selftune

Self-improving skills for AI agents.

Your agent skills learn how you work. Detect what's broken. Improve low-risk skill behavior automatically.

Install · Use Cases · How It Works · Commands · Platforms · Docs

---

Your skills do not understand how you talk. You say "make me a slide deck" and nothing happens: no error, no signal, no clue why the right skill never fired. selftune reads the transcripts and telemetry your agent already saves, learns how you actually speak, and improves skill descriptions to match. It validates changes before deployment, watches for regressions after, and rolls back when needed.

Built for Claude Code. Also works with Codex, OpenCode, and OpenClaw. Zero runtime dependencies.

Install

npx skills add selftune-dev/selftune

Then tell your agent: "initialize selftune"

Two minutes. No API keys. No external services. No configuration ceremony. Uses your existing agent subscription.

Quick proof path:

npx selftune@latest doctor
npx selftune@latest sync
npx selftune@latest status
npx selftune@latest dashboard

Use --force only when you explicitly need to rebuild local state from scratch.

Autonomy quick start:

npx selftune@latest init --enable-autonomy
npx selftune@latest orchestrate --dry-run
npx selftune@latest schedule --install --dry-run

CLI only (no installed skill):

npx selftune@latest doctor

Before / After

selftune learned that real users say "slides", "deck", "presentation for Monday" — none of which matched the original skill description. It rewrote the description to match how people actually talk. Validated against the eval set. Deployed with a backup. Done.

Built for How You Actually Work

I write and use my own skills — You built skills for your workflow but your descriptions don't match how you actually talk. selftune learns your language from real sessions and evolves descriptions to match — no more manual tuning. selftune status · selftune evolve · selftune baseline

I publish skills others install — Your skill works for you, but every user talks differently. selftune ships skills that get better for every user automatically — adapting descriptions to how each person actually works. selftune status · selftune evals · selftune badge

I manage an agent setup with many skills — You have 15+ skills installed. Some work. Some chain together. Some conflict. selftune shows which combinations repeat, which ones help, and where the friction is. selftune dashboard · selftune composability · selftune workflows

How It Works

A continuous feedback loop that makes your skills learn and adapt from real work.

Observe — selftune reads the transcripts and telemetry your agents already save. On Claude Code, hooks can add low-latency hints, but transcripts and logs are the source of truth. Use selftune sync to ingest current activity and selftune replay to backfill older Claude Code sessions.

Detect — selftune finds the gap between how you talk and how your skills are described. It spots missed triggers, underperforming descriptions, noisy environments, and regressions in real usage.

Evolve — For low-risk changes, selftune can autonomously rewrite skill descriptions to match how you actually work. Every proposal is validated before deploy. Full skill-body or routing changes stay available for higher-touch workflows.

Watch — After deploying changes, selftune monitors trigger quality and post-deploy evidence. If something regresses, it can roll back automatically. The goal is autonomous improvement with safeguards, not blind self-editing.

What's New in v0.2.x

• Source-truth sync — selftune sync now leads the product loop, using transcripts/logs as truth and hooks as hints
• SQLite-backed local app — selftune dashboard now serves the React SPA by default with faster overview/report routes on top of materialized local data
• Autonomous low-risk evolution — description evolution is autonomous by default, with explicit review-required mode for stricter policies
• Autonomous scheduling — selftune init --enable-autonomy and selftune schedule --install make the orchestrated loop the default recurring runtime
• Full skill body evolution — evolve routing tables and entire skill bodies using teacher-student model with structural, trigger, and quality gates
• Synthetic eval generation — selftune evals --synthetic generates eval sets from SKILL.md for cold-start skills
• Cheap-loop evolution — selftune evolve --cheap-loop uses haiku for proposal generation and validation, sonnet only for the final deployment gate
• Per-stage model control — --validation-model, --proposal-model, and --gate-model give fine-grained control over each evolution stage
• Sandbox test harness — automated coverage, including devcontainer-based LLM testing
• Workflow discovery + codification — selftune workflows finds repeated multi-skill sequences from telemetry and can append them to ## Workflows in SKILL.md

Commands

Command	What it does
selftune doctor	Health check: logs, config, permissions, dashboard build/runtime expectations
selftune sync	Ingest source-truth activity from supported agents and rebuild local state
selftune status	See which skills are undertriggering and why
selftune dashboard	Open the React SPA dashboard (SQLite-backed)
selftune orchestrate	Run the core loop: sync, inspect candidates, evolve, and watch
selftune schedule --install	Install platform-native scheduling for the autonomous loop
selftune evals --skill <name>	Generate eval sets from real session data (--synthetic for cold-start)
selftune evolve --skill <name>	Propose, validate, and deploy improved descriptions (--cheap-loop, --with-baseline)
selftune evolve-body --skill <name>	Evolve full skill body or routing table (teacher-student, 3-gate validation)
selftune watch --skill <name>	Monitor after deploy. Auto-rollback on regression.
selftune replay	Backfill data from existing Claude Code transcripts
selftune baseline --skill <name>	Measure skill value vs no-skill baseline
selftune unit-test --skill <name>	Run or generate skill-level unit tests
selftune composability --skill <name>	Measure synergy and conflicts between co-occurring skills, with workflow-candidate hints
selftune workflows	Discover repeated multi-skill workflows and save a discovered workflow into SKILL.md
selftune import-skillsbench	Import external eval corpus from SkillsBench
selftune badge --skill <name>	Generate skill health badge SVG
selftune cron setup	Optional scheduler helper for OpenClaw-oriented automation

Full command reference: selftune --help

Why Not Just Rewrite Skills Manually?

Approach	Problem
Rewrite the description yourself	No data on how users actually talk. No validation. No regression detection.
Add "ALWAYS invoke when..." directives	Brittle. One agent rewrite away from breaking.
Force-load skills on every prompt	Doesn't fix the description. Expensive band-aid.
selftune	Learns from real usage, rewrites descriptions to match how you work, validates against eval sets, auto-rollbacks on regressions.

Different Layer, Different Problem

Observability tools trace LLM calls. Skill authoring tools help you write skills. Neither knows whether the right skill fired for the right person. selftune does — and fixes it automatically.

Dimension	selftune	Braintrust / Langfuse	skill-creator / SkillForge
Layer	Skill-specific	LLM call / agent trace	Skill authoring
When	Runtime (real sessions)	Runtime (traces)	Authoring time (manual)
Detects	Missed triggers, false negatives, conflicts	Token usage, latency, chain failures	—
Improves	Descriptions, body, routing — automatically	—	Helps you write better manually
Closed loop	Yes — observe → evolve → watch → repeat	No	No
Setup	Zero deps, zero API keys	Self-host or cloud	Included with agent
Price	Free (MIT)	Freemium / Paid	Free

Platforms

Claude Code (primary) — Reads saved transcripts and telemetry directly. Hooks install automatically and add low-latency hints. selftune replay backfills older Claude Code sessions. Full feature support.

Codex — selftune wrap-codex -- <args> or selftune ingest-codex

OpenCode — selftune ingest-opencode

OpenClaw — selftune ingest-openclaw. selftune cron setup remains available as an optional OpenClaw-oriented scheduler helper, but the main product loop is still selftune orchestrate plus generic scheduling.

Requires Bun or Node.js 18+. No extra API keys.

---

Architecture · Contributing · Security · Integration Guide · Sponsor

MIT licensed. Free forever. Built for Claude Code.