v1.3.0

What's New

• /autoimprove:audit — Scan your codebase for deficiencies with a prioritized fix plan, efficiency ranking, and token usage estimates. Offers to start fixing interactively, area by area.
• /autoimprove:continue — Resume interrupted sessions from where they left off. Inherits iterations, focus, and baseline from the log.
• Setup auto-runs audit after generating config, so first-time users see the full report immediately.
• Session tracking — Log now includes structured session headers with status tracking (IN_PROGRESS / COMPLETE).

Changed

• Setup protects existing config (asks before re-detecting stack)
• Improved audit behavior for 100/100 scores (consistent output format)

Removed

• /autoimprove:measure command (subsumed by /autoimprove:audit — measure skill kept as internal utility)

See the full changelog for details.

v1.2.0

What's Changed

Added

• Progress indicators in the improvement loop — visible status lines at every step
• .gitignore to exclude generated autoimprove data from the plugin repo
• CLAUDE.md with release checklist rules
• CHANGELOG.md backfilled for all versions

Fixed

• hooks/sessionstart.sh missing execute permission

Removed

• Project-specific .claude/autoimprove/config.md from repo

Full Changelog

v1.1.0...v1.2.0

v1.1.0

What's New

Auto-update check

• SessionStart hook checks GitHub for new releases once per day (3s timeout, 24h cache)
• Users see update notifications directly in the status line
• New /autoimprove:upgrade command to install latest version

via HAPI

v1.0.0

autoimprove v1.0.0

Autonomous codebase improvement loop for Claude Code.

Installation

/plugin marketplace add benmarte/autoimprove
/plugin install autoimprove@autoimprove

Features

• Auto-detect stack and generate autoimprove.config.md via /autoimprove:setup
• Run improvement loop with isolated git worktrees — main branch is never touched mid-session
• Composite quality score (type check, build, tests, lint) on a 0–100 scale
• Winning experiments squash-merged to main, losers deleted cleanly
• Detailed logging to autoimprove-log.md
• Inline focus mode: /autoimprove:improve 10 "Replace all any types" — no config editing needed
• Support for 10+ languages (TypeScript, Python, Go, Rust, Ruby, Java, C#, PHP, Swift, and more)

Commands

• /autoimprove:setup — detect stack, generate config, show baseline score
• /autoimprove:improve [N] ["focus"] — run N iterations, optionally focused on a specific task
• /autoimprove:measure — check current score without changes
• /autoimprove:status — summary of all runs from the log