collect-pr-info

Collect data, draft complete PR content, cleanup, return ready-to-use JSON.

Arguments

Workflow

1. Run ${CLAUDE_PLUGIN_ROOT}/skills/collect-pr-info/scripts/collect-info.sh -a <target> --lang <lang>
2. Check for errors
3. Read temp files (commits, diffs)
4. Draft PR title and body
5. Prepare preceding PR update (if exists)
6. Run ${CLAUDE_PLUGIN_ROOT}/skills/collect-pr-info/scripts/cleanup.sh <temp_dir>
7. Return complete PR data

Output JSON

{
  "current_branch": "feature/x",
  "target_branch": "origin/main",
  "push_needed": true,
  "matched_spec": "auth",
  "pr": {
    "title": "feat(auth): add login module",
    "body": "## Summary\n..."
  },
  "existing_pr": {
    "exists": false,
    "number": null
  },
  "preceding_pr": {
    "exists": true,
    "number": 122,
    "updated_body": "## PR Order\n- this PR\n- #NEW\n\n<original body>"
  }
}

Agent replaces #NEW with actual PR number after creation.

Title Format

<prefix> type(scope): description

• Prepend --prefix value if provided
• Type: feat/fix/refactor/chore/docs/test (from commits)
• Scope: matched_spec if single, omit if multiple/none
• Max 50 chars (excluding prefix), lowercase, imperative

Body Format

Language: lang.effective

## Summary
What and why (2-3 sentences)

## Changes
- Key changes

## Testing
How to verify

## Related Issues
closes #N

Preceding PR Update

If preceding PR exists, prepare updated_body:

## PR Order
- this PR
- #NEW

<original body>

Use ## PR順序 for Japanese.

Large Diff Strategy

When split: read manifest.json, prioritize source files, skip generated.

Error Handling

{"error": "message", "error_code": "NOT_GIT_REPO"}