80a Compare June 20, 2026 05:19

openclaw-barnacle Bot added size: M and removed size: S labels Jun 20, 2026

harjothkhara added 2 commits June 22, 2026 13:39

harjothkhara force-pushed the fix/sessions-send-fire-and-forget-92257 branch from 426774a to 1f040bd Compare June 22, 2026 20:43

vincentkoc self-assigned this Jul 2, 2026

vincentkoc added 2 commits July 1, 2026 17:14

Merge branch 'main' into fix/sessions-send-fire-and-forget-92257

4dbd299

fix(agents): align cron A2A fallback baselines

5528dca

openclaw-barnacle Bot added channel: telegram Channel integration: telegram app: ios App: ios labels Jul 2, 2026

vincentkoc added 2 commits July 1, 2026 17:24

chore: prepare branch refresh

3f002f0

Merge branch 'main' into fix/sessions-send-fire-and-forget-92257

9cda03c

openclaw-barnacle Bot added size: S and removed channel: telegram Channel integration: telegram app: ios App: ios size: M labels Jul 2, 2026

clawsweeper Bot added the mantis: telegram-visible-proof Mantis should capture Telegram visible proof. label Jul 2, 2026

clawsweeper Bot temporarily deployed to qa-live-shared July 2, 2026 00:28 Inactive

vincentkoc merged commit c49714f into openclaw:main Jul 2, 2026
113 of 116 checks passed

Haderach-Ram mentioned this pull request Jul 2, 2026

Ecosystem Digest — 2026-07-02 Haderach-Ram/openclaw-radar#67

Open

This was referenced Jul 4, 2026

fix: A2A handoffs can duplicate message-tool replies #97259

Open

fix(codex): resolve dynamic tool hard timeout from timeoutSeconds when timeoutMs is absent #100800

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Uh oh!

fix(agents): don't inject A2A turns into isolated-cron sessions_send (#92257)#92283

fix(agents): don't inject A2A turns into isolated-cron sessions_send (#92257)#92283
vincentkoc merged 7 commits into
openclaw:mainfrom
harjothkhara:fix/sessions-send-fire-and-forget-92257

harjothkhara commented Jun 11, 2026 •

edited

Loading

Uh oh!

clawsweeper Bot commented Jun 11, 2026 •

edited

Loading

Uh oh!

harjothkhara commented Jun 20, 2026

Uh oh!

clawsweeper Bot commented Jun 20, 2026 •

edited

Loading

Uh oh!

Uh oh!

vincentkoc commented Jul 2, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Uh oh!

Uh oh!

Conversation

harjothkhara commented Jun 11, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Linked context

Real behavior proof (required for external PRs)

Tests and validation

Risk checklist

Current review state

Uh oh!

clawsweeper Bot commented Jun 11, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

harjothkhara commented Jun 20, 2026

Uh oh!

clawsweeper Bot commented Jun 20, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Uh oh!

vincentkoc commented Jul 2, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

harjothkhara commented Jun 11, 2026 •

edited

Loading

clawsweeper Bot commented Jun 11, 2026 •

edited

Loading

clawsweeper Bot commented Jun 20, 2026 •

edited

Loading