agent-skills vs autoresearch
AI agent development workflows compared
agent-skills provides reusable skills for taking AI-assisted development from specifications through implementation, review, testing, and pull requests. autoresearch focuses on goal-driven iteration, making bounded changes and keeping or reverting them based on a measurable result.

agent-skills: Structure AI-Assisted Development from Specs to PRs
A collection of reusable skills for AI coding agents, covering specification writing, implementation, review, testing, and autonomous sprint workflows. Useful for teams that want repeatable processes and clearer quality gates across agent sessions.

autoresearch: Automate Goal-Driven Iteration in Coding Agents
Built on Andrej Karpathy's autoresearch principles, this skill turns Claude Code, OpenCode, and OpenAI Codex into goal-directed improvement loops. It makes focused changes, checks them against a mechanical metric, and keeps or reverts each result.
| agent-skills | autoresearch | |
|---|---|---|
| Language | Shell | Shell |
| License | MIT | MIT |
| Stars | 2 | 6.7k |
| Forks | 0 | 486 |
| Last analyzed | Oct 3, 2026 | Oct 21, 2026 |
Key differences
- agent-skills emphasizes specification-driven workflows and documented project rules; autoresearch emphasizes repeated experiments against a mechanical metric.
- agent-skills supports specification checks, task implementation, reviews, testing, delegation, and an optional sprint harness; autoresearch centers on metric-based iteration and also offers commands for debugging, security reviews, and shipping.
- agent-skills describes workflows for agents supporting the SKILL.md format, while autoresearch lists installation for Claude Code, OpenCode, and Codex, with platform-specific command formats.
- Both projects use Shell and have MIT licenses. autoresearch reports 6.7k stars and 486 forks, while agent-skills reports 2 stars and 0 forks.
- agent-skills cautions that compatibility may vary by agent and runtime, and that its optional harness depends on other skills; autoresearch's core loop relies on Git commits and verification commands.
- agent-skills includes cmux-specific skills with macOS and session requirements; autoresearch provides Claude Code hooks for guardrails, which are not available on OpenCode or Codex.
Choose agent-skills if you…
- need to turn feature requests into requirements, designs, tasks, and acceptance test plans.
- want structured implementation, review, and testing workflows that follow documented project rules.
- plan to delegate work or trial a sprint-level harness within human-defined boundaries.
Choose autoresearch if you…
- want an agent to make focused changes and keep or revert them based on a measurable metric.
- need iteration outcomes recorded in TSV files for later review.
- use Claude Code, OpenCode, or Codex and want shared core skills with platform-specific command formats.
Compare the numbers side by side, or add more projects
This comparison is generated with AI from the OSRepos analyses of both projects. Always check each project's repository and documentation before choosing.