Rig Tidy

Post-Merge Cleanup

Audit and clean up the diff from the last N commits. Catches dead code, unused imports, stale comments, and obvious duplicates that slipped in during rapid development.

Configuration

Reads .rig/config.json:

• sourceScope[0] — the default directory to audit when no scope is given (default: src).
• test.command — the suite run in the verify step (default: npm test).
• agents.reviewer — the project's name for the canonical reviewer role (default: reviewer).

If the file is absent, use the defaults above and note you're running unconfigured.

Arguments

The user invoked this with: $ARGUMENTS

What to do

1. Parse the arguments.

   • scope: directory to audit (default: the default source scope from .rig/config.json (sourceScope[0], default src)).
   • commits: number of recent commits to review (default: 10)

   Examples:

   • /rig-tidy → default scope, 10 commits
   • /rig-tidy path/to/dir 5 → that directory, 5 commits
   • /rig-tidy 20 → default scope, 20 commits

2. Show what will be audited.

   git log --oneline -${commits}
   git diff HEAD~${commits} --stat -- ${scope} | tail -20
   

3. Audit the diff via a fresh-context subagent. Spawn a reviewer agent (mapped through agents.reviewer) with this prompt (don't do the audit inline — the implementer context is too close to the code to spot dead-on-arrival abstractions):

   "Audit the diff git diff HEAD~${commits} -- ${scope} for cleanup candidates. Categories: (a) dead code — exported symbols with no callers in the tree, internal helpers used once; (b) unused imports; (c) stale comments — TODOs completed by the diff, comments describing prior behavior, doc lines that no longer match the code; (d) duplicates — near-identical helpers, repeated inline expressions that could collapse; (e) inconsistencies — naming drift, mixed patterns introduced during the window.

   For each finding give file:line, the category, one-line justification, and the proposed deletion or merge. Skip anything you're not confident about. If the window is clean, say so."

4. Apply the high-confidence findings. For each item the reviewer flagged with concrete file:line + proposed edit, apply the Edit. Skip anything ambiguous and surface it to the user instead.

5. Verify. Run the test command from .rig/config.json (test.command, default npm test):

   <test.command>
   

   If anything fails, revert the offending edits (git checkout -- <file>) and report which removal broke tests so the user can decide whether to keep or rework.

6. Report. Lines removed, files touched, anything skipped and why.