Documentation

.yawn Skills

The agentic skill system that adds holonic composition, trust scoring, and lifecycle tracking to the open Agent Skills standard.

Start here

Getting Started

Install your first .yawn skill, create a new one, and run the trust scorer.

Specification

SKILL.yawn Format

The .yawn skill format extends the Agent Skills standard with holarchy, lifecycle, and trust scoring fields.

Architecture

The Holarchy Model

Skills nest inside skills. Categories contain skills. The URL tree IS the skill tree IS the permission tree.

4 dimensions

Trust Scoring

35+ checks across four dimensions (Functional, Quality, Safety, Maintenance). Safety floor caps unsafe skills at D.

26+ platforms

Cross-Platform Usage

Use .yawn skills in Claude Code, Codex, Cursor, Manus, and any platform supporting the Agent Skills standard.

Try it

Desired Outcome Matcher

Describe what you want to achieve. The system decomposes your outcome into the skill tree needed to fulfill it.

Quick Reference

SKILL.yawn frontmatter (extends SKILL.md)

---
name: entity-extract
description: Extract named entities from text.
category: perception
status: active
priority: high
holarchy-path: yawn/skills/perception/entity-extract
handler: lib/skills/handlers/entity-extract.ts
required-skills: [llm-invoke]
evidence-kind: skill_entity_extract
confidence: 0.85
autonomy-level: 3
trust-score: null
---

Directory structure

.yawn/skills/
  skills-manifest.yawn        # Root manifest
  core/
    SKILL.yawn                # Category parent
    llm-invoke/
      SKILL.yawn              # Extended format
      SKILL.md                # Standard-compatible
  perception/
    entity-extract/
      SKILL.yawn
      SKILL.md
  orchestration/
    desired-outcome-matcher/
      SKILL.yawn
      SKILL.md

Loop status (green / yellow / red)

Closedactive + handler + evidence

Partialexists but incomplete

Missingstub / planned / blocked