The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to use the Bash tool for build operations, environment checks, and proof validation, including commands like lake build, hostname, and grep.
[REMOTE_CODE_EXECUTION]: The skill contains instructions to execute commands on a remote Mac Mini server using ssh mini if the agent is not running on the designated server.
[DYNAMIC_EXECUTION]: The skill workflow involves writing user-authored Lean source code to a file and executing the lake build system on it. Lean 4 is a powerful language where the compiler and tactic engine can execute arbitrary logic, including I/O, during the formal verification process.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a vulnerability surface by ingesting untrusted data (lemma statements and proofs) and processing it as executable Lean code. 1. Ingestion points: User-supplied lemma and proof text in the SKILL.md workflow. 2. Boundary markers: Absent; instructions focus on fidelity rather than isolation. 3. Capability inventory: Write, Edit, and Bash (via lake build) are used to process and execute the user input. 4. Sanitization: Absent; no validation or filtering is applied to the user-provided code before execution.

lean-check