The Agent Skills Directory

[COMMAND_EXECUTION]: The provided code implementation examples for adding new tools or custom wrappers use asyncio.create_subprocess_exec combined with user-provided arguments. Specifically, the pattern cmd.extend(options.split()) allows a user to pass arbitrary flags and parameters to the underlying system binaries, leading to Argument Injection vulnerabilities.
[EXTERNAL_DOWNLOADS]: The installation instructions require cloning multiple third-party GitHub repositories (spiderfoot, ghunt, blackbird) and installing their dependencies via pip install -r requirements.txt. While these are established OSINT tools, they introduce significant third-party code dependencies into the execution environment.
[PROMPT_INJECTION]: The skill facilitates 'Indirect Prompt Injection' (Category 8) due to the following attack surface:
Ingestion points: The skill reads large volumes of untrusted data from external social media platforms, websites, and search engine results (via tools like sherlock, maigret, and spiderfoot) and feeds it directly into the agent's context.
Boundary markers: No boundary markers or 'ignore embedded instructions' warnings are present in the provided tool response templates (TextContent).
Capability inventory: The agent has the capability to execute shell commands via asyncio.create_subprocess_exec for each of the OSINT tools.
Sanitization: The implementation examples show raw stdout from external tools being decoded and returned to the AI without any filtering or sanitization of potentially malicious instructions embedded in the gathered data.

osint-tools-mcp-server