Use when a known-good manual baseline exists but runtime logs or a modified harness may be misleading; verify rendered artifacts first, trust the baseline over altered instrumentation, and recover by changing one variable at a time.
Use when authoring or validating systems code/scripts/configuration where distro/version compatibility, service semantics, policy composition, and deployment-environment variance can cause generated artifacts to fail.
Use when debugging generated structured configs or deriving language tooling from authoritative corpora; compare against upstream examples, derive strict AST rules, optionally emit EBNF, build parsers, and generate interpreters with runtime-configurable operator and stdlib mappings.
Learn and maintain language-specific coding style preferences from user-authored files, functions, commit ranges, and external style guides; apply those preferences during later feature, bugfix, and refactor work in matching languages.
Use when creating or rewriting commit messages so AI-generated commits are semantically meaningful, evidence-backed, and concise. Enforces a strong subject line and a bounded, context-rich body.