The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill is designed to generate new, project-specific agent skills by creating directories under .agents/skills/ and symlinks to .claude/skills/. This enables persistence of verification logic within a repository, which is the primary intended function of the skill.
[COMMAND_EXECUTION]: The instructions involve identifying and running formal verification tools (e.g., TLA+, SMT verifiers) on the local codebase to prove properties or identify race conditions. These operations are scoped to the development environment.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted code from the repository to define verification models and properties. Ingestion points: codebase files. Boundary markers: absent. Capability inventory: skill generation, file system writes, and command execution. Sanitization: absent. While this creates a surface where malicious code could attempt to influence the agent's logic or verification properties, the risk is inherent to the tool's purpose and is mitigated by the requirement to manually prove the verification catches real bugs before completion.

create-formal-verification-skill