The Agent Skills Directory

[COMMAND_EXECUTION]: The skill requires the agent to identify and execute shell commands for formal verification tools (such as TLA+, proof assistants, or SMT-backed verifiers). It specifically instructs the agent to 'Follow the generated skill once' and 'run the check' to confirm the verification logic works as intended.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze arbitrary content from a codebase, creating a risk if that codebase contains malicious instructions disguised as logic or comments.
Ingestion points: Accesses and interprets project files, functions, and commit history to determine verification targets (referenced in SKILL.md).
Boundary markers: The instructions lack markers or specific guidelines to ignore instructions embedded within the code being analyzed.
Capability inventory: The agent possesses the capability to execute shell commands and write new instruction files (skills) based on the input data.
Sanitization: There is no mention of sanitizing or validating the code-derived data before it is used to generate new prompts or commands.
[DYNAMIC_EXECUTION]: A core component of the skill is the 'Generate the verify skill' step. This involves the agent dynamically constructing a new set of instructions (a project-local skill) including specific shell commands, which are then executed within the agent's environment.

create-formal-verification-skill