# selftune

> Skill-level observability for AI agents. selftune watches how your skills perform in real sessions, detects missed triggers, and automatically rewrites descriptions to match how you actually talk. For consumers, it's invisible — install it, forget it. For creators, it's a comparison grid with confidence scores, trigger rates, and crowdsourced evolution from real users. Open source, MIT licensed.

Skills fail silently. You say "make me a slide deck" and nothing happens — no error, no log, no signal. Skill descriptions are written once and frozen, but your language evolves. selftune closes that gap with a continuous observe-detect-evolve loop.

No API keys required. No new subscriptions. Works with your existing agent platform — Claude Code, Codex, OpenCode, OpenClaw, or Pi. Two minutes to install.

## Two Users, One Product

selftune serves two audiences with different information appetites:

- **Skill consumers**: Install it and forget it. A CLI status line tells you what improved. No dashboard. No noise. Your skills get better automatically. If something regresses, it rolls back.
- **Skill creators**: See which skills work, which don't, and why. Comparison grid with confidence scores, trigger rates, and miss patterns. Your users can opt in to share anonymous signals — so your skills improve based on how everyone talks, not just you.

## Core Features

- **Self-improving skills**: selftune rewrites skill descriptions and bodies based on real usage data, validated against eval sets, with automatic rollback on regression
- **Session observation**: hooks into your agent's session lifecycle to capture real usage patterns
- **Missed trigger detection**: identifies when skills should have fired but didn't, with confidence scoring
- **Three information layers**: CLI status line (consumer default), comparison grid (creator default), skill detail view (drill-down)
- **Crowdsourced evolution**: creators add a config file to their skill package; users opt in to share anonymous signals; selftune aggregates across 20+ users and generates evolution proposals
- **Eval generation**: creates eval sets from real user language for validation
- **Auto-rollback**: monitors post-deploy trigger rates and restores originals if anything regresses

## CLI Commands

- `selftune status`: one-line skill health summary — skills monitored, skills improved, trigger rate trend
- `selftune dashboard`: comparison grid with confidence scores, trigger rates, and miss patterns
- `selftune evolve --skill <name>`: propose, validate, and deploy improved descriptions from real usage
- `selftune evals --skill <name>`: generate eval sets from real usage data
- `selftune doctor`: health check for hooks, logs, and configuration
- `selftune init`: initialize selftune and install session hooks
- `selftune replay`: backfill observations from existing transcripts
- `selftune contribute proposals`: view crowdsourced evolution proposals from user signals
- `selftune badge`: generate and update skill health badges

## Pricing

- **Free** ($0/forever): Full CLI, unlimited skills, local dashboard, badges, all platforms, community contribute (opt-in)
- **Pro** ($19/month): Cloud dashboard, 30-day trends, regression alerts, dynamic badge, shareable URLs
- **Team** ($49/month + $15/seat): Contribution pipeline — receive anonymous signals from users, aggregate evolution proposals, multi-user workspaces, CI/CD integration, 90-day history
- **Enterprise** (Custom): Self-hosted relay, SSO/RBAC, audit logs, compliance reporting, unlimited history

## Documentation

- [Getting Started](https://selftune.dev/docs/getting-started): Install selftune and start improving your skills in minutes
- [Skills Discovery](https://selftune.dev/docs/skills-discovery): Agent skills discovery protocol and integration
- [About](https://selftune.dev/about): Problem statement, approach, and competitive landscape
- [For Skill Users](https://selftune.dev/for/users): Consumer-focused overview — install it, forget it
- [For Skill Creators](https://selftune.dev/for/creators): Creator-focused overview — comparison grid, crowdsourced evolution
- [Pricing](https://selftune.dev/pricing): Free forever, scale when you're ready

## Links

- [Homepage](https://selftune.dev): Main landing page with features, how it works, and installation
- [GitHub Repository](https://github.com/selftune-dev/selftune): Source code, issues, and contributions
- [Blog](https://selftune.dev/blog): Updates, guides, and announcements

## Skills Discovery

- [Skills Index](https://selftune.dev/.well-known/skills/index.json): Agent skills discovery endpoint (Cloudflare Agent Skills Discovery RFC)

## Optional

- [Sitemap](https://selftune.dev/sitemap.xml): XML sitemap for all pages