AGENTS — skill-design

Agent Overview

skill-design is the skill-authoring skill that keeps the skill tree clean, correct, and consolidated. It runs one of three modes per invocation: --create builds a publishable SKILL.md plus its references/ from a described job, --improve audits an existing skill against the shared lint standard and applies the repair edit recipe, and --prune collapses duplicate or same-purpose skills into one folder and archives the dropped duplicates. Agents generating code, docs, or suggestions must only edit files under skills/skill-authoring/skill-design/ and must keep every flag and mode term in lockstep across SKILL.md and its reference files.

Runtime

Project Structure

.
├── PRD.md                      # What the skill does and why
├── ROADMAP.md                  # What ships next, in order
├── README.md                   # How to use/run
├── AGENTS.md                   # This file — how to work here safely
├── CONTRIBUTING.md             # Branch, commit, and PR conventions
├── ARCHITECTURE.md             # Layer contract and linter rule catalog
├── SKILL.md                    # Skill frontmatter + routing table
├── lint_arwaky.config.yaml     # Lint gate configuration
├── template/                   # Canonical SKILL.md and reference templates
├── scripts/                    # Non-trivial logic referenced by SKILL.md
└── modules/skills_design/      # AES layer implementation
    ├── src/                    # agent + capability layers
    ├── shared/src/skills_design/  # shared kernel (contract, taxonomy, utility)
    └── tests/                  # unit / integration / e2e / smoke tests

Commands

Reference