[{"data":1,"prerenderedAt":4},["ShallowReactive",2],{"readme:skillopt":3},"\u003Ch1>SkillOpt: Executive Strategy for Self-Evolving Agent Skills\u003C\u002Fh1>\n\u003Cp>\u003Cem>Train agent skills like you train neural networks — with epochs, (mini-)batchsize, learning rates, and validation gates — but without touching model weights.\u003C\u002Fem>\u003C\u002Fp>\n\u003Cp>\u003Ca href=\"https:\u002F\u002Fmicrosoft.github.io\u002FSkillOpt\u002F\" rel=\"nofollow ugc noopener\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FProject%20Page-SkillOpt-8dbb3c\" alt=\"Project Page\" \u002F>\u003C\u002Fa> \u003Ca href=\"https:\u002F\u002Farxiv.org\u002Fabs\u002F2605.23904\" rel=\"nofollow ugc noopener\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FPaper-arXiv-b31b1b\" alt=\"Paper\" \u002F>\u003C\u002Fa> \u003Ca href=\"https:\u002F\u002Fyoutu.be\u002FJUBMDTCiM0M\" rel=\"nofollow ugc noopener\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FProject%20Video-Watch%20Demo-ff0000\" alt=\"Project Video\" \u002F>\u003C\u002Fa> \u003Ca href=\"https:\u002F\u002Fpypi.org\u002Fproject\u002Fskillopt\u002F\" rel=\"nofollow ugc noopener\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FPyPI-skillopt-green.svg\" alt=\"PyPI\" \u002F>\u003C\u002Fa> \u003Ca href=\"https:\u002F\u002Fwww.python.org\u002F\" rel=\"nofollow ugc noopener\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FPython-3.10%2B-blue.svg\" alt=\"Python 3.10+\" \u002F>\u003C\u002Fa> \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002Fskillopt\u002Fblob\u002FHEAD\u002FLICENSE\" rel=\"nofollow ugc noopener\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FLicense-MIT-yellow.svg\" alt=\"License: MIT\" \u002F>\u003C\u002Fa>\u003C\u002Fp>\n\u003Cp align=\"center\">\n  \u003Ca href=\"https:\u002F\u002Ftrendshift.io\u002Frepositories\u002F38498?utm_source=repository-badge&utm_medium=badge&utm_campaign=badge-repository-38498\" rel=\"nofollow ugc noopener\">\u003Cimg src=\"https:\u002F\u002Ftrendshift.io\u002Fapi\u002Fbadge\u002Frepositories\u002F38498\" alt=\"microsoft%2FSkillOpt | Trendshift\" width=\"250\" height=\"55\" \u002F>\u003C\u002Fa>\n  \u003Ca href=\"https:\u002F\u002Ftrendshift.io\u002Frepositories\u002F38498?utm_source=trendshift-badge&utm_medium=badge&utm_campaign=badge-trendshift-38498\" rel=\"nofollow ugc noopener\">\u003Cimg src=\"https:\u002F\u002Ftrendshift.io\u002Fapi\u002Fbadge\u002Ftrendshift\u002Frepositories\u002F38498\u002Fweekly?language=Python\" alt=\"microsoft%2FSkillOpt | Trendshift\" width=\"250\" height=\"55\" \u002F>\u003C\u002Fa>\n\u003C\u002Fp>\u003Cblockquote>\n\u003Cp>📖 \u003Cstrong>For installation, data preparation, training\u002Feval commands, configuration, and framework internals, start with the versioned \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002FSkillOpt\u002Fblob\u002Fmain\u002Fdocs\u002Findex.md\" rel=\"nofollow ugc noopener\">SkillOpt documentation\u003C\u002Fa>. A concise rendered overview is available in the \u003Ca href=\"https:\u002F\u002Fmicrosoft.github.io\u002FSkillOpt\u002Fdocs\u002Fguideline.html\" rel=\"nofollow ugc noopener\">Documentation & Reproduction Guide\u003C\u002Fa>, and longer-form engineering analysis appears on the \u003Ca href=\"https:\u002F\u002Fmicrosoft.github.io\u002FSkillOpt\u002Fblog\u002F\" rel=\"nofollow ugc noopener\">Technical Blog\u003C\u002Fa>. We also maintain a \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002Fskillopt\u002Fblob\u002FHEAD\u002FCHANGELOG.md\" rel=\"nofollow ugc noopener\">Changelog\u003C\u002Fa> for released and unreleased changes.\u003C\u002Fstrong>\u003C\u002Fp>\n\u003C\u002Fblockquote>\n\u003Chr \u002F>\n\u003Ch2>News 🔥🔥🔥\u003C\u002Fh2>\n\u003Cul>\n\u003Cli>\u003Cstrong>[2026-07-24]\u003C\u002Fstrong> 📰 \u003Cstrong>SkillOpt in the news.\u003C\u002Fstrong> Read the official \u003Ca href=\"https:\u002F\u002Fwww.microsoft.com\u002Fen-us\u002Fresearch\u002Fblog\u002Fskillopt-agent-skills-as-trainable-parameters\u002F\" rel=\"nofollow ugc noopener\">Microsoft Research feature\u003C\u002Fa>, along with recent coverage from \u003Ca href=\"https:\u002F\u002Fventurebeat.com\u002Forchestration\u002Fmicrosofts-open-source-skillopt-automatically-upgrades-ai-agent-skills-without-touching-model-weights\" rel=\"nofollow ugc noopener\">VentureBeat\u003C\u002Fa>, \u003Ca href=\"https:\u002F\u002Fmp.weixin.qq.com\u002Fs\u002FpMlyj3a3KOh8L7cIHClRXA\" rel=\"nofollow ugc noopener\">Synced (机器之心)\u003C\u002Fa>, \u003Ca href=\"https:\u002F\u002Fflowtivity.ai\u002Fblog\u002Fmicrosoft-skillopt-train-ai-agent-skills\u002F\" rel=\"nofollow ugc noopener\">Flowtivity\u003C\u002Fa>, and \u003Ca href=\"https:\u002F\u002Fthe-decoder.com\u002Fmicrosofts-skillopt-boosts-gpt-5-5-by-using-nothing-but-a-trained-markdown-file\u002F\" rel=\"nofollow ugc noopener\">The Decoder\u003C\u002Fa>.\u003C\u002Fli>\n\u003Cli>\u003Cstrong>[2026-07-02]\u003C\u002Fstrong> 🚀 \u003Cstrong>SkillOpt \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002FSkillOpt\u002Freleases\u002Ftag\u002Fv0.2.0\" rel=\"nofollow ugc noopener\">v0.2.0\u003C\u002Fa> is out on \u003Ca href=\"https:\u002F\u002Fpypi.org\u002Fproject\u002Fskillopt\u002F\" rel=\"nofollow ugc noopener\">PyPI\u003C\u002Fa>!\u003C\u002Fstrong> Headline feature: \u003Cstrong>SkillOpt-Sleep\u003C\u002Fstrong>, a nightly offline self-evolution engine (harvest → mine → replay → consolidate behind a held-out validation gate), now shipped as the \u003Ccode>skillopt-sleep\u003C\u002Fcode> CLI. It also includes experimental multi-objective, replay, and dream-rollout controls; the main CLI keeps conservative defaults and does not expose every experiment-harness control as a flag. The release source adds integration shells for \u003Cstrong>Claude Code, Codex, Copilot, and Devin\u003C\u002Fstrong>, plus an \u003Cstrong>OpenClaw reference adaptation\u003C\u002Fstrong>; these plugin\u002FMCP files live in the repository rather than the PyPI wheel. It also adds SearchQA split materialization, Windows robustness, and hardened JSON parsing. See the \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002FSkillOpt\u002Freleases\u002Ftag\u002Fv0.2.0\" rel=\"nofollow ugc noopener\">release notes\u003C\u002Fa> for full release details and contributor acknowledgements.\u003C\u002Fli>\n\u003Cli>\u003Cstrong>[2026-06-15]\u003C\u002Fstrong> 😴 \u003Cstrong>SkillOpt-Sleep (preview)\u003C\u002Fstrong> — a nightly offline self-evolution companion for local coding agents (Claude Code \u002F Codex \u002F Copilot): review past sessions, replay recurring tasks, and consolidate validated skills behind a held-out gate. See **[`docs\u003C\u002Fli>\n\u003C\u002Ful>\n",1791678840426]