fix(compaction): skip replay-unsafe session sends by mushuiyu886 · Pull Request #93293 · openclaw/openclaw

mushuiyu886 · 2026-06-15T13:06:45Z

Summary

Filters only the completed active target-side sessions_send tail from compaction branch fallback. Historical turns and unfinished turns ending in tool calls/results remain recoverable.
Converts source-side sessions_send calls into inert summary text, preserving call IDs, arguments, returned results, failures, empty results, and missing-result state without leaving executable tool-call blocks to replay.
Recognizes provider-prefixed tool names and keeps ordinary compaction fallback unchanged.

Root cause

When normal compaction preparation contained no real conversation, the safeguard reused the session branch verbatim. That could replay an already-delivered target-side inter-session turn or an already-executed source-side sessions_send call. The previous candidate filter was too broad: it also removed historical and unfinished work.

The repaired patch stays at the compaction fallback owner boundary. It drops a target-side turn only when the original branch ends in terminal assistant text. Source-side send calls/results become inert evidence instead of executable tool history.

Compatibility and scope

No public API, config, schema, protocol, persistence, provider-routing, or delivery-execution changes.
No normal compaction behavior change when branch fallback is not needed.
Root CHANGELOG.md remains release-owned; this PR body carries release-note context.

Merge risk

merge-risk: session-state: over-filtering could lose recovery context.
merge-risk: message-delivery: under-filtering could replay a side effect.

The focused tests cover completed target turns, historical target turns, ordinary and nested unfinished tool work, resolved/failed/missing source-side sends, namespaced tool names, and normal fallback preservation.

Proof

Rebased current head: e9dafd53382c7abc45d285dddf98d882abf85f33

Blacksmith Testbox run: https://github.com/openclaw/openclaw/actions/runs/29480862149

oxfmt --check (2 changed files): passed
plugin-sdk boundary DTS and targeted oxlint: passed
src/agents/agent-hooks/compaction-safeguard.test.ts: 111 passed
git diff --check: passed
fresh aggregate autoreview: no accepted/actionable findings

Behavior verification remains source-level and focused rather than a live timing capture across two running agents. The linked intermittent path is therefore partially, not fully, reproduced end to end.

Contributor credit

Original fix and regression direction by @mushuiyu886. Maintainer repair commits retain Co-authored-by: 杨浩宇0668001029 <[email protected]>.

clawsweeper · 2026-06-15T13:09:07Z

Codex review: needs maintainer review before merge. Reviewed July 6, 2026, 10:30 PM ET / 02:30 UTC.

Summary
The branch filters replay-unsafe sessions_send and messaging-send artifacts from compaction-safeguard branch fallback while preserving safe assistant text, with regression coverage for target-side, source-side, and mixed-content fallback cases.

PR surface: Source +112, Tests +237. Total +349 across 2 files.

Reproducibility: yes. at the source level, but not as a live timing repro in this review. Current main still falls back to unfiltered branch messages, sessions_send records matching inter-session provenance, and the PR adds targeted regressions for the unsafe replay shapes.

Review metrics: 1 noteworthy metric.

Fallback replay cases covered: 3 regression scenarios added. The tests cover target-side inter-session turns, source-side send tool artifacts, and mixed assistant text preservation at the changed fallback boundary.

Root-cause cluster
Relationship: fixed_by_candidate
Canonical: #84139
Summary: This PR is the focused candidate fix for the canonical compaction-safeguard sessions_send fallback replay issue; nearby A2A duplicate-message PRs are adjacent but do not supersede this branch.

Members:

canonical: [2026.5.18] Compaction safeguard causes duplicate messages on sessions_send interactions #84139 - The PR uses a closing reference for this issue, and the issue describes the exact compaction fallback replay mode touched by the branch.
adjacent_distinct: fix: clear cross-run messaging-tool sent-text state after every compaction #80285 - That PR concerns stale cross-run messaging-tool sent-text state after compaction, not branch-fallback replay of sessions_send artifacts.
adjacent_distinct: fix: prevent A2A target from reverse-calling sessions_send to requester (#39476) #91161 - That PR targets reverse-calling sessions_send during A2A target turns for a separate duplicate-delivery path.
adjacent_distinct: fix: A2A handoffs can duplicate message-tool replies #97259 - That PR explicitly says it does not change the compaction replay behavior covered by this PR.
adjacent_distinct: fix(agents): keep post-compaction user re-issue of a kept-tail prompt during compaction rotation #94328 - That merged PR fixes post-compaction prompt rotation loss rather than replay-unsafe sessions_send branch fallback.
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.

Risk before merge

[P2] The change intentionally alters which recovered branch entries can become compaction summary input; under-filtering could replay already-executed send artifacts, while over-filtering could lose safe fallback context.
[P1] The PR head is behind current main, so maintainers should refresh/recheck the merge result before landing even though the stale-base drift alone is not a patch defect.
[P1] The contributor proof exercises the production hook path with terminal output and tests, but it does not include a full live two-agent timing capture of the intermittent duplicate-delivery scenario.

Maintainer options:

Refresh and land the focused filter (recommended)
After rebasing or otherwise proving the exact merge head, proceed with the compaction-boundary filter because the source path and regression tests cover the reported replay mode.
Require a live timing capture
Maintainers can ask for a full two-agent compaction timing proof before merge if terminal production-hook proof is not enough for this risk level.
Pause for a broader A2A design
Pause this PR only if maintainers decide the fallback filter should be replaced by a broader delivery/API ownership change instead of a narrow compaction repair.

Next step before merge

No ClawSweeper repair job is needed because no concrete patch defect was found; the remaining action is maintainer review, stale-base refresh, and normal landing gates.

Security
Cleared: The diff touches only agent compaction runtime logic and colocated tests; I found no concrete security or supply-chain regression.

Review details

Best possible solution:

Land this focused compaction fallback filter after refreshing the merge result and exact-head gates, while leaving the adjacent A2A reverse-send and message-tool mirror duplicate paths to their separate PRs.

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

Yes at the source level, but not as a live timing repro in this review. Current main still falls back to unfiltered branch messages, sessions_send records matching inter-session provenance, and the PR adds targeted regressions for the unsafe replay shapes.

Is this the best way to solve the issue?

Yes. The compaction safeguard fallback is the narrow owner boundary for this bug because it is the point where current main reuses recovered branch messages for summarization; the related A2A API/delivery PRs address adjacent duplicate modes rather than this fallback replay.

AGENTS.md: found and applied where relevant.

Codex review notes: model internal, reasoning high; reviewed against b2cecd8ece47.

Label changes

Label changes:

add proof: sufficient: Contributor real behavior proof is sufficient. The PR body includes after-fix terminal proof over the production compaction safeguard hook path plus focused tests, lint, and diff checks; a live two-agent timing capture would be stronger but is not required for this nonvisual runtime path.
add rating: 🦞 diamond lobster: Overall readiness is 🦞 diamond lobster; proof is 🦞 diamond lobster and patch quality is 🦞 diamond lobster.
add status: 👀 ready for maintainer look: ClawSweeper has no concrete contributor-facing blocker left for this PR. Sufficient (terminal): The PR body includes after-fix terminal proof over the production compaction safeguard hook path plus focused tests, lint, and diff checks; a live two-agent timing capture would be stronger but is not required for this nonvisual runtime path.

Label justifications:

P2: This is a focused fix for an intermittent duplicate-delivery/session-compaction bug with real but limited blast radius.
merge-risk: 🚨 session-state: The PR changes recovered session-branch content used for compaction summaries when normal preparation omitted real conversation content.
merge-risk: 🚨 message-delivery: The same boundary must avoid replaying already-executed sessions_send artifacts without suppressing safe conversation context.
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 (terminal): The PR body includes after-fix terminal proof over the production compaction safeguard hook path plus focused tests, lint, and diff checks; a live two-agent timing capture would be stronger but is not required for this nonvisual runtime path.
proof: sufficient: Contributor real behavior proof is sufficient. The PR body includes after-fix terminal proof over the production compaction safeguard hook path plus focused tests, lint, and diff checks; a live two-agent timing capture would be stronger but is not required for this nonvisual runtime path.

Evidence reviewed

PR surface:

Source +112, Tests +237. Total +349 across 2 files.

View PR surface stats

Area	Files	Added	Removed	Net
Source	1	114	2	+112
Tests	1	237	0	+237
Docs	0	0	0	0
Config	0	0	0	0
Generated	0	0	0	0
Other	0	0	0	0
Total	2	351	2	+349

What I checked:

Root and scoped policy read: Read root AGENTS.md, src/agents/AGENTS.md, the PR template, issue template config, bug form, and CODEOWNERS; the review-depth, agent-test-performance, and PR evidence rules were applicable. (AGENTS.md:1, b2cecd8ece47)
Current main still has the unfiltered fallback: Current main falls back to collectSessionBranchMessages(ctx.sessionManager) when preparation omitted real conversation content, so the linked issue is still relevant and not implemented on main. (src/agents/agent-hooks/compaction-safeguard.ts:907, b2cecd8ece47)
PR filters at the implicated owner boundary: The PR applies filterReplayUnsafeSessionBranchMessages exactly where the compaction safeguard switches to recovered branch messages before summarization. (src/agents/agent-hooks/compaction-safeguard.ts:1020, 69fa892b99fa)
sessions_send provenance matches the filter: sessions_send marks nested input as kind: "inter_session" with sourceTool: "sessions_send", matching the target-side replay filter condition. (src/agents/tools/sessions-send-tool.ts:601, b2cecd8ece47)
Messaging sends are side-effecting: isMessagingToolSendAction treats every sessions_send invocation as a send action, supporting the choice to remove those replay artifacts from fallback history. (src/agents/embedded-agent-messaging.ts:70, b2cecd8ece47)
Regression coverage hits both replay and preservation cases: The PR adds tests for target-side inter-session branch turns, source-side sessions_send tool calls/results, and mixed assistant text where safe text is preserved while send artifacts are removed. (src/agents/agent-hooks/compaction-safeguard.test.ts:2466, 69fa892b99fa)

Likely related people:

steipete: Peter Steinberger authored the inter-session input provenance work and much of the earlier sessions_send delivery semantics that this filter relies on. (role: feature owner / adjacent owner; confidence: high; commits: 85409e401b65, a25922a21f86; files: src/agents/tools/sessions-send-tool.ts, src/sessions/input-provenance.ts)
vincentkoc: Vincent Koc recently touched sessions_send explicit keys, delivery evidence, and nearby agent hook/runtime surfaces. (role: recent area contributor; confidence: high; commits: 3659ff8bbf71, 0ea08076c3b5, 6cb631afc9c5; files: src/agents/tools/sessions-send-tool.ts, src/agents/embedded-agent-messaging.ts, src/agents/agent-hooks/compaction-safeguard.ts)
Takhoffman: Tak Hoffman authored adjacent sessions_send stale-reply carryover work in the same delivery-fallback problem family. (role: adjacent sessions_send contributor; confidence: medium; commits: 5b4669632a9d; files: src/agents/tools/sessions-send-tool.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.

Review history (2 earlier review cycles)

reviewed 2026-06-29T08:18:17.433Z sha 69fa892 :: needs maintainer review before merge. :: none
reviewed 2026-07-07T01:48:55.680Z sha 69fa892 :: needs maintainer review before merge. :: none

mushuiyu886 · 2026-06-15T14:08:52Z

@clawsweeper re-review

clawsweeper · 2026-06-15T14:08:56Z

🦞🧹
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.

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/27552225974
Updated: 2026-06-15T14:17:34.996Z

Co-authored-by: 杨浩宇0668001029 <[email protected]>

steipete · 2026-07-16T08:46:17Z

Merged via squash.

Prepared head SHA: e9dafd53382c7abc45d285dddf98d882abf85f33
Landed commit: f771aa54e8fd6eafbee9dcdd185f1455f8e5182c

* fix(compaction): skip replay-unsafe session sends * fix(compaction): preserve safe text around session sends * fix(compaction): narrow replay filtering to sessions_send Co-authored-by: 杨浩宇0668001029 <[email protected]> * test(compaction): use review-safe key fixtures Co-authored-by: 杨浩宇0668001029 <[email protected]> * test(compaction): use placeholder API key Co-authored-by: 杨浩宇0668001029 <[email protected]> * test(compaction): limit fixture cleanup to new cases Co-authored-by: 杨浩宇0668001029 <[email protected]> * fix(compaction): preserve completed sessions_send history Co-authored-by: 杨浩宇0668001029 <[email protected]> * fix(compaction): preserve unfinished delegated work Co-authored-by: 杨浩宇0668001029 <[email protected]> * fix(compaction): detect completion before replay filtering Co-authored-by: 杨浩宇0668001029 <[email protected]> * fix(compaction): retain unresolved send status Co-authored-by: 杨浩宇0668001029 <[email protected]> * fix(compaction): preserve inert send outcomes Co-authored-by: 杨浩宇0668001029 <[email protected]> * fix(compaction): narrow terminal tail type Co-authored-by: 杨浩宇0668001029 <[email protected]> * fix(compaction): guard replay input lookup Co-authored-by: 杨浩宇0668001029 <[email protected]> --------- Co-authored-by: Peter Steinberger <[email protected]>

openclaw-barnacle Bot added agents Agent runtime and tooling size: M proof: supplied External PR includes structured after-fix real behavior proof. labels Jun 15, 2026

github-actions Bot mentioned this pull request Jun 16, 2026

🦞 OpenClaw 生态日报 2026-06-16 96loveslife/big_model_radar#7

Open

openclaw-barnacle Bot removed the proof: sufficient ClawSweeper judged the real behavior proof convincing. label Jun 18, 2026

mushuiyu886 force-pushed the feat/issue-84139 branch from 3b641e9 to 9ca43f2 Compare June 18, 2026 14:14

This was referenced Jul 1, 2026

fix: prevent A2A target from reverse-calling sessions_send to requester (#39476) #91161

Closed

A2A sessions_send: target agent can call sessions_send back, causing duplicate messages #39476

Open

steipete self-assigned this Jul 16, 2026

mushuiyu886 and others added 13 commits July 16, 2026 00:51

fix(compaction): skip replay-unsafe session sends

a7aa3ca

fix(compaction): preserve safe text around session sends

1f601ad

fix(compaction): narrow replay filtering to sessions_send

853b545

Co-authored-by: 杨浩宇0668001029 <[email protected]>

test(compaction): use review-safe key fixtures

0ba5137

Co-authored-by: 杨浩宇0668001029 <[email protected]>

test(compaction): use placeholder API key

94c07c0

Co-authored-by: 杨浩宇0668001029 <[email protected]>

test(compaction): limit fixture cleanup to new cases

d7b779a

Co-authored-by: 杨浩宇0668001029 <[email protected]>

fix(compaction): preserve completed sessions_send history

9624979

Co-authored-by: 杨浩宇0668001029 <[email protected]>

fix(compaction): preserve unfinished delegated work

8e6f366

Co-authored-by: 杨浩宇0668001029 <[email protected]>

fix(compaction): detect completion before replay filtering

ce087a5

Co-authored-by: 杨浩宇0668001029 <[email protected]>

fix(compaction): retain unresolved send status

8198d8f

Co-authored-by: 杨浩宇0668001029 <[email protected]>

fix(compaction): preserve inert send outcomes

28c8532

Co-authored-by: 杨浩宇0668001029 <[email protected]>

fix(compaction): narrow terminal tail type

66d82b3

Co-authored-by: 杨浩宇0668001029 <[email protected]>

fix(compaction): guard replay input lookup

e9dafd5

Co-authored-by: 杨浩宇0668001029 <[email protected]>

steipete force-pushed the feat/issue-84139 branch from 69fa892 to e9dafd5 Compare July 16, 2026 07:52

openclaw-barnacle Bot added size: L triage: needs-pr-context Candidate: external PR body lacks required problem context or evidence. and removed size: M labels Jul 16, 2026

steipete merged commit f771aa5 into openclaw:main Jul 16, 2026
134 of 140 checks passed

github-actions Bot mentioned this pull request Jul 16, 2026

📡 Upstream Digest — 2026-07-16 10:08 UTC curtismercier/openclaw-mods#1311

Open

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Uh oh!

fix(compaction): skip replay-unsafe session sends#93293

fix(compaction): skip replay-unsafe session sends#93293
steipete merged 13 commits into
openclaw:mainfrom
mushuiyu886:feat/issue-84139

mushuiyu886 commented Jun 15, 2026 •

edited by steipete

Loading

Uh oh!

clawsweeper Bot commented Jun 15, 2026 •

edited

Loading

Uh oh!

mushuiyu886 commented Jun 15, 2026

Uh oh!

clawsweeper Bot commented Jun 15, 2026 •

edited

Loading

Uh oh!

Uh oh!

steipete commented Jul 16, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Uh oh!

Uh oh!

Conversation

mushuiyu886 commented Jun 15, 2026 • edited by steipete Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Root cause

Compatibility and scope

Merge risk

Proof

Contributor credit

Uh oh!

clawsweeper Bot commented Jun 15, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

mushuiyu886 commented Jun 15, 2026

Uh oh!

clawsweeper Bot commented Jun 15, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Uh oh!

steipete commented Jul 16, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

mushuiyu886 commented Jun 15, 2026 •

edited by steipete

Loading

clawsweeper Bot commented Jun 15, 2026 •

edited

Loading

clawsweeper Bot commented Jun 15, 2026 •

edited

Loading