Tencent/SkillHone

168 stars · Last commit 2026-09-19

Continual agent skill evolution through persistent decision history. Whole-skill optimisation (SKILL.md + scripts + references) with every decision landing as a local Git issue / PR / wiki. Runs on any agentskills.io runtime — Claude Code, Codex, OpenClaw, Hermes.

README preview

<div align="center">

# SkillHone

### Turn the Skill failures your Agents encounter into lasting improvements.

Skills break in real work: a referenced script is missing, an API changes, or an
instruction stops producing the right result. That evidence usually disappears
inside a chat. SkillHone captures the failure while the Agent is working, repairs
the complete Skill repository, runs the regression test, and leaves a local PR
for review.

*Continual Agent Skill Evolution Through Persistent Decision History*

[![Paper](https://img.shields.io/badge/Paper-EMNLP%202026%20Industry-8b1a1a)](https://arxiv.org/abs/2606.08671)
[![Gitleaks](https://github.com/Tencent/SkillHone/actions/workflows/gitleaks.yml/badge.svg)](https://github.com/Tencent/SkillHone/actions/workflows/gitleaks.yml)
[![License](https://img.shields.io/badge/License-MIT-green)](LICENSE)

[Install](docs/install/skillhone.md) ·
[中文](docs/README.zh.md) ·

View full repository on GitHub →