Source
DeepReport intelligence briefing — 2026-05-11. References: #31509 (open bug filed today by [aw] Failure Investigator), Safe Output Health Report (discussion #31430).
Problem
The link_sub_issue safe-output handler treats "sub-issue is already a sub-issue of parent" as a fatal error instead of a no-op. The agent job succeeds, but the post-agent safe-output validation step fails — producing a misleading "workflow failed" signal and a spurious auto-ticket downstream. This is self-affecting: [aw] Failure Investigator (6h) itself trips on this when re-linking already-linked sub-issues to its shared parent #30961.
Reproducer
Run §25657654477 — agent succeeded after 18.5 min ($7.94, 114 turns) but safe_outputs failed in 24s on a no-op link_sub_issue re-assertion.
Suggested fix
- In the
link_sub_issue handler: if GitHub reports the sub-issue is already linked to the parent, log a ##[warning] (already done) and mark the message as success — do not fail the safe-output job.
- Audit
unlink_sub_issue, add_labels, remove_labels for the same warn-then-fail pattern.
- Add a unit/integration test asserting that linking an already-linked sub-issue exits cleanly.
Effort / Impact
- Effort: < 2 hours (single handler + test)
- Impact: High — throws away the most expensive work in the portfolio when triggered.
- Suggested agent: Code-modification agent (Claude / Copilot)
Success criteria
- Re-running
[aw] Failure Investigator (6h) with an already-linked sub-issue does not fail the safe-output job.
- New regression test in the safe-outputs handler suite.
Related: #31509, #31430
Generated by DeepReport - Intelligence Gathering Agent · ● 12.8M · ◷
Source
DeepReport intelligence briefing — 2026-05-11. References: #31509 (open bug filed today by [aw] Failure Investigator), Safe Output Health Report (discussion #31430).
Problem
The
link_sub_issuesafe-output handler treats "sub-issue is already a sub-issue of parent" as a fatal error instead of a no-op. The agent job succeeds, but the post-agent safe-output validation step fails — producing a misleading "workflow failed" signal and a spurious auto-ticket downstream. This is self-affecting:[aw] Failure Investigator (6h)itself trips on this when re-linking already-linked sub-issues to its shared parent #30961.Reproducer
Run §25657654477 — agent succeeded after 18.5 min ($7.94, 114 turns) but
safe_outputsfailed in 24s on a no-oplink_sub_issuere-assertion.Suggested fix
link_sub_issuehandler: if GitHub reports the sub-issue is already linked to the parent, log a##[warning](already done) and mark the message as success — do not fail the safe-output job.unlink_sub_issue,add_labels,remove_labelsfor the same warn-then-fail pattern.Effort / Impact
Success criteria
[aw] Failure Investigator (6h)with an already-linked sub-issue does not fail the safe-output job.Related: #31509, #31430