For skill users
Install it. Forget it.
Your skills get better.
You have dozens of skills installed. You don't want to manage them. selftune watches your real sessions, detects what's drifting, and quietly fixes it. You never open a dashboard.
$ npx skills add selftune-dev/selftune Two minutes to install. Zero API keys. Works with Claude Code, Codex, OpenCode, OpenClaw, and Pi.
What it does
Seen and not heard
Silent improvement
selftune learns from your real sessions. Skills match how you actually talk — not how someone guessed you'd talk when they wrote the description.
Auto-rollback
Every change is validated before deploy. If a skill regresses, selftune rolls it back automatically. Skills only get better, never worse.
Your data stays local
Everything runs on your machine. No cloud. No telemetry. If a skill creator asks you to opt in to anonymous signals, you choose. Nothing leaves without your consent.
Three steps. Then you forget about it.
selftune installs in two minutes and runs silently from there. Check in when you want to — or don't.
Add selftune to your agent
One command. No API keys. No config files. selftune detects your agent platform and starts watching sessions automatically.
- Works with any MCP-compatible agent
- Zero configuration
- Backfills from existing transcripts
Skills evolve from how you work
selftune detects which skills fire and which miss. It rewrites descriptions to match your vocabulary. Every change is validated against your real sessions before deploy.
- Learns your vocabulary
- Validates before deploying
- Backed up automatically
One line tells you everything
'selftune status' shows one line: skills monitored, skills improved, trigger rate trend. If you never go deeper, that's fine — your skills still get better.
- CLI status line
- No dashboard required
- Drill down only if you want to
What you never have to do
Zero maintenance
Open a dashboard
CLI status line has everything you need.
Write eval sets
Generated from your real sessions automatically.
Rewrite skill descriptions
selftune evolves them based on how you actually talk.
Debug skill conflicts
Composability analysis detects overlaps before they cause problems.
Common questions
What if it breaks a skill that was working?
Every change is validated against baseline metrics before deploy. If a skill regresses after evolution, selftune auto-rolls back to the previous version. You can also manually approve or reject any change.
Does it use my data or send anything to the cloud?
Everything runs locally on your machine. No cloud services, no telemetry. If a skill creator has enabled the contribution pipeline, you'll be asked to opt in — and even then, only anonymized trigger signals are shared, never your prompts or session content.
Does it slow down my agent?
No. selftune runs as a post-session hook, not inline with your agent's responses. It analyzes transcripts after they're written. Zero impact on response latency.
How many skills can it handle?
Unlimited. Whether you have 5 skills or 500, selftune monitors all of them. The CLI status line compresses everything into one glanceable view.
Seen and not heard.
Your skills just get better.
Two minutes to install. Zero API keys. Works with your existing agent.
$ npx skills add selftune-dev/selftune MIT licensed. Free forever.
Skill observability updates
New features, skill tuning patterns, and ecosystem news. No spam. Unsubscribe anytime.
Monthly or less