agent-friendly
Installation
SKILL.md
Agent friendly
The one catalogue of what makes a codebase safe for agents to work in unattended, and the order to build it in. Each item states the rule in a line, the section says why it matters, references/audit-checklist.md says how to check it, references/build-playbook.md says how to install it, and the owner named at the end of each section holds the detail. When an owner and this file disagree, the owner wins and this file is stale.
The build order comes from a production monorepo of dozens of apps written mostly by agents; each step there exists because an agent-written pull request hit the failure it prevents.
- IS: installing the guardrails, instructions and verification in order (Build), explaining which rule applies and why (Explain), scoring a repo across every section with evidence (Audit), and fixing the highest-leverage gap (Apply).
- IS NOT: the detail behind any rule (read its owner), module and interface design (
codebase-architecture), the content of AGENTS.md (agents-md), a product's verify CLI and feature map (app-verification), or CI speed (ci-speedup).
Files
| File | Read when |
|---|---|
references/build-playbook.md |
Build mode, or Apply mode when the gap is a missing step: the ordered steps, each with the incident behind it, its check and its owner |
references/audit-checklist.md |
Audit mode: rule, check and owner per row, and the verdict rules |
references/doctrine.md |
Explain mode: worked examples of each rule in a real repo |
references/diagnostics.md |
Writing a lint, type, build, startup or guard message that names the fix |
references/laws.md |
Adding or reviewing a law or a repo-level promise: fast-check, census with a planted mutant, TLA+, Lean, and their limits |