[0.2.0] — 2026-07-02

The headline of this release is SkillOpt-Sleep: a nightly offline
self-evolution engine that harvests a coding agent's real session
transcripts, mines recurring tasks, replays them offline, and consolidates
short-term experience into long-term memory and skills — all behind the same
held-out validation gate that keeps SkillOpt training honest. It ships as a
decoupled top-level package (skillopt_sleep/, zero dependency on the
research code) and as the new skillopt-sleep CLI.

Added

• SkillOpt-Sleep engine — nightly offline self-evolution cycle
  (harvest → mine → replay → consolidate) behind a validation gate, exposed
  as the skillopt-sleep console script and python -m skillopt_sleep.
  • Multi-objective reward (accuracy / tokens / latency) with user preferences.
  • Multi-rollout contrastive reflection under a token/time budget.
  • Experience replay + controllable dream rollouts (opt-in).
  • Slow-update long-term memory field (runs even with the gate off).
  • 3-way train/val/test split with gate_mode on|off.
  • Verifier-discipline validation gate, with a stress-test suite
    (thanks Tanmay Garg (@Tanmay9223), #87).
• Cross-tool backends & plugin shells for Claude Code, Codex, Copilot,
  Devin, and OpenClaw:
  • Codex Desktop transcript harvesting, skill-first Codex integration, and a
    reviewed task-file flow (thanks Kostarev Kirill (@Kirchberg), #48, #49, #60).
  • GitHub Copilot backend (CopilotCliBackend) + research-engine MCP plugin
    (thanks Dongbum(DB) Lee (@Dongbumlee), #50).
  • Devin plugin: MCP server + ATIF-v1.7 harvest (thanks @xerxes-y, #88).
  • OpenClaw shell for SkillOpt-Sleep (thanks Elzlxx, #59).
• SearchQA split materialization helper and fail-fast on systemic rollout
  failures, with a searchqa install extra (thanks summerview1997,
  #63, #64, #65).
• WebUI environment loading and backend preflight (thanks summerview1997, #63).

Changed

• Decoupled the Sleep engine into a standalone top-level skillopt_sleep/
  package with zero dependency on the research code.
• Made EnvAdapter.reflect a shared default so reflect kwargs are no longer
  dropped (thanks Shunsuke (@imshunsuke), #44).
• English-only pass across the engine, plugins, and docs.

Fixed

• Windows robustness for the Claude/Codex backends, plus a hardened JSON
  fallback path (thanks Yifan Yang (@Yif-Yang), #79).
• Reject prose pseudo-JSON wrapped in single quotes/backticks (#82).
• Surface Codex auth/model/version failures instead of silently scoring 0
  (thanks Daniel Martinez Morales (@dmmdea), #92).
• Redact secrets before persisting cycle diagnostics.
• Configure the qwen_chat/minimax backends so local LLM endpoints work
  (thanks Gergely Imreh (@imrehg), #85).
• Forward the Qwen target timeout and gate enable_thinking for vLLM targets
  (thanks Matt Van Horn (@mvanhorn), #40).
• Make --bare conditional on ANTHROPIC_API_KEY (#68), add a
  SKILLOPT_SLEEP_PYTHON override with a lookback-hours first-run fallback
  (#74), and fix ALFWorld gamefile paths relative to ALFWORLD_DATA.

Packaging

• Bump skillopt, skillopt.__version__, and skillopt_sleep.__version__
  to 0.2.0.
• Restore skillopt_webui to the built wheel (it was dropped when the
  packages.find include list was made explicit).
• Add the searchqa extra and include json_repair in the claude, qwen,
  and all extras.

Acknowledgements 🙏

v0.2.0 landed thanks to our community contributors — thank you!

• Kostarev Kirill (@Kirchberg) — Codex Desktop harvesting, skill-first Codex integration,
  reviewed task-file flow (#48, #49, #60)
• Dongbum(DB) Lee (@Dongbumlee) — GitHub Copilot backend + research-engine MCP plugin (#50)
• summerview1997 — SearchQA materialization, rollout fail-fast, WebUI
  preflight (#63, #64, #65)
• @xerxes-y — Devin plugin: MCP server + ATIF-v1.7 harvest (#88)
• Elzlxx — OpenClaw shell for SkillOpt-Sleep (#59)
• Shunsuke (@imshunsuke) — shared EnvAdapter.reflect default + docs fixes (#43, #44)
• Matt Van Horn (@mvanhorn) — Qwen timeout forwarding + enable_thinking gating (#40)
• Daniel Martinez Morales (@dmmdea) — surface Codex auth/model/version failures (#92)
• Tanmay Garg (@Tanmay9223) — verifier-discipline stress test (#87)
• Gergely Imreh (@imrehg) — configure_qwen_chat for local LLM endpoints (#85)
• samuelgoofus-boop — community contributions

Special thanks to Yifan Yang (@Yif-Yang) for driving the SkillOpt-Sleep engine.

Full changelog: v0.1.0...v0.2.0

SkillOpt v0.1.0 — Initial PyPI Release

The first public release of SkillOpt, now available on PyPI.

Install

pip install skillopt

Highlights

• Core training loop: rollout → reflect → aggregate → select → update → evaluate
• Multi-backend support: OpenAI, Azure OpenAI, Claude, Qwen (vLLM)
• Built-in benchmarks: ALFWorld, SpreadsheetBench, SearchQA, DocVQA, OfficeQA, LiveMath
• Hierarchical patch merging and gradient clipping
• YAML-based configuration system
• Optional WebUI dashboard (pip install skillopt[webui])

Links

• 📦 PyPI: https://pypi.org/project/skillopt/0.1.0/
• 📖 Docs: https://microsoft.github.io/SkillOpt