Codex review: needs maintainer review before merge.

Summary
The PR removes the active-requester early return in subagent announcement delivery and updates the focused test to expect the existing requester-agent gateway handoff to run.

Reproducibility: yes. at source level: set expectsCompletionMessage, resolve an active requester session, make embedded-run queueing return not_streaming or no_active_run, and current main returns before the requester-agent handoff. The focused current-main test also asserts callGateway is not called for that path.

Real behavior proof
Sufficient (logs): The PR body supplies after-fix live Telegram gateway logs from a manually patched production install showing 3/3 subagent announcements delivered; the clean PR-build gap remains normal maintainer validation, not a contributor proof blocker.

Next step before merge
This is a real bug-fix PR with sufficient proof, but it is conflicting and overlaps a protected maintainer queue-first delivery PR, so the next step is maintainer consolidation and current-head validation rather than an automated repair branch.

Security
Cleared: The latest diff only changes in-repo agent delivery control flow and a focused test expectation, with no dependency, workflow, package, secret, or artifact-handling changes.

Revision — removed sendMessage fallback

Thanks for the thorough review @clawsweeper. The previous version (commit 8c74048) added a sendMessage fallback layer that:

1. Violated the delivery contract from fix(agents): clean subagent fallback scaffolding #78700 — raw child-output sends bypass the requester-agent rewrite/sanitization
2. Had type errors — AgentInternalEvent uses result, not summary/text
3. Introduced phantom path types — direct-fallback/direct-thread-fallback aren't in the SubagentDeliveryPath union

This revision (commit 73d712b) takes the simpler approach: just remove the early return. The callGateway handoff path below was always there and already does the right thing — it starts a proper requester-agent turn that rewrites and delivers through the session.

The fix is now: -11 lines, +5 comment lines. No new code, types, imports, or dependencies. Test expectations updated to reflect that callGateway is now reached.

node --check passes on both files.

Real Behavior Proof

Test suite: all 30 tests pass on PR branch

Ran the full test suite for the modified file on branch `fix/subagent-announce-delivery-drop-v2` (commit `73d712b`) against upstream main `95a1c91`:

$ node scripts/test-projects.mjs src/agents/subagent-announce-delivery.test.ts

[test] starting test/vitest/vitest.agents.config.ts

 RUN  v4.1.5 /tmp/openclaw-fork

 ✓  agents  src/agents/subagent-announce-delivery.test.ts (30 tests) 230ms

 Test Files  1 passed (1)
      Tests  30 passed (30)
   Start at  19:26:13
   Duration  3.66s (transform 1.39s, setup 349ms, import 2.91s, tests 230ms, environment 0ms)

[test] passed 1 Vitest shard in 9.37s

Key modified test: passes

The test that exercises the exact bug scenario — "queues when an active Telegram requester cannot be woken directly" — now verifies:

1. `queueEmbeddedPiMessage` returns false (active but non-consuming)
2. `callGateway` IS called (new — previously unreachable due to early return)
3. Direct-primary phase reaches callGateway, returns `delivered: false` with `"completion agent did not produce a visible reply"` (gateway mock returns empty)
4. Queue-fallback succeeds → `delivered: true, path: "queued"`

This confirms the fix: the early-return dead-end is removed, execution falls through to callGateway (requester-agent handoff), and the dispatch layer's queue-fallback path is reached on gateway mock failure.

What this proves

Contract preserved

• No raw child-output sends (no `sendMessage` fallback)
• All delivery goes through requester-agent handoff (`callGateway`) or queue-retry
• `SubagentDeliveryPath` union unchanged (`"queued" | "steered" | "direct" | "none"`)
• Consistent with PR fix(agents): clean subagent fallback scaffolding #78700 delivery contract

Syntax checks

$ node --check src/agents/subagent-announce-delivery.ts  # PASS
$ node --check src/agents/subagent-announce-delivery.test.ts  # PASS

Note: This is source-level proof from running the test suite on the PR branch. Live Telegram/push-channel reproduction would require a running OpenClaw gateway with a Telegram bot configured — not available in this contributor environment. The test suite exercises the exact seam identified in the clawsweeper review.

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/25630746101
• Updated: 2026-05-10T14:11:48.275Z

The Real behavior proof CI check requires live runtime evidence (screenshots, logs, recordings) or a maintainer proof: override.

This contributor environment does not have a live Telegram bot or push-channel setup to produce runtime screenshots. The PR body includes terminal output from the full test suite (30/30 pass), source verification, and before/after behavior analysis.

Requesting maintainer review for proof: override if the source-level evidence is sufficient, or guidance on what additional evidence would be needed.

Closing as superseded by #82834, merged in d887eb8.

That landed the active-requester wake fallthrough from this PR, added regression coverage that callGateway is reached after the wake attempt, and preserved contributor credit. The broader timeout/reconcile issues remain tracked separately.