Skip to content
Back to search

SkillHone

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.

Safety scan

A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.

Checked today
Badge for your README
findsafeskills safety rating

Markdown

[![findsafeskills safety rating](/api/listings/tencent--skillhone/badge.svg)](/listings/tencent--skillhone)

HTML

<a href="/listings/tencent--skillhone"><img src="/api/listings/tencent--skillhone/badge.svg" alt="findsafeskills safety rating" /></a>

Install

i

Install instructions from the README

As written by the repo's author. findsafeskills has not verified these steps.

Give this to your coding Agent:

> Install SkillHone from the `main` branch of > `https://github.com/Tencent/SkillHone`. Follow > `docs/install/skillhone.md`, make the SkillHone Skills and CLI available to > this Agent, and verify the installation without changing unrelated files. > Use the documented `--install-links=true` Git installation command.

SkillHone installs directly from GitHub. There is no package registry account to configure and no hosted service to deploy.

What was scanned

The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.

README.md · README · 105 lines

View this file on GitHub (read at the repository's current version: the scan did not record which commit it read)

This file has changed since it was scanned, so this is not exactly the text that was scanned, and no line is marked. The findings are listed with the safety scan above.

  1. <div align="center">
  2. # SkillHone
  3. **Continually improve your Agent Skills — and keep what was learned.**
  4. [![Paper](https://img.shields.io/badge/EMNLP_2026-Industry_Track-286c3b)](https://aclanthology.org/2026.emnlp-industry.105/)
  5. [![Version](https://img.shields.io/badge/version-2.0.0--rc.1-315fbd)](https://github.com/Tencent/SkillHone/releases)
  6. [![Agent Skills](https://img.shields.io/badge/Agent_Skills-Compatible-7657a5)](skills/)
  7. [![CI](https://github.com/Tencent/SkillHone/actions/workflows/ci.yml/badge.svg)](https://github.com/Tencent/SkillHone/actions/workflows/ci.yml)
  8. [![License](https://img.shields.io/badge/License-MIT-286c3b)](LICENSE)
  9. [Get started](#get-started) · [中文](docs/README.zh.md) ·
  10. [Public example](examples/phoenix-tracing/) · [Paper](https://aclanthology.org/2026.emnlp-industry.105/)
  11. ![SkillHone: a real failure becomes an Issue, a tested repair, a reviewable local PR, and persistent decision history.](docs/assets/skillhone-hero.svg)
  12. </div>
  13. ## Get started
  14. Give your coding Agent this prompt:
  15. > Install SkillHone from `https://github.com/Tencent/SkillHone`, following
  16. > `docs/install/skillhone.md`. Make the CLI and Skills available to this Agent,
  17. > use the documented `--install-links=true` Git installation command, and
  18. > verify the installation.
  19. Prefer the terminal? Start with the [installation guide](docs/install/skillhone.md)
  20. for installation, provider configuration, and verification.
  21. ## Why SkillHone
  22. **Accepted at EMNLP 2026 Industry Track.** The paper reports **+15.8 points on
  23. GAIA** and **+3.2 points on WebWalkerQA-EN** against its commercially backed
  24. deep-research baseline, without a pre-integrated search stack.
  25. [Read the paper and evaluation context](https://aclanthology.org/2026.emnlp-industry.105/).
  26. An Agent hits a missing script, a broken reference, or an outdated instruction.
  27. SkillHone captures the evidence, repairs the **complete Skill repository**, and
  28. returns a tested local PR. The Issue, test, diff, run and decision stay together,
  29. so the next Agent can build on what was learned.
  30. Works with CLI-capable Agents, including **Codex, Claude Code, Cursor, Pi and
  31. ZCode**. New installations use the current Agent for repair assignments,
  32. preserving its model and account. [DeepSeek Harness](docs/deepseek-harness.md)
  33. is an optional managed executor; existing installations retain their saved
  34. choice. The host verifies the committed repair and linked tests before a PR.
  35. ## SkillHone in 35 seconds
  36. https://github.com/user-attachments/assets/69bafb25-5ba9-44cb-b5ff-8a2a94f8e51a
  37. <details>
  38. <summary>Inspect the current workbench</summary>
  39. ![Current SkillHone workbench showing Skills, per-Skill approval counts, local PRs and repair history](docs/assets/skillhone-workbench-current.png)
  40. See which Skills need attention, expand their repair records, and inspect each local PR before merging.
  41. </details>
  42. **Reproduce a real repair.** The [Phoenix tracing walkthrough](examples/phoenix-tracing/) starts with four referenced documents that were never shipped, then follows the failure, regression test and repair evidence. Public source: [Issue #2567](https://github.com/github/awesome-copilot/issues/2567) · [merged fix #2568](https://github.com/github/awesome-copilot/pull/2568).
  43. ## What stays with the fix
  44. | You need | SkillHone keeps |
  45. |---|---|
  46. | A problem the next Agent can reproduce | One Issue with sanitized evidence and a linked regression test |
  47. | A complete repair | Changes to `SKILL.md`, scripts, references, assets and tests in one Git diff |
  48. | A result you can review | Test outcomes, repair run, commits and a local PR |
  49. | A reason the next Agent can reuse | Per-Skill Wiki records and persistent decision history |
  50. Repairs use isolated branches. **Review is the default**; automatic local merge
  51. is opt-in and requires linked tests to pass. Pushing remains a separate user
  52. action. The [local workbench](docs/install/developer.md#run-the-workbench) brings
  53. Issues, runs, PRs, approval state and Wiki records into one view.
  54. ## Two ways to improve a Skill
  55. | | Runtime maintenance | Evaluation-driven optimization |
  56. |---|---|---|
  57. | Start with | A reproducible failure encountered during real work | A frozen dataset and verifier in a separate Eval repository |
  58. | Use it for | Missing files, broken scripts, stale instructions and API drift | Broader capability and quality improvements |
  59. | Review | Issue, regression, repair and local PR | Baseline, candidate iterations, validation and local PR |
  60. | Learn more | [Runtime workflow](skills/skillhone-auto-optimization/SKILL.md) | [Benchmark workflow](docs/benchmark-optimization.md) |
  61. ## Research and citation
  62. > Zhiwei Li and Yong Hu. *SkillHone: A Harness for Continual Agent Skill
  63. > Evolution Through Persistent Decision History.* EMNLP 2026 Industry Track.
  64. > [ACL Anthology](https://aclanthology.org/2026.emnlp-industry.105/) · [arXiv:2606.08671](https://arxiv.org/abs/2606.08671)
  65. [BibTeX](docs/skillhone.bib) · [CITATION.cff](CITATION.cff)
  66. ## Acknowledgements
  67. SkillHone uses [DeepSeek Harness](https://github.com/deepseek-ai/deepseek-harness)
  68. as an optional managed repair executor. Native assignments use the current
  69. Agent instead. Both paths keep test-gated local PRs, evidence, and decision
  70. history in SkillHone; installing DSH is not required for native repair.
  71. [Security](SECURITY.md) · [Contributing](CONTRIBUTING.md) ·
  72. [MIT License](LICENSE) · [Third-party notices](THIRD_PARTY_NOTICES.md)

Ratings and comments

Sign in to rate this skill.

Sign in

Free to use. If it’s helped, pass it on.