The Agent Skills Directory

[SAFE]: The skill instructions are focused on standard software engineering practices for formal methods and verification. It does not initiate unauthorized network requests or access sensitive user credentials.
[INDIRECT_PROMPT_INJECTION]: The skill possesses an ingestion surface as it reads the project codebase to generate new instructions for a project-local skill. The capability inventory includes the ability to write a new skill file. Although explicit boundary markers and sanitization for the generated content are absent, the risk is categorized as safe because the instructions require the agent to validate the generated skill against a 'scratch change' that intentionally breaks the code, ensuring the verification logic is functionally correct regardless of potentially malicious code comments.

create-formal-verification-skill