Retro agent should search for related PRs and recent fixes before filing issues

[fullsend-ai-retro]

What happened

The retro agent filed issue #7305 proposing AGENTS.md documentation about data-acceptable-bundles GC risk. This triggered the full pipeline: triage (run 27749792636), code (run 27750013305), and review (run 27750234906). Maintainer yftacherzog closed PR #7306 and the issue, explaining that PR #7164 had already implemented an upstream fix and the documentation was unnecessary. The entire pipeline (4 agent runs, ~40 minutes) was wasted work.

What could go better

The retro agent proposed a documentation change for a problem (issue #6997) that already had an active fix in PR #7164. Had the retro agent searched for existing PRs mentioning issue #6997 or the data-acceptable-bundles image, it would have found PR #7164 and could have either skipped the proposal or noted the existing fix attempt.

Existing issue #1948 covers searching for related open issues but does not extend to searching for related PRs (open or recently merged/closed) that may already address the root cause. This is a meaningful gap: in this case, the fix was a PR, not an issue, so issue-only dedup would not have caught it.

Confidence: High. The causal chain is clear — the retro agent filed an issue without checking for existing PRs addressing the same root cause, leading to wasted pipeline runs.

Proposed change

Extend the retro agent's pre-filing validation (in the retro agent definition or the retro-analysis skill's 'Before proposing' section) to include a PR search step alongside the existing issue search. Specifically:

1. Before filing a new issue, the retro agent should search for open and recently-closed PRs in the target repo that reference the same root-cause issue or touch the same files/components.
2. The search should use both gh api search/issues (which covers PRs) with queries like <keywords> repo:<target> is:pr and direct searches for referenced issue numbers.
3. If a related PR exists (open or closed within last 90 days), the retro agent should either skip the proposal or note the existing work and explain why additional action is still needed.

This refines the scope of #1948 to include PR-based dedup, not just issue-based dedup.

Validation criteria

In the next 10 retro agent runs that propose issues related to problems with existing PRs, the retro agent should either (a) skip the proposal and note the existing PR in its summary, or (b) explicitly justify why a new issue is needed despite the existing PR. Zero instances of filing issues for problems already addressed by existing PRs.

---

Generated by retro agent from konflux-ci/konflux-ci#7306

[fullsend-ai-triage]

🤖 Finished Triage · ✅ Success · Started 10:18 AM UTC · Completed 10:20 AM UTC
Commit: 4e21a60 · View workflow run →

[fullsend-ai-triage]

Triage Summary

Severity: medium · Category: feature

Problem

The retro agent's pre-filing validation only searches for duplicate issues, not PRs. When a root-cause fix exists as a PR rather than an issue, the retro agent misses it and files a redundant proposal — triggering a full pipeline cycle (triage → code → review) that produces wasted work.

The concrete example: the retro agent filed an issue proposing documentation for a GC risk in konflux-ci, unaware that a PR already implemented the upstream fix. The entire pipeline (~40 minutes, 4 agent runs) was wasted, and a maintainer had to manually close the result.

Root Cause

The retro-analysis skill (internal/scaffold/fullsend-repo/skills/retro-analysis/SKILL.md) instructs the agent to search for existing issues before proposing, but has no step for searching PRs — open or recently closed/merged.

Recommended Fix

Add a PR search step to the "Before proposing" section of the retro-analysis skill:

1. Search for open and recently-closed PRs in the target repo using gh api search/issues with is:pr queries matching the root-cause issue number and relevant keywords.
2. If a related PR exists (open or closed within ~90 days), either skip the proposal or explicitly justify why additional action is still needed.
3. Update the retro agent definition (internal/scaffold/fullsend-repo/agents/retro.md) if it has independent pre-filing guidance.

Related Issues

• Retro agent should search for related open issues before proposing fixes #1948 — Retro agent should search for related open issues (subsystem-level search). This issue extends that to cover PRs.
• Retro dedup recipe should include author-scoped search for self-duplicates #1563 — Retro dedup should include author-scoped self-duplicate search. Different dimension (same-author dedup).

Proposed Test Case

Scenario: PR-level dedup prevents redundant retro proposals
  Given a recently-merged PR in the target repo that fixes the root cause
  And the retro agent analyzes a closed PR related to the same root cause
  When the retro agent runs its 'Before proposing' checks
  Then the PR search surfaces the existing fix
  And the retro agent skips the proposal or justifies why a new issue is needed

Edge cases:
  - PR closed without merging (fix rejected) — proposal may still be valid
  - PR merged >90 days ago — should not suppress proposals for recurring issues
  - Multiple related PRs with mixed states

---

Labels: Retro agent feature targeting the harness/skills layer, consistent with labeling on related issue #1948.

[fullsend-ai-prioritize]

RICE Priority Score: 1.2

Score breakdown

Dimension	Score	Reasoning
Reach	1.5	The retro agent runs across multiple repos/customers, and any repo with active PRs fixing root causes is susceptible to redundant issue filings. This affects all orgs using the retro agent pipeline, scoring 1.5 for multiple strategic customers.
Impact	1	Each redundant filing wastes ~40 minutes of pipeline time (triage, code, review runs) and requires maintainer intervention to close. This is a noticeable workflow improvement but has a clear workaround (maintainers manually close), scoring 1 for medium impact.
Confidence	0.8	High confidence at 0.8. The issue provides a concrete example with full causal chain, the root cause is clearly identified in the retro-analysis skill, and the fix scope is well-defined. The only gap is uncertainty about edge cases (closed-without-merge PRs, timing thresholds).
Effort	1	Medium effort at 1. The fix requires adding a PR search step to the retro-analysis skill definition and possibly the retro agent definition, touching a few files. The implementation uses existing gh API search capabilities and follows the pattern of the existing issue-dedup logic from #1948.

Formula: (1.5 x 1 x 0.8) / 1 = 1.2