SkillOpt
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 is a text-space optimizer that trains reusable natural-language skills for frozen LLM agents through a disciplined training loop inspired by neural network optimization. It treats the skill document itself as trainable state, applying trajectory-driven edits, validation-gated updates, and epoch-wise optimization to produce compact best_skill.md artifacts (300–2,000 tokens) that improve agent performance without changing model weights. The system supports multiple backends (OpenAI, Claude, Azure, Qwen, MiniMax) and execution harnesses (direct chat, Codex CLI, Claude Code CLI), and includes SkillOpt-Sleep, a nightly self-evolution engine for local coding agents. Across six benchmarks and seven target models, optimized skills lift GPT-5.5 average accuracy by +19.1 to +24.8 points depending on harness.
Key Features
Use Cases
- 01Improving accuracy of coding agents like Claude Code or Codex on recurring task patterns
- 02Training domain-specific skills for question-answering benchmarks and specialized workflows
- 03Nightly self-evolution for local coding assistants by replaying and consolidating session learnings
- 04Transferring optimized skills across model scales and between different agent harnesses
- 05Bootstrapping agent capabilities on new benchmarks without fine-tuning model weights
- 06Creating reusable skill artifacts that enhance multiple frozen LLMs in production
Related Skills
View moresuperpowers
An agentic skills framework & software development methodology that works.
skills
Skills for Real Engineers. Straight from my .agents directory.
skills
Public repository for Agent Skills
ponytail
Makes your AI agent think like the laziest senior dev in the room. The best code is the code you never wrote.
SkillOpt — FAQ
What is SkillOpt?+
SkillOpt is a training framework that optimizes natural-language skill documents for frozen LLM agents using epochs, validation gates, and trajectory-driven edits, producing compact best_skill.md files that improve agent performance without modifying model weights.
How do I install SkillOpt?+
Install SkillOpt from PyPI with pip install skillopt (requires Python 3.10 or later). For the nightly self-evolution engine, the skillopt-sleep CLI is included in v0.2.0 and later releases.
Which AI clients and agents does SkillOpt work with?+
SkillOpt integrates with Claude Code, Codex, Copilot, Devin, and Cursor through execution harnesses, and supports direct chat backends for OpenAI, Azure OpenAI, Claude, Qwen, and MiniMax models. Integration shells are included in the repository.
Do I need API keys to use SkillOpt?+
Yes, you need API credentials for the target LLM provider you choose (OpenAI, Azure, Anthropic Claude, Alibaba Qwen, or MiniMax). The optimizer model that generates skill edits also requires API access.
Is SkillOpt free and open source?+
SkillOpt is open source under the MIT License and free to use. However, running training and inference incurs API costs from your chosen LLM provider.
How does SkillOpt-Sleep work for coding agents?+
SkillOpt-Sleep is a nightly offline self-evolution engine that harvests past coding sessions, mines recurring patterns, replays tasks, and consolidates validated skill improvements behind a held-out validation gate. It runs as the skillopt-sleep CLI and is designed for local coding assistants like Claude Code and Copilot.
How do I install SkillOpt?+
Open the source repository on GitHub and follow its README. SkillOpt is a skill — MCP Agents Market links you directly to the official repo.
Is SkillOpt free?+
SkillOpt is an open-source project hosted on GitHub. Check the repository for its license and any usage requirements.