From SKILL.md to 11 AI Coding Agents

Anatomy of a skill and the single-script conversion pipeline

Skill Anatomy Conversion Output Formats # SKILL.md --- name: senior-architect version: 2.1.0 domain: engineering tools: [scripts/*.py] --- # Instructions, workflows, # decision frameworks... # (structured Markdown) scripts/ *.py (stdlib-only, zero pip) CLI tools: --help, --json, --csv references/ Templates, checklists assets/ Domain knowledge Personas (3 built-in) startup-cto, growth-marketer, solo-founder Curated skill loadouts + communication style ./scripts/convert.sh --tool all (15 seconds) Parses SKILL.md frontmatter + body ./scripts/install.sh --tool cursor --target /project skill-security-auditor Scan before install: PASS / WARN / FAIL /plugin marketplace One-command install for Claude Code Claude Code ~/.claude/skills/ OpenAI Codex ~/.codex/skills/ Gemini CLI GEMINI.md Cursor .cursor/rules/*.mdc Aider CONVENTIONS.md Windsurf .windsurf/skills/ Kilo Code .kilocode/rules/ + OpenCode, Augment, Antigravity Native format per tool
Key insight: Every skill is authored once as a SKILL.md with optional Python scripts and references. The conversion script transforms all 204 skills into native formats for each tool in about 15 seconds. Zero manual adaptation needed.