AGENTS.md

Workflow

1. Inspect the repo's current instruction surfaces before writing anything: AGENTS.md, nested AGENTS.md, .github/copilot-instructions.md, .github/instructions/*.instructions.md, CLAUDE.md, and any docs that already define setup or validation commands.
2. Extract only non-inferable, repo-specific facts:
   • install, build, test, lint, and format commands
   • repo entry points and important directories
   • protected paths, approval gates, and secret-handling rules
   • fast vs slow validation expectations
   • common failure and recovery steps
3. Keep the instruction file short and command-first. Link to deeper docs instead of copying architecture or onboarding prose into the instruction file.
4. Scope guidance to the smallest useful boundary. Prefer nested AGENTS.md files when packages or subtrees have different commands, tooling, or safety rules.
5. Write concrete, checkable rules. State when to ask before running expensive, destructive, or networked actions.
6. Validate the documented commands, links, and formatting when feasible.