{"full_name":"microsoft/SkillOpt","owner":"microsoft","name":"SkillOpt","shard":"1","url":"https://github.com/microsoft/SkillOpt","description":"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.","homepage":"https://aka.ms/skillopt","topics":["agent-skills","self-evolving-agents"],"tag_pages":["agent-skills"],"language":"Python","license":"MIT","avatar":"https://avatars.githubusercontent.com/u/6154722?v=4","created_at":"2026-05-08T06:41:01Z","pushed_at":"2026-11-10T20:10:50Z","metrics":{"stars":19088,"forks":1830,"watchers":67,"issues":116,"prs":295,"releases":2,"issues_open":40,"issues_closed":76,"prs_open":52,"prs_closed":38,"prs_merged":205,"discussions":0},"growth":{"day":{"stars":10,"forks":0,"base_day":"2026-11-15"}},"history":{"dates":["2026-11-10","2026-11-11","2026-11-12","2026-11-13","2026-11-14","2026-11-15","2026-11-16"],"stars":[18967,18992,19025,19038,19066,19078,19088],"forks":[1809,1816,1819,1822,1824,1830,1830],"watchers":[null,66,66,66,66,67,67],"issues_open":[null,30,40,40,40,40,40],"prs_open":[null,43,51,51,52,52,52]},"boards":{},"same_owner":[{"full_name":"microsoft/markitdown","name":"markitdown","shard":"1","boards":{"daily":151,"weekly":176,"monthly":103,"total":36}},{"full_name":"microsoft/vscode","name":"vscode","shard":"1","boards":{"daily_watchers":21,"total":37}},{"full_name":"microsoft/PowerToys","name":"PowerToys","shard":"1","boards":{"daily_watchers":111,"total":81}},{"full_name":"microsoft/generative-ai-for-beginners","name":"generative-ai-for-beginners","shard":"1","boards":{"daily_forks":40,"total":98}},{"full_name":"microsoft/TypeScript","name":"TypeScript","shard":"1","boards":{"daily_watchers":120,"total":120}},{"full_name":"microsoft/terminal","name":"terminal","shard":"1","boards":{"total":133}},{"full_name":"microsoft/playwright","name":"playwright","shard":"1","boards":{"total":143}},{"full_name":"microsoft/Web-Dev-For-Beginners","name":"Web-Dev-For-Beginners","shard":"1","boards":{"total":155}},{"full_name":"microsoft/ML-For-Beginners","name":"ML-For-Beginners","shard":"1","boards":{"total":173}},{"full_name":"microsoft/AI-For-Beginners","name":"AI-For-Beginners","shard":"1","boards":{}},{"full_name":"microsoft/AirSim","name":"AirSim","shard":"1","boards":{}},{"full_name":"microsoft/BitNet","name":"BitNet","shard":"1","boards":{}},{"full_name":"microsoft/Bringing-Old-Photos-Back-to-Life","name":"Bringing-Old-Photos-Back-to-Life","shard":"1","boards":{}},{"full_name":"microsoft/CNTK","name":"CNTK","shard":"1","boards":{}},{"full_name":"microsoft/Data-Science-For-Beginners","name":"Data-Science-For-Beginners","shard":"1","boards":{}},{"full_name":"microsoft/FluidFramework","name":"FluidFramework","shard":"1","boards":{}},{"full_name":"microsoft/IoT-For-Beginners","name":"IoT-For-Beginners","shard":"1","boards":{}},{"full_name":"microsoft/MMdnn","name":"MMdnn","shard":"1","boards":{}},{"full_name":"microsoft/Ontology-Playground","name":"Ontology-Playground","shard":"1","boards":{}},{"full_name":"microsoft/Procexp-Mac","name":"Procexp-Mac","shard":"1","boards":{"daily_rate":96,"weekly_rate":32,"monthly_rate":87}}],"issues":[{"number":132,"title":"Help wanted: integrate SkillOpt with Superpowers for measurable agent efficiency","comments":12,"created_at":"2026-07-13T16:28:05Z"},{"number":154,"title":"Mining is limited to programmatic success checks — intent-level improvements get dropped or Goodharted into shallow proxy rules","comments":11,"created_at":"2026-07-19T04:48:27Z"},{"number":158,"title":"Several reported scores are incompatible with the released test split sizes","comments":7,"created_at":"2026-07-19T09:42:12Z"},{"number":286,"title":"skillopt-sleep Codex harvest ingests its own headless replay sessions","comments":6,"created_at":"2026-09-20T04:32:54Z"},{"number":155,"title":"Proposal: AgenticReplayBackend — replay mined tasks with a real agentic CLI in a throwaway worktree, gate on the diff","comments":6,"created_at":"2026-07-19T04:48:39Z"},{"number":288,"title":"Release cut for the Aug adopt/webui hardening + 2 residual staging gaps","comments":4,"created_at":"2026-09-20T17:36:03Z"},{"number":247,"title":"skillopt-sleep: adopt() writes a backup that nothing can restore — add a `revert` command","comments":4,"created_at":"2026-08-22T21:49:45Z"},{"number":97,"title":"SkillOpt-Sleep: Support harvest from Vscode/ Visual studio copilot sessions","comments":4,"created_at":"2026-07-02T17:18:42Z"}],"discussions":[],"latest_release":{"name":"SkillOpt v0.2.0 — SkillOpt-Sleep","tag":"v0.2.0","published_at":"2026-07-02T14:11:43Z","body":"## [0.2.0] — 2026-07-02\n\nThe headline of this release is **SkillOpt-Sleep**: a nightly offline\nself-evolution engine that harvests a coding agent's real session\ntranscripts, mines recurring tasks, replays them offline, and consolidates\nshort-term experience into long-term memory and skills — all behind the same\nheld-out validation gate that keeps SkillOpt training honest. It ships as a\ndecoupled top-level package (`skillopt_sleep/`, zero dependency on the\nresearch code) and as the new `skillopt-sleep` CLI.\n\n### Added\n- **SkillOpt-Sleep engine** — nightly offline self-evolution cycle\n  (harvest → mine → replay → consolidate) behind a validation gate, exposed\n  as the `skillopt-sleep` console script and `python -m skillopt_sleep`.\n  - Multi-objective reward (accuracy / tokens / latency) with user preferences.\n  - Multi-rollout contrastive reflection under a token/time budget.\n  - Experience replay + controllable dream rollouts (opt-in).\n  - Slow-update long-term memory field (runs even with the gate off).\n  - 3-way train/val/test split with `gate_mode on|off`.\n  - Verifier-discipline validation gate, with a stress-test suite\n    (thanks @Tanmay9223, #87).\n- **Cross-tool backends & plugin shells** for Claude Code, Codex, Copilot,\n  Devin, and OpenClaw:\n  - Codex Desktop transcript harvesting, skill-first Codex integration, and a\n    reviewed task-file flow (thanks @Kirchberg, #48, #49, #60).\n  - GitHub Copilot backend (`CopilotCliBackend`) + research-engine MCP plugin\n    (thanks @Dongbumlee, #50).\n  - Devin plugin: MCP server + ATIF-v1.7 harvest (thanks @xerxes-y, #88).\n  - OpenClaw shell for SkillOpt-Sleep (thanks @Elzlxx, #59).\n- **SearchQA** split materialization helper and fail-fast on systemic rollout\n  failures, with a `searchqa` install extra (thanks @summerview1997,\n  #63, #64, #65).\n- WebUI environment loading and backend preflight (thanks @summerview1997, #63).\n\n### Changed\n- Decoupled the Sleep engine into a standalone top-level `skillopt_sleep/`\n  package with zero dependency on the research code.\n- Made `EnvAdapter.reflect` a shared default so reflect kwargs are no longer\n  dropped (thanks @imshunsuke, #44).\n- English-only pass across the engine, plugins, and docs.\n\n### Fixed\n- Windows robustness for the Claude/Codex backends, plus a hardened JSON\n  fallback path (thanks @Yif-Yang, #79).\n- Reject prose pseudo-JSON wrapped in single quotes/backticks (#82).\n- Surface Codex auth/model/version failures instead of silently scoring 0\n  (thanks @dmmdea, #92).\n- Redact secrets before persisting cycle diagnostics.\n- Configure the `qwen_chat`/`minimax` backends so local LLM endpoints work\n  (thanks @imrehg, #85).\n- Forward the Qwen target timeout and gate `enable_thinking` for vLLM targets\n  (thanks @mvanhorn, #40).\n- Make `--bare` conditional on `ANTHROPIC_API_KEY` (#68), add a\n  `SKILLOPT_SLEEP_PYTHON` override with a lookback-hours first-run fallback\n  (#74), and fix ALFWorld gamefile paths relative to `ALFWORLD_DATA`.\n\n### Packaging\n- Bump `skillopt`, `skillopt.__version__`, and `skillopt_sleep.__version__`\n  to `0.2.0`.\n- Restore `skillopt_webui` to the built wheel (it was dropped when the\n  `packages.find` include list was made explicit).\n- Add the `searchqa` extra and include `json_repair` in the `claude`, `qwen`,\n  and `all` extras.\n\n### Acknowledgements 🙏\nv0.2.0 landed thanks to our community contributors — thank you!\n\n- @Kirchberg — Codex Desktop harvesting, skill-first Codex integration,\n  reviewed task-file flow (#48, #49, #60)\n- @Dongbumlee — GitHub Copilot backend + research-engine MCP plugin (#50)\n- @summerview1997 — SearchQA materialization, rollout fail-fast, WebUI\n  preflight (#63, #64, #65)\n- @xerxes-y — Devin plugin: MCP server + ATIF-v1.7 harvest (#88)\n- @Elzlxx — OpenClaw shell for SkillOpt-Sleep (#59)\n- @imshunsuke — shared `EnvAdapter.reflect` default + docs fixes (#43, #44)\n- @mvanhorn — Qwen timeout forwarding + `enable_thinking` gating (#40)\n- @dmmdea — surface Codex auth/model/version failures (#92","is_prerelease":false},"readme":true,"readme_zh":false,"news":[],"stale":false,"snapshot_at":"2026-11-16T09:23:14Z","history_start":"2026-09-06"}