autoresearch
Claude Autoresearch Skill — Autonomous goal-directed iteration for Claude Code. Inspired by Karpathy's autoresearch. Modify → Verify → Keep/Discard → Repeat forever.
Autoresearch is an autonomous iteration engine for Claude Code, OpenCode, and Codex that transforms AI coding assistants into relentless optimization systems. Inspired by Karpathy's autoresearch, it follows a modify-verify-keep/discard loop that automatically improves any measurable metric through bounded iteration cycles. The skill provides 14 specialized commands covering debugging, security audits, documentation generation, requirement analysis, and systematic optimization, with automatic git-based rollback of failed experiments. Each iteration makes one atomic change, commits it, verifies the metric, and keeps improvements while auto-reverting failures.
Key Features
Use Cases
- 01Increase test coverage from 72% to 90% by iteratively adding tests
- 02Reduce API bundle size or response time through systematic optimization
- 03Hunt and fix all bugs, type errors, or lint issues in a codebase autonomously
- 04Run STRIDE/OWASP security audits with adversarial red-team analysis
- 05Generate and maintain documentation that stays in sync with code changes
- 06Validate pull requests with regression testing across 8 stability dimensions
Related Skills
View moresuperpowers
An agentic skills framework & software development methodology that works.
skills
Skills for Real Engineers. Straight from my .agents directory.
skills
Public repository for Agent Skills
ponytail
Makes your AI agent think like the laziest senior dev in the room. The best code is the code you never wrote.
autoresearch — FAQ
What is autoresearch and how does it work?+
Autoresearch is an agent skill that turns Claude Code, OpenCode, or Codex into an autonomous optimization loop. It makes one atomic change per iteration, commits it to git, verifies against a metric, and keeps improvements while auto-reverting failures until reaching the goal or iteration limit.
How do I install autoresearch in Claude Code?+
Run 'npx skills add uditgoenka/autoresearch' and restart Claude Code, or use '/plugin marketplace add uditgoenka/autoresearch' followed by '/plugin install autoresearch@autoresearch'. The guided installer './scripts/install.sh --claude --global' is also available.
Which AI clients does autoresearch work with?+
Autoresearch works with Claude Code (full feature set including hooks), OpenCode (core skill and 14 commands with underscore naming), and OpenAI Codex (via mention syntax). Hook-based guardrails are Claude Code-only.
Do I need API keys or prerequisites?+
You need git installed and a working Claude Code, OpenCode, or Codex environment. No additional API keys are required. The skill uses the agent's native capabilities to execute commands and verify metrics.
Is autoresearch free to use?+
Yes, autoresearch is open source under the MIT license and free to use. The skill itself has no cost; you only need access to Claude Code, OpenCode, or Codex.
How do I stop an autoresearch loop?+
Press Ctrl+C to interrupt, or specify 'Iterations: N' when starting the command to set a bounded limit. Every successful change is committed to git, so your last working state is always preserved.
How do I install autoresearch?+
Open the source repository on GitHub and follow its README. autoresearch is a skill — MCP Agents Market links you directly to the official repo.
Is autoresearch free?+
autoresearch is an open-source project hosted on GitHub. Check the repository for its license and any usage requirements.