Codex review: needs maintainer review before merge. Reviewed May 29, 2026, 3:10 PM ET / 19:10 UTC.

Summary
The PR sends generated-media completion attachments directly before requester-agent handoff, marks generated-media direct failures terminal, and updates agent delivery/dispatch tests plus one run-wait timing test.

PR surface: Source +21, Tests +52. Total +73 across 5 files.

Reproducibility: yes. The linked issue and PR comments give a concrete shipped-beta Discord reproduction with one image_generate completion producing 12 posts, and current main source still routes active generated-media completions through requester wake before fallback direct delivery; I did not run the live repro locally.

Review metrics: 1 noteworthy metric.

• Fallback semantics changed: 1 generated-media failure path made terminal. This is the maintainer-visible behavior change behind the message-delivery risk: failed direct attempts no longer fall back to requester-agent delivery.

Merge readiness
Overall: 🐚 platinum hermit
Proof: 🦀 challenger crab
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:

• [P2] Have maintainers explicitly accept the direct-send failure tradeoff, or request the artifact-idempotency alternative before merge.

Mantis proof suggestion
A Telegram Desktop run would add useful sibling proof that generated-media completions post once in a topic and do not duplicate through requester wake. A maintainer can ask Mantis to capture proof by posting a new PR comment that starts with the OpenClaw Mantis account mention, followed by:

telegram desktop proof: verify a generated-media completion in a Telegram topic posts one media message and does not duplicate or wake requester message.send.

Risk before merge

• [P2] If direct media upload fails before posting anything, generated-media completion now fails instead of falling back to requester-agent delivery; that prevents partial-send duplicates but can turn a recoverable channel failure into a missed completion message.
• [P1] The generic generated-media completion path affects visible delivery across Discord, Slack, Telegram, and other message channels; the PR has strong Discord live proof and sibling unit coverage, but maintainers may still want transport-specific proof before accepting the delivery tradeoff.

Maintainer options:

1. Accept terminal direct-send semantics
   Merge as-is if maintainers prefer preventing duplicate generated media over requester-agent rescue after any attempted direct media send fails.
2. Preserve fallback with artifact idempotency
   Revise the PR to add an artifact-and-target idempotency guard so confirmed or possible partial sends cannot duplicate while clear pre-send failures may still use requester-agent fallback.
3. Pause for sibling transport proof
   Hold merge until a real Telegram or other sibling transport run proves the same direct-once behavior on the generic path maintainers care about most.

Next step before merge

• [P2] No narrow automated repair remains; maintainers need to accept or reject the message-delivery fallback tradeoff before merge.

Security
Cleared: The diff only changes agent completion delivery logic and tests; it does not touch dependencies, workflows, permissions, secrets, lockfiles, or package execution surfaces.

Reopened because #87989 reproduced again after updating the affected instance to 2026.5.28-beta.1 (84fd629).

New evidence: one image_generate:5e66cfd7-9cff-4edd-86b5-85d5aec8e2fe:ok completion produced 12 successful message.send calls and 12 visible Discord posts for the same generated file, image-1---34799bf3-36d1-462a-8e08-3490a7e58a28.jpg.

The bot closure pointed at the message-tool terminal guard, but the reproduced tool calls all had explicit route keys (channel, target, or both), and current main returns false for hasExplicitMessageRoute(args). That guard is present but bypassed by this delivery path. See #87989 for the full receipt list and source citations.

Live Dicky validation

2026.5.28-beta.1 (84fd629) patched with this PR's runtime change on Dicky:

Patched file: /home/dicky/.npm-global/lib/node_modules/openclaw/dist/subagent-announce-delivery-meOMMrhe.js
Patched sha256: 54be39526423c74139c17b32c4e73c287b83dc9bda7ea8d2f73066289a77f3b0
Backup: /home/dicky/.openclaw/state/backups/openclaw-pr87991-runtime/20260529T110329Z/subagent-announce-delivery-meOMMrhe.js
Gateway: running, RPC probe ok

Retest result:

Prompt: @Dicky make an image of the last hour of this channel
Run: image_generate:26e54be5-8afb-4736-8534-d8c6fb52b902:ok
Media path: ~/.openclaw/media/tool-image-generation/image-1---815ea96c-e360-4b6e-9b0a-5522e5c730cb.jpg
Public Discord posts with that attachment filename: 1
Discord message id: 1509876170701799486
Attachment id: 1509876164879974593
Caption/content: The generated image is ready.
Requester session message tool calls with that image: 0
Requester session message tool results with that image: 0
Delivery mirrors: 1
Delivery mirror idempotency key: image_generate:26e54be5-8afb-4736-8534-d8c6fb52b902:ok:generated-media-direct

This proves the patched path delivered the generated image directly and did not wake the requester session into the old message.send loop.

@clawsweeper please re-review.

Addressed the P2 fallback edge in e6e0569:

• generated-media direct delivery failures are now marked terminal
• completion dispatch now stops on terminal direct failures instead of fallback-steering into the requester-agent path
• added regression coverage for partial media-send failure, active-requester wake fallback, requester handoff lock/error fallback, pending announce fallback, and the dispatcher terminal path

Validation:
OPENCLAW_TEST_FAST=1 node_modules/.bin/vitest run src/agents/subagent-announce-dispatch.test.ts src/agents/subagent-announce-delivery.test.ts --reporter=dot

Result: 4 files passed, 174 tests passed.

@clawsweeper please re-review latest head 7d2ed55.

Follow-up changes:

• Added exact ## Real behavior proof section fields to the PR body using the Dicky live evidence, and locally verified the proof policy now returns passed.
• Stabilized the unrelated CI failure in src/agents/run-wait.test.ts: the NaN drain-timeout test normalized to a 1ms deadline and could race on loaded CI runners. The test now uses fake time for that deadline assertion.

Validation:
OPENCLAW_TEST_FAST=1 node_modules/.bin/vitest run src/agents/run-wait.test.ts src/agents/subagent-announce-dispatch.test.ts src/agents/subagent-announce-delivery.test.ts --reporter=dot

Result: 6 files passed, 214 tests passed.

@clawsweeper re-review

Rebased onto current main and pushed updated head 19dfdad770efb2f60a5486490946ff5487a8afdb. The previous OpenGrep failure was caused by the workflow scanning unrelated base-branch drift from the PR merge ref; the fresh OpenGrep run is now green. Focused local validation after rebase: OPENCLAW_TEST_FAST=1 node_modules/.bin/vitest run src/agents/run-wait.test.ts src/agents/subagent-announce-dispatch.test.ts src/agents/subagent-announce-delivery.test.ts --reporter=dot => 214 passed.

🦞🧹
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.

Follow-up proof for the fallback edge ClawSweeper flagged, plus a live repro of the exact failure mode.

A patched test instance was still running the earlier PR dist shape where primary generated-media direct delivery only stopped fallback when it returned delivered: true:

const generatedMediaDelivery = await tryGeneratedMediaPrimaryDirectDelivery();
if (generatedMediaDelivery?.delivered) return generatedMediaDelivery;

That is not enough. If the platform send posts media successfully and then throws during post-send bookkeeping/mirroring, the direct attempt returns a failed result even though Discord already has the file. The old branch then falls through to active wake/requester handoff and duplicates the generated media.

Live evidence from the same generated artifact:

• 2026-05-29T15:02:30.466Z: Discord message 1509934971605749963, attachment image-1---6750b96e-9d36-4afe-b472-4a3e74e9b81d.jpg, size 1266775, caption The generated image is ready.
• 2026-05-29T15:02:34.125Z: gateway log: Active requester session could not be woken for subagent completion; falling back to requester-agent handoff: ... queue_message_failed reason=no_active_run ...
• 2026-05-29T15:03:10.197Z: Discord message 1509935138249904181, same attachment filename and size, caption The last hour at The Tidepool ...

So this was direct/generated-media delivery followed by requester-agent fallback on the same artifact. The current PR head fixes that by treating any generated-media direct attempt as terminal for this path, and by marking direct delivery failure terminal: true, so a post-send failure cannot hand off and duplicate the artifact.

Local targeted gate on current head:

node_modules/.bin/vitest run src/agents/subagent-announce-delivery.test.ts
Test Files  2 passed (2)
Tests       152 passed (152)

I also deployed the current branch delta to the same test instance and restarted the gateway; the patched dist now has:

if (generatedMediaDelivery) return generatedMediaDelivery;
...
terminal: true

Gateway came back healthy: CLI/Gateway 2026.5.28-beta.1, runtime running, connectivity probe ok, admin-capable.

@clawsweeper please re-review this head.

@clawsweeper re-review

🦞👀
ClawSweeper picked this up.

Command router queued. I will update this comment with the next step.

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/26656568740
• Updated: 2026-05-29T19:11:18.572Z

Thanks for the report and patch. I checked this against today's main history, and this PR is now superseded by the maintainer-side completion-delivery changes that landed after it:

• fix: route generated media completions through requester agent #88141 changed generated media completions to use the requester-agent delivery contract.
• Fix subagent DM completion delivery after yield #88182 fixed the direct-message completion edge cases on that same path.
• refactor(agents): type media completion delivery misses #88250 added typed miss reasons for the remaining media completion delivery failures.

Current main also already has the partial-send terminal guard and dispatch stop behavior this PR was trying to add, while the remaining direct-first generated-media change conflicts with the newer requester-agent contract and the same tests now merge-conflict. Closing this as superseded rather than merging an older delivery shape.

If this still reproduces on current main, please open a fresh issue/PR with a new repro against the current requester-agent media completion path.