The Agent Skills Directory

[DATA_EXPOSURE]: The skill manages state and configuration files within the user's home directory at ~/.pipa/. Specifically, it interacts with ~/.pipa/profile.md for business facts and ~/.pipa/communication-style.md for presentation rules. This is legitimate application-specific data management.
[EXTERNAL_DOWNLOADS]: The skill can fetch content from user-provided URLs to assist in drafting the business profile. This is limited to information gathering and requires explicit user confirmation of the proposed facts before any data is saved to the local profile.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external websites. It mitigates potential injection risks by:
Ingestion points: Website URLs provided by the user during the setup workflow.
Boundary markers: The skill does not use technical delimiters but implements a procedural boundary by requiring the agent to propose facts for human review.
Capability inventory: File writing to ~/.pipa/ and delegation of tool discovery to pipa-connectors.
Sanitization: The instructions explicitly command the agent to confirm proposed facts with the user before writing and to ignore external presentation rules if they conflict with safety or tool-use gates.
[COMMAND_EXECUTION]: The skill performs file system operations (create directory, write markdown files) within the scoped ~/.pipa/ directory to maintain its operational state.

pipa-setup