[REMOTE_CODE_EXECUTION]: The README.md file instructs users to install the Lean toolchain manager via a piped shell command from https://elan.lean-lang.org/elan-init.sh. This is the official and standard installation method for the Lean language environment.
[EXTERNAL_DOWNLOADS]: The Lean project configuration in lean/lake-manifest.json specifies dependencies from the official leanprover-community and leanprover GitHub organizations, which are well-known and trusted sources within the formal methods community.
[COMMAND_EXECUTION]: The skill requires the agent to execute lake build to verify Lean 4 proofs. This command is scoped to the mathematical verification tasks central to the skill's purpose.
[INDIRECT_PROMPT_INJECTION]: The skill ingests user-provided mathematical conjectures and reads from a math-journal.md file, creating a surface for indirect prompt injection. However, the risk is mitigated by strict voice guidelines and a structured 'covenant' that prevents the agent from following arbitrary instructions embedded in the math data.
Ingestion points: User-provided conjectures in the chat and the math-journal.md file in the working directory.
Boundary markers: The agent is guided by a 'covenant' that requires falsifiable statements and a 'hint ladder' which restricts the agent's interaction patterns.
Capability inventory: The agent can perform file system operations (read/write journal) and execute proof verification commands (lake build).
Sanitization: The skill enforces ASD-STE100 technical English and explicitly forbids conversational filler or references to other AI products, limiting the effectiveness of potential injection attacks.

math-schema