The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes the repository's codebase to identify verification targets and properties, which allows potentially malicious external data to enter the agent's context.
Ingestion points: The target repository's codebase and specific bug reports provided by the user.
Boundary markers: No specific delimiters or safety instructions are defined to distinguish between source code and embedded natural language instructions.
Capability inventory: The skill can execute shell commands, install software, generate new agent skills, and perform temporary code mutations.
Sanitization: There is no evidence of sanitization or filtering of the codebase content before it is used to influence the agent's logic for skill generation.
[COMMAND_EXECUTION]: The instructions direct the agent to select, install, and run third-party formal verification tools.
The skill specifies that the agent should "Follow native permissions for tool installation and external access" and mentions tools such as TLA+, SMT-backed verifiers, and proof assistants.
Orchestration involves running commands to execute these tools and evaluate their output within the repository environment.
[DYNAMIC_EXECUTION]: The skill's primary function is to generate new, persistent instructions in the form of project-local skills stored in .agents/skills/.
It requires the agent to perform "scratch mutations" (temporary code changes) to verify that the formal check can successfully fail when a property is broken.
The generation of executable agent instructions from analyzed codebase data constitutes dynamic context creation.

cursor-create-formal-verification-skill