The Agent Skills Directory

[PROMPT_INJECTION]: The skill documentation and the embedded auditing script contain common injection patterns such as "ignore previous instructions" and "do not reveal". These are explicitly used as reference examples and detection signatures for security auditing purposes, rather than active attempts to override the current agent's behavior.
[EXTERNAL_DOWNLOADS]: The provided Python auditing script utilizes the standard urllib.request library to perform network requests to external MCP endpoints. This is the core functionality intended for scanning and analyzing the safety of remote tools.
[COMMAND_EXECUTION]: Includes a "Runnable Probe" Python script that users are encouraged to run to audit MCP servers. The script is self-contained and uses standard libraries to perform its analysis.
[DATA_EXFILTRATION]: Mentions sensitive file paths including .env, ~/.ssh/id_rsa, and /etc within detection logic and example attack scenarios. These references serve to help auditors identify potential exfiltration targets used by malicious tool developers.
[OBFUSCATION]: The auditing script includes logic to detect the presence of Unicode tag characters (U+E0000–U+E007F) in tool descriptions, identifying them as a technique for hiding malicious instructions from human review while remaining readable by the agent.

hunt-agent-mcp-supply-chain