---
slug: "autoharness"
name: "autoharness"
effect: "Claude Code plugin that learns, merges, updates and prunes skills from your real sessions"
category: "devtool"
tier: "rising"
url: "https://github.com/tigerless-labs/autoharness"
github: "tigerless-labs/autoharness"
pricing: "open-source"
license: "MIT"
tags: ["claude-code", "plugin", "skills", "self-learning", "session-learning"]
platforms: ["claude-code"]
install_kind: "cli"
date_added: "2026-10-19"
rank: 20
rising_score: 54
stars: 11145
star_velocity: 84.7
velocity_basis: "since creation"
created_at: "2026-06-09T21:05:42Z"
pushed_at: "2026-10-15T16:22:50Z"
signals_status: "ok"
signals_fetched_at: "2026-10-19T11:39:01.539Z"
page: "https://agenttooldrop.com/tools/autoharness"
---

# autoharness

Claude Code plugin that learns, merges, updates and prunes skills from your real sessions

Category: Devtools. Tier: new and rising, rank 20 by new and rising score (54 of 100). Pricing: Open source. License: MIT.

Platforms: claude-code.

## Why you would want it

autoharness is a Claude Code plugin that distills skills from the sessions you are already having. A background reflection fires once a session has done enough work, counted in tool calls, and /learn distills the current session on demand. A new lesson is compared against existing skills, and same-scenario skills are folded into one instead of stacking near-duplicates. Each session opens with an index of the skills it wrote, and every create or update is logged to a per-skill ledger with the transcript slice behind it. It only touches skills it wrote itself, and it needs no daemon or benchmark.

## Best for

- Turning Claude Code sessions into reusable skills
- Keeping a skill library from filling with near-duplicates
- Tracing why each learned skill exists

## Use it when

You use Claude Code on ongoing work and want repeated workflows and corrections to become native skills under .claude/skills/ without writing them by hand.

## Skip it if

You want to write and review every skill yourself, or you do not want session content captured (the README says it grabs each turn's user input, agent output and tool I/O, redacted at egress). Reflections run in the background in a separate subagent, and the README says its default thresholds are placeholders pending empirical calibration. It needs Python 3.11+ as python3 on your PATH, or its hooks do not fire. Uninstalling leaves the learned skills and state on disk until you delete them.

## Editor note

The repo has no GitHub releases; per the README, the installed plugin is cached by the version field in .claude-plugin/plugin.json. Run /reload-plugins or restart Claude Code after installing.

## Install

Install as a Claude Code plugin:

```sh
/plugin marketplace add tigerless-labs/autoharness
/plugin install autoharness@autoharness
```

Source: https://github.com/tigerless-labs/autoharness#install

## Live signals

Fetched 2026-10-19T11:39:01.539Z (status: ok).

- Repository: tigerless-labs/autoharness
- Repository created: 2026-06-09
- Stars: 11,145 (GitHub's count; it cannot be authenticated)
- Stars per day, since repo creation: 84.7 (an average over the repo's life, not a growth rate)
- Forks: 629
- Open issues and PRs: 65
- License (GitHub): MIT
- Last push: 2026-10-15
- Latest release: none

## Links

- Website: https://github.com/tigerless-labs/autoharness
- GitHub: https://github.com/tigerless-labs/autoharness
- Install instructions verified at: https://github.com/tigerless-labs/autoharness#install
- Source for this entry: https://github.com/tigerless-labs/autoharness#readme
- AgentToolDrop page: https://agenttooldrop.com/tools/autoharness
- Full catalog (JSON): https://agenttooldrop.com/tools.json