jnzader/skill-recipes

Purpose

Define a pattern where skills are self-installing: an INSTALL.md contains executable setup instructions that an agent follows to configure dependencies, MCP servers, or config.

Recipe Format

own/skills/my-skill/
  SKILL.md         <- the skill itself
  INSTALL.md       <- optional setup recipe

INSTALL.md Structure

markdown
# Install: {skill-name}

## Prerequisites
- [ ] Node.js >= 20

## Steps
### 1. Install dependencies
\`\`\`bash
pnpm add -D playwright
\`\`\`

### 2. Verify
\`\`\`bash
npx playwright --version
\`\`\`

## Uninstall
\`\`\`bash
pnpm remove playwright
\`\`\`

Agent Execution Protocol

1. Read INSTALL.md completely before executing
2. Check Prerequisites — verify each is met, stop if not
3. Execute Steps sequentially (Bash for commands, Write for files)
4. Run Verify step to confirm success
5. Report results

Safety Rules

1. NEVER modify files outside project directory without confirmation
2. NEVER install global packages without confirmation
3. If a step fails, STOP — do not continue
4. Show commands to user BEFORE executing

Critical Rules

1. INSTALL.md is OPTIONAL — skills work without it
2. Recipes MUST be idempotent — running twice must not break anything
3. Every recipe MUST have a Verify step
4. Recipes MUST NOT contain secrets or API keys
5. Agents MUST show what will execute before running