The Agent Skills Directory

[DATA_EXFILTRATION]: The skill connects to an external MCP server at https://server.promptwatch.com/mcp to retrieve visibility and sentiment data. This behavior is consistent with the skill's stated purpose and communicates with the vendor's own infrastructure.
[PROMPT_INJECTION]: The skill ingests and processes untrusted data from external AI responses via the list_responses tool. 1. Ingestion points: AI-generated responses are sampled through the list_responses tool in SKILL.md. 2. Boundary markers: The skill does not explicitly define delimiters to isolate sampled responses from the agent's own instructions. 3. Capability inventory: The skill's workflow is focused on analysis and reporting; it lacks high-risk capabilities such as arbitrary command execution, file system writes, or network access beyond the defined vendor API. 4. Sanitization: There is no mention of sanitizing or validating the content of the AI responses before they are analyzed.

visibility-audit