skills
anthropics
Public repository for Agent Skills
microsoft/SkillOpt
SkillOpt is a tool for DeepSeek Harness. SkillOpt is a text-space optimizer that trains reusable natural-language skills for frozen LLM agents through trajectory-driven edits, validation-gated updates, and deployable best_skill.md artifacts.

REVIEW SCORECARD
✓ 无Risky检出 · 自动化扫描结果仅供参考,非官方背书
AI DEEP REVIEW
A well-documented, actively maintained Microsoft optimizer that trains reusable agent skills without touching model weig
README references versioned docs, a changelog, PyPI releases, and a structured training loop (rollout→reflect→aggregate→select→update→evaluate), but no CI or test details are visible in the excerpt.
No plaintext secrets, download-and-execute instructions, or excessive permissions appear; install is a standard pip package, though the optimizer calls external LLM backends and the excerpt gives no deeper security detail.
It addresses a real pain point by turning hand-crafted or one-shot agent skills into validation-gated, reusable best_skill.md artifacts with zero inference-time model calls at deployment.
Last updated 2026-09-05 with v0.2.0 released 2026-07-02, an active changelog, news entries, and third-party integrations, indicating ongoing development.
The README links to versioned docs, a rendered guideline, a technical blog, a changelog, and release notes, though the excerpt itself only summarizes rather than showing full install/usage steps.
Generated by AI after reading the project README, as a decision aid; neutral scores are given when information is thin. Not an official endorsement.
正在读取 GitHub README...
README 暂时无法读取。
前往仓库INSTALL REFERENCE
pip install skilloptThis command comes from an explicit install reference in the README and is provided for reference only — it is never auto-executed by a DSH host.
PROJECT TOPICS
VALIDATION LADDER
FAQ
SkillOpt is a text-space optimizer that trains reusable natural-language skills for frozen LLM agents through trajectory-driven edits, validation-gated updates, and deployable best_skill.md artifacts.
SkillOpt has 16,916 stars and 1,588 forks on GitHub, last updated 2026-09-05.
SkillOpt is distributed under the MIT license.
SkillOpt is listed in the DSH Universe directory as a tool for DeepSeek Harness. This plugin is listed in the DSH Universe directory and covered by its validation pipeline.
CLASSIFICATION EVIDENCE
System reads GitHub Topics first, then compares against the in-site category dictionary and word-root rules. Current match:agent-skills。