improve

Parse $ARGUMENTS:

• If a number is present, use it as the iteration count. Otherwise default to 5.
• If a quoted string is present (e.g. "Replace fetch() with apiClient"), use it as the FOCUS for this run.
• Both can be combined: /autoimprove:improve 10 "Fix any types in convex/" → 10 iterations, focused on that task.

Start the autoimprove loop for the parsed iteration count.

Pre-flight:

1. Check .claude/autoimprove/config.md exists. If not, stop: "Run /autoimprove:setup first."
2. Confirm git working tree is clean. If not, stop: "Please commit or stash changes first."
3. Run worktree setup (create .claude/autoimprove/worktrees/, update .gitignore).
4. Run measure skill in the main directory — record as BASELINE.
5. Print the pre-flight summary with progress indicators:

   ━━━ Pre-flight ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
   ✓ Config found
   ✓ Git working tree clean
   ✓ Worktree directory ready
   ✓ Baseline score: XX/100
   ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
   Starting N iterations. Focus: [focus or "all improvement areas"]
   Main branch is safe — experiments run in isolated worktrees.
   

Run the improve-loop skill for the requested number of iterations, passing the FOCUS string if one was provided. Pass session_mode: new so it creates a fresh session header in the log.

After all iterations — session summary:

• Clean up all remaining worktrees (worktree skill cleanup step)
• Report: total iterations, wins vs discards, baseline → final score
• List all commits added to main (one per winning experiment)
• Show: git log --oneline -N where N = number of wins
• Remind user: "Review with git log --oneline and git diff HEAD~N before pushing."