The Agent Skills Directory

[SAFE]: The skill implements a secure workflow for rule management with multiple validation layers. It includes a classifier that rejects procedural or one-off instructions, ensuring that only declarative coding conventions are written to the rule directory.
[SAFE]: File system operations are performed with strict user oversight. The deletion workflow displays the full content of the target file and requires explicit confirmation before removal. Filenames and glob patterns are sanitized and validated to prevent path manipulation or unintended shell expansions.
[SAFE]: The 'extract' functionality, which could serve as an indirect prompt injection surface by ingesting content from CLAUDE.md, is mitigated by a mandatory verification process where the user must approve or amend every extraction verdict before files are modified.
[SAFE]: The skill performs context-aware checks by reading project manifests like package.json or pyproject.toml to ensure new rules align with the detected technology stack, preventing the introduction of conflicting or irrelevant guidelines.

rule-creator