autoharness
An agent that gets better by working, not by being rebuilt. AutoHarness turns every session into lasting capability, so the harness improves itself as the model moves on.
10956stars

An open-source AI lab shipping tools, agents, and experiments across the AI development stack.
Tools that change how your coding agent works — and show you what it is actually doing.
An agent that gets better by working, not by being rebuilt. AutoHarness turns every session into lasting capability, so the harness improves itself as the model moves on.
10956stars

Usage logs tell you what a turn cost; Cost X-ray tells you why, attributing every token in the context to its source.
4698stars

A memory layer beneath the agent: knowledge that connects like a graph, reads like a vault, and matures with use. Runs locally, shared across agents.
3399stars

Skills that put the agent to work on everything around the code — design, research, distribution.
Turns your PRs and the day's hot threads into topics, X and Reddit post drafts, and reply drafts — drafts only, never posts for you.
633stars

A crawler's-eye audit of your site's SEO and GEO foundation: 30 deterministic checks, pass or fail.
701stars

Feed your agent papers and half-formed ideas; it links them into a system design you can defend. An Agent Skill for Claude Code and any SKILL.md-compatible agent.
230stars

Finds the AI papers 28 labs put on arXiv over any date range, and separates the ones a company led from the ones it only appears on.
219stars

Free, open-source influencer discovery: finds creators to partner with across 15 major platforms — TikTok, Instagram, X and more — with their public contact info.
213stars

Tools for safer and more reliable AI development workflows.
PHI boundary reports for healthcare and insurance AI workflows — across prompts, RAG, tools, memory, and logs.
204stars

Lightweight, rule-based checks for missing, weak, and uncovered tests in pull requests.
139stars

More in the works — built in the open, shipped fast.