Summary

What problem does this PR solve?

• Agent Teams could announce failed: subagent run lost active execution context while the same completion event still contained readable child assistant output, so parents treated successful work as a plain failure.

Why does this matter now?

• Issue #90299: sessions_spawn / sessions_yield orchestration depends on terminal status matching delivered child results; a false failed status breaks multi-agent trust and can trigger unnecessary retries or stalled workflows.

What is the intended outcome?

• When the subagent sweeper loses live run context and session-store reconciliation cannot settle terminal state, reconcile readable child transcript output first and complete as ok instead of emitting a plain lost-context failure.

What is intentionally out of scope?

• No new shared-memory layer, Agent Teams plugin changes, or announce formatting changes for unrelated error paths; true failures without usable child output still complete as errors.

What does success look like?

• Stale active runs with captured assistant output complete with outcome.status: "ok"; runs without output still report subagent run lost active execution context.

What should reviewers focus on?

• resolveStaleActiveSubagentOutcome in subagent-lost-context-completion.ts and the sweeper branch in subagent-registry.ts that calls it before sweeper-lost-context completion.

Linked context

Which issue does this close?

Closes #90299

Which issues, PRs, or discussions are related?

• ClawSweeper review on [Bug]: Agent Teams subagent completion can report "lost active execution context" while still delivering child output #90299 (clawsweeper:queueable-fix, clawsweeper:fix-shape-clear)
• Related lifecycle work in subagent registry / announce surfaces (no duplicate open PR for this issue at submission time)

Was this requested by a maintainer or owner?

• Community report; user confirmed fix in automation workflow.

Real behavior proof (required for external PRs)

• Behavior or issue addressed: Lost-context sweeper fallback vs. child output mismatch on Agent Teams completion events.
• Real environment tested: Local checkout E:\Projects\skills\work\openclaw on branch fix/issue-90299-lost-context-with-output, Node via node scripts/run-vitest.mjs.
• Exact steps or command run after this patch:
  • node scripts/run-vitest.mjs src/agents/subagent-lost-context-completion.test.ts
  • node scripts/run-vitest.mjs src/agents/subagent-registry.test.ts -t "completes stale active runs as ok when child session has readable output"
• Evidence after fix:

 Test Files  2 passed (2)
      Tests  2 passed | 120 skipped (122)
   Duration  17.80s

• Observed result after fix: Unit tests assert resolveStaleActiveSubagentOutcome returns ok when readSubagentOutput yields text, and the registry sweep test completes run-lost-context-with-output with outcome.status: "ok" (no lost-context error) when chat.history contains assistant output.
• What was not tested: Live Agent Teams sessions_spawn + sessions_yield end-to-end on a gateway host (read-only review path per ClawSweeper).
• Proof limitations: Regression coverage is source-level via vitest mocks for chat.history and sweep timing, not a reproduced bench transcript from the original report.
• Before evidence (optional): Sweeper always completed stale active runs without session terminal state using outcome.status: "error" and error: "subagent run lost active execution context" even when announce later read the same child output for the parent message.

Tests and validation

Which commands did you run?

• node scripts/run-vitest.mjs src/agents/subagent-lost-context-completion.test.ts
• node scripts/run-vitest.mjs src/agents/subagent-registry.test.ts -t "completes stale active runs as ok when child session has readable output"

What regression coverage was added or updated?

• subagent-lost-context-completion.test.ts — outcome resolver with/without output.
• subagent-registry.test.ts — sweep completes stale active run as ok when chat.history has assistant text.

What failed before this fix, if known?

• Sweeper sweeper-lost-context path emitted error outcome regardless of readable child transcript.

If no test was added, why not?

• N/A — regression tests added.

Risk checklist

Did user-visible behavior change? (Yes/No)

Yes.

Did config, environment, or migration behavior change? (Yes/No)

No.

Did security, auth, secrets, network, or tool execution behavior change? (Yes/No)

No.

What is the highest-risk area?

• False-positive ok completion if stale transcript output is unrelated to the current run.

How is that risk mitigated?

• Output is read only for the specific childSessionKey; session-store terminal reconciliation and notBeforeMs freshness checks still run first; empty/non-readable output keeps the existing lost-context error.

Current review state

What is the next action?

• Maintainer review

What is still waiting on author, maintainer, CI, or external proof?

• CI on fork PR

Which bot or reviewer comments were addressed?

• ClawSweeper narrow fix recommendation: reconcile child output before lost-context failure.