The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructions in SKILL.md direct the agent to execute shell commands such as lake build, lake env lean, and rg to validate Lean code and search the reference knowledge base.
[PROMPT_INJECTION]: The skill processes untrusted external data in the form of .lean source files and compiler diagnostics, creating a surface for indirect prompt injection. 1. Ingestion points: Local .lean source files, build configuration files (lakefile.lean, lakefile.toml), and diagnostic messages from the Lean compiler. 2. Boundary markers: The instructions do not define explicit delimiters or warnings to ignore instructions embedded in the processed code. 3. Capability inventory: The agent can execute lake and lean commands, which provides a path for untrusted code to be executed or for the agent to be manipulated into performing unintended searches. 4. Sanitization: No specific sanitization or validation of input files is described beyond standard Lean elaboration.
[EXTERNAL_DOWNLOADS]: The documentation references official Lean 4 resources and repositories from organizations including leanprover, leanprover-community, and microsoft. These references target well-known and established services within the software development and formal methods community.

lean4