The Agent Skills Directory

[PROMPT_INJECTION]: The skill does not contain any instructions aimed at bypassing AI safety protocols or overriding system prompts. It explicitly defines principles for mathematical honesty, such as acknowledging unsolved problems and avoiding fabricated sources. The 'Adversarial Cases' identified in the test suites refer to mathematical logic traps (like division by zero) rather than security exploits.
[DATA_EXFILTRATION]: There are no commands or patterns indicating the collection or transmission of sensitive user data. The skill does not access sensitive file paths (e.g., SSH, AWS, or .env files). Network references in the 'Search Policy' are confined to authoritative mathematical and scientific resources (e.g., arXiv, Wolfram MathWorld) used for verification purposes.
[REMOTE_CODE_EXECUTION]: No instructions for the agent to download and run untrusted external code were found. The installation command provided in the documentation (npx skills add) is a standard deployment mechanism for the platform and refers to the vendor's own verified package name.
[COMMAND_EXECUTION]: The skill does not utilize the dynamic context injection syntax (!command) or any other means of triggering silent pre-execution shell commands. All operations are designed to be executed within the agent's normal reasoning process.
[CREDENTIALS_UNSAFE]: A review of all 44 files, including modules, examples, and test cases, confirmed the total absence of hardcoded API keys, tokens, or other sensitive credentials.

math-skill