Documentation Index
Fetch the complete documentation index at: https://docs.selftune.dev/llms.txt
Use this file to discover all available pages before exploring further.
The open standard
Agent Skills is an open format for giving AI agents new capabilities. It defines how skills are structured, discovered, and activated across compatible clients. selftune builds on this standard. Every skill that follows the agent skills spec works with selftune — and selftune makes those skills better by adding observability, grading, and evolution.Compatible agents
Skills built to the spec work across all compatible agents. selftune does not offer the same depth on every runtime, so the practical support matrix is:| Platform | Session capture | LLM-backed judge and evolve | Runtime replay validation | Notes |
|---|---|---|---|---|
| Claude Code | Full hooks + ingest | Yes | Yes | Primary platform with the deepest integration |
| OpenAI Codex | Hooks, wrapper, and ingest | Yes | Experimental | Optimizer agents are inlined into codex exec |
| OpenCode | Plugin hooks + ingest | Yes | Experimental | No prompt-submit hook event and no hard-blocking guards |
| Pi | Extension hooks + ingest | Yes | No | selftune uses pi -p and inlines optimizer instructions because Pi has no native subagent flag |
| OpenClaw | Ingest + cron | No | No | Experimental; no real-time hooks |
| Cline | Partial hooks | No | No | Experimental; post-tool and task lifecycle coverage only |
Where to find skills
skills.sh
skills.sh is the community registry for agent skills. Browse, search, and install skills:Anthropic’s example skills
The anthropics/skills repository contains reference implementations and examples from the team that created the spec.GitHub
Any GitHub repo with a validSKILL.md can be installed directly:
selftune registry
The selftune registry adds contributor signal data on top of the skill — trigger pass rates, evolution history, and contributor insights:How selftune extends the spec
The agent skills spec defines how skills are structured and discovered. selftune adds what happens after discovery:| Agent skills spec | selftune adds |
|---|---|
| Skill structure (SKILL.md, scripts, references) | Observability (session capture, grading) |
| Progressive disclosure (3-tier loading) | Evolution (automatic description improvement) |
| Description-based triggering | Trigger accuracy measurement |
| Compatible client list | Cross-platform session ingestion |
| Manual skill authoring | Continuous improvement from real usage |
Key resources
Specification and reference
Agent Skills Spec
The complete open standard — structure, discovery, activation, scripts, and client implementation.
Spec GitHub Repo
Source for the spec, reference library, and validation tools.
Skills and examples
skills.sh
Community skill registry. Browse, search, and install skills.
Anthropic Example Skills
Reference implementations from the creators of the spec.
selftune
selftune GitHub
Open-source CLI and skill. MIT licensed.
selftune Cloud
Team dashboards, contributor signals, and the skill registry.
Community
Agent Skills Discord
The official community for the agent skills spec.
selftune Discussions
Ask questions, share skills, and discuss evolution strategies.
Validation tools
The spec provides a reference library for validating skills:- Frontmatter fields (name format, description length, required fields)
- Directory structure
- Name-directory match
- Description quality
Next steps
Quickstart
Install selftune and see your first skill health report.
Writing Descriptions
Write descriptions that trigger reliably across all compatible agents.
Publishing Skills
Package and share your skills with the community.
Structuring Skills
Organize skills that scale beyond simple use cases.