Skip to content
Sign in
  • Browse all toolsFilter by category and pricing
  • Browse categoriesWriting, code, image, audio and more
  • Compare toolsPut up to three side by side
  • Read the blogGuides and comparisons

SkillOpt

SkillOpt trains reusable natural-language skills for frozen LLM agents without changing model weights.

SkillOpt is a text-space optimizer that trains reusable natural-language skills for frozen LLM agents through trajectory-driven edits and validation-gated updates. SkillOpt serves agent developers and researchers, produces deployable skill documents, and is available as an open-source MIT-licensed project through PyPI and GitHub.

Updated 9 Nov 20266 sections

SkillOpt website screenshot

What is SkillOpt?

SkillOpt is an open-source optimizer that trains natural-language skill documents for LLM agents while leaving the target model unchanged. SkillOpt treats the document as trainable state, using feedback from scored agent runs to guide revisions. The project is published in the microsoft GitHub repository and distributed under the MIT license.

SkillOpt applies an optimization workflow with epochs, batch sizes, learning-rate budgets, and validation gates to text rather than model weights. SkillOpt produces a compact best_skill.md artifact that can accompany the target agent at deployment. The repository links to versioned documentation, a reproduction guide, a technical blog, and a changelog.

What SkillOpt can do

SkillOpt combines skill editing, evaluation, and deployment preparation in a training loop. SkillOpt uses a separate optimizer model to propose bounded changes informed by agent trajectories. The project also provides monitoring and extension paths, allowing developers to work with supported model providers or implement additional backends and benchmark environments.

  • SkillOpt turns scored rollouts into add, delete, and replace edits.
  • SkillOpt checks candidate edits against held-out validation scores.
  • SkillOpt exports the selected skill as best_skill.md.
  • SkillOpt supports OpenAI-compatible chat backends.
  • SkillOpt provides an optional WebUI monitoring dashboard.
  • SkillOpt-Sleep harvests sessions, mines patterns, replays tasks, and consolidates skills.

How SkillOpt works

SkillOpt follows a rollout, reflect, aggregate, select, update, and evaluate loop. SkillOpt accepts a candidate edit in the default paper-style path only when the edit strictly improves a held-out validation score. A textual learning-rate budget, rejected-edit buffer, and epoch-wise slow or meta update help control how the skill document changes.

SkillOpt can be installed with pip install skillopt, with installation, data preparation, configuration, and training commands covered in the documentation. SkillOpt supports chat and execution backends, including Codex and Claude Code harnesses. The optional WebUI uses a separate installation step, and exposing the dashboard beyond its default loopback host requires authentication.

Who SkillOpt is for

SkillOpt is suited to developers and researchers building or evaluating LLM agents that use reusable skill documents. SkillOpt provides backend and benchmark extension contracts for technical users adapting the optimizer to other environments. SkillOpt-Sleep addresses local coding-agent workflows by reviewing past sessions and validating consolidated skills before accepting updates.

  • SkillOpt supports experiments on reusable skills for frozen target models.
  • SkillOpt supports skill deployment across Codex and Claude Code harnesses.
  • SkillOpt supports custom benchmarks with adapters, loaders, scored rollouts, and YAML configuration.
  • SkillOpt-Sleep supports offline consolidation of recurring coding tasks.

SkillOpt pricing and access

SkillOpt is available as an open-source project under the MIT license, with source code on GitHub and a package on PyPI. SkillOpt documentation provides the installation and execution instructions. Plugin and MCP integration files for coding-agent integrations are supplied in the repository rather than included in the PyPI wheel.

SkillOpt strengths and limitations

SkillOpt separates skill optimization from model-weight training and deploys the resulting document without adding inference-time model calls. SkillOpt also provides validation-gated updates and documented extension paths. Practical constraints include technical installation and configuration, separate repository-based integration files, and experimental controls that are not all available through the main command-line interface.

  • SkillOpt requires installation, data preparation, and configuration for training workflows.
  • SkillOpt keeps plugin and MCP files outside the PyPI wheel.
  • SkillOpt does not expose every experimental harness control as a main CLI flag.
  • SkillOpt requires authentication when the WebUI is exposed beyond its default local host.

SkillOpt FAQ

What is SkillOpt?

SkillOpt is an open-source text-space optimizer that trains reusable natural-language skills for frozen LLM agents. SkillOpt uses scored trajectories, bounded document edits, and validation-gated updates to produce a deployable best_skill.md artifact without changing target model weights.

Is SkillOpt free?

SkillOpt is available as an open-source project under the MIT license, with source code on GitHub and a package on PyPI. SkillOpt can be installed using pip install skillopt, and the repository provides documentation for configuration, training, and evaluation.

Does SkillOpt change model weights?

SkillOpt does not change the target model's weights; SkillOpt trains the natural-language skill document instead. A separate optimizer model proposes bounded edits, and the default paper-style workflow accepts candidate changes only when held-out validation scores strictly improve.

Which model backends does SkillOpt support?

SkillOpt lists support for OpenAI, Azure, Claude, Qwen, and MiniMax, alongside chat and execution backend options. SkillOpt also provides an openai_compatible backend for providers implementing the OpenAI Chat Completions protocol and documentation for adding custom backends.

What does SkillOpt-Sleep do?

SkillOpt-Sleep is a nightly offline self-evolution companion for local coding agents. SkillOpt-Sleep reviews past sessions, replays recurring tasks, and consolidates validated skills behind a held-out validation gate, with the workflow available through the skillopt-sleep CLI.

Does SkillOpt include coding-agent integrations?

SkillOpt provides integration shells for Claude Code, Codex, Copilot, and Devin, plus an OpenClaw reference adaptation. SkillOpt supplies these plugin and MCP files in the repository rather than the PyPI wheel, so package installation alone does not include those files.

From the community

No reviews yet

Be the first to review this tool. What you write helps the next person choose.

More tools like SkillOpt

autoresearch

www.github.com

Open-source autonomous AI experiments for single-GPU LLM training

Free

GenericAgent

www.github.com

Open-source autonomous agent framework for local computer control, browser automation, and reusable task skills.

Free

agent-vision-toolkit

www.github.com

Open-source vision tools and skills for text-only coding agents, including image Q&A, OCR, and GUI automation.

Free

GPT Engineer

www.github.com

Open-source CLI for generating code from natural-language instructions and improving existing projects.

Free