The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill is designed to programmatically generate and write new project-local agent skills to the .agents/skills/ directory. These generated skills contain executable instructions that the agent will follow in subsequent sessions.
[EXTERNAL_DOWNLOADS]: The instructions direct the agent to select and install external formal verification tools (such as TLA+, SMT verifiers, or proof assistants). This involves downloading and executing software from the internet that is not pre-defined in the skill.
[COMMAND_EXECUTION]: The skill performs several high-privilege shell operations, including installing software, running verification harnesses, and committing changes to the repository if authorized.
[INDIRECT_PROMPT_INJECTION]: The skill exhibits a significant attack surface for indirect prompt injection.
Ingestion points: It reads repository instructions, the active skill catalog, and the entire codebase to decide how to generate the verification logic.
Boundary markers: There are no instructions to use delimiters or ignore potentially malicious instructions embedded in the analyzed repository files.
Capability inventory: The agent has the ability to write files (creating new skills), install software, and execute shell commands.
Sanitization: The skill lacks explicit sanitization or validation of the data ingested from the repository before using it to generate new skill content or command arguments.

create-formal-verification-skill