AICodingFlow: Skill for Claude Code

update-dedupe

Use this skill to turn strong GitHub duplicate-closure evidence into concise updates to the repo-local dedupe-issue companion skill.

This skill owns only the self-evolution logic: how to interpret aggregated duplicate feedback and propose local guidance. The GitHub Actions runner owns data collection, write-surface validation, commits, pushes, and PR creation. When run inside GitHub Actions, .agents may be read-only. Write proposed changes only to update-dedupe-output/; the runner applies them.

Workflow

1. Read the aggregated duplicate feedback JSON provided by the runner.
2. Validate that the data contains only structured duplicate evidence.
3. Identify repeated duplicate clusters where two or more independent issues were closed as duplicates of the same canonical issue.
4. Compare the repeated clusters with the existing .github/skills/dedupe-issue-repo/SKILL.md content when available.
5. Convert uncovered repeated clusters into concise repo-specific guidance.
6. Write proposed local companion skill content to update-dedupe-output/.
7. Stop; the runner validates and publishes the result.

Output Contract

Always write update-dedupe-output/status.json:

{
  "status": "changed",
  "reason": "Brief evidence summary.",
  "updated_files": [".github/skills/dedupe-issue-repo/SKILL.md"]
}

Allowed statuses:

• changed when repeated maintainer duplicate evidence should update guidance
• no_change when evidence is insufficient or already covered
• error when the feedback cannot be interpreted safely

Use no_change when there is no repeated cluster. A single duplicate closure, comments that only suggest a duplicate, title similarity, or agent-only inference is not enough to update guidance.

For changed, write the complete replacement content for:

• update-dedupe-output/dedupe-issue-repo/SKILL.md

Do not edit .agents directly.

Evidence Rules

Only learn from structured evidence supplied by the aggregation script: