katas-path-conditional-rules

Path-conditional loading uses the `paths:` frontmatter of a skill (e.g., `paths: ["src/**/*.py"]`) or a path matcher in a PreToolUse hook. The `## When editing <glob>:` heading inside CLAUDE.md is only an organizational convention, not an automatic runtime loader.

Security policies should be universal: imported directly in the root CLAUDE.md with no glob so they apply to every edit. Language heuristics like style, lints, and testing conventions should be conditional, scoped to globs such as `src/**/*.py` or `*.tf`.

No. That heading is a structural convention for organizing rules, not a runtime mechanism. Automatic activation by path requires skill `paths:` frontmatter, explicit imports via `.claude/rules/`, or a path-aware hook matcher.

Both matching rule sets load, and the more specific subpath wins on point conflicts. For example, if `src/**/*.py` and `src/payments/**/*.py` both match, the deeper payments rule takes precedence in conflicts.

Compare `input_tokens` per turn between a session editing a README and one editing a `.py` file. A README edit should load only universal rules, while the Python edit additionally loads the glob-matched language heuristics.

Avoid them for security policies, which must always load, and for tasks unrelated to repository rule structure, such as writing application code. Also avoid overly broad globs that reload rules on unexpected files.