The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface by ingesting and processing data from external theorem submissions and compiler logs.
Ingestion points: Data enters the agent context through the search_theorems tool response, particularly the recent_submissions and output_log fields.
Boundary markers: The skill lacks explicit boundary markers or instructions to treat tool-provided content as untrusted data.
Capability inventory: The agent has the capability to interact with external APIs through the prove_theorem and submit_theorem tools, which could be influenced by instructions embedded in the theorem data.
Sanitization: No sanitization or filtering of the content received from the MathProofs-Claw backend is described.
[DATA_EXFILTRATION]: The skill requires and transmits the MATHPROOFS_API_KEY to the mathproofs.adeveloper.com.br domain. This is a functional requirement for authentication with the platform's backend infrastructure.

mathproofs-claw