AutoHarness

AutoHarness is a Claude Code plugin that treats the skill library as a living system. While you work, hooks capture tool-heavy episodes, a reflector proposes add/merge/patch/delete intents, and a promoter is the only writer that lands validated skills into .claude/skills. It only touches skills it authored, keeps a per-skill ledger of why each change happened, and opens every session with an index of its own library so recall does not depend on the host guessing.

Static skill packs rot the week after you install them. Founders running Claude Code all day need the harness to keep its own playbooks current from real work — not another dump of SKILL.md files and not a separate labeling job. AutoHarness is that missing lifecycle: learn, fold, graduate, archive.

In Claude Code run marketplace add tigerless-labs/autoharness, install autoharness@autoharness, then reload plugins. Zero config after that: tool-call cadence triggers background reflection, /learn forces a distill of the current session, and env knobs tune reflect/consolidate rates and capacity. State lives under .claude/autoharness/ and ~/.claude/autoharness/; skills land as normal native files the host already knows how to load.

Skill packs and one-shot recorders freeze a lesson once. AutoHarness keeps running beside the host: merge same-scenario skills instead of stacking twins, prune by usage rate after probation, and never rewrite hand-written skills. Same model, different harness — the bet is the skill layer can maintain itself.

• MCP server / client support

• Skill Recorder ↗
• SkillOpt ↗
• Impeccable ↗