fix(#61): add summary length guidance and increase schema limit

[fullsend-ai-coder]

Add conciseness guidance to the retro-analysis skill for evidence notes routed into the summary field: one sentence per existing issue, prioritize most impactful evidence if space constrained.

Increase summary maxLength in retro-result.schema.json from 16,384 to 32,768 to provide additional headroom. GitHub supports up to 65,536 chars in issue comments, so this is within platform limits.

Both changes address the overflow risk introduced by PR #59, which routed all evidence for existing issues into summary without length guidance or increased schema capacity.

---

Closes #61

Post-script verification

• Branch is not main/master (agent/61-summary-length-guidance)
• Secret scan passed (gitleaks — 19a59848232667f38610e8c53967cb1b0d2a1d5d..HEAD)
• Pre-commit hooks passed (authoritative run on runner)
• Tests ran inside sandbox

[fullsend-ai-retro]

🤖 Finished Retro · ✅ Success · Started 3:14 PM UTC · Completed 3:19 PM UTC
Commit: 14bfba8 · View workflow run →

[fullsend-ai-retro]

Retro: PR #62 — summary length guidance and schema limit increase

Pipeline performance: excellent. PR #62 is a clean, minimal fix that addressed issue #61 (filed by the retro agent after analyzing PR #59). The full cycle — from retro detecting the overflow risk on PR #59 to PR #62 merging the fix — took 25 minutes, including human review and approval. The coder agent produced a single commit with 2 lines changed across 2 files, implementing both recommended options from the issue (conciseness guidance in the skill + doubling the schema maxLength).

No new proposals. The improvement opportunities identified below are already tracked in existing issues.

Evidence for existing issues

• Auto-file tracking issues for unresolved medium+ severity review findings fullsend#1956 (auto-file tracking issues for unresolved medium+ severity review findings): PR fix(retro): prohibit filing "evidence for" issues #59's review by qodo-code-review flagged the summary overflow risk as finding fix(#1): update harness dest paths and restructure for forge #3 (reliability category). Human reviewers acknowledged but did not address or track this finding before merging. The retro agent independently rediscovered it 7 minutes later, triggering a full issue→triage→code→review→merge cycle. Auto-filing a tracking issue for the unresolved qodo finding would have eliminated this extra cycle.

• Review agent missed vouch-check missing issues: write permission (caught by qodo) #46 (review agent missed what qodo caught): This is the second instance of this pattern in this repo. On PR docs: add CLAUDE.md, COMMITS.md, CONTRIBUTING.md, LICENSE, and repo workflows #32, qodo caught a missing issues: write permission that the fullsend review agent missed. On PR fix(retro): prohibit filing "evidence for" issues #59, qodo caught the schema overflow risk. In both cases, the fullsend review agent did not independently identify the finding, and human reviewers dismissed it without tracking.

• feat(retro): post evidence comments on related issues #60 (post evidence comments on related issues): Once implemented, this would reduce pressure on the summary field by routing evidence to the relevant issues directly, making the maxLength increase less critical as a defense mechanism.