The Agent Skills Directory

[DATA_EXPOSURE]: The skill accesses local configuration files within the user's home directory at ~/.pipa/communication-style.md and ~/.pipa/CONNECTORS.md. While these are hidden directories, they appear to be standard configuration storage for the 'pipa' tool suite.
[INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection because it processes data from untrusted external platforms.
Ingestion points: The skill reads data from external tools including Slack, Notion, Linear, and Google Drive (as seen in SKILL.md workflow and evals.json).
Boundary markers: The instructions do not define specific delimiters or instructions to ignore embedded commands within the ingested data.
Capability inventory: The agent can perform external writes and contact stakeholders, though these are protected by explicit approval requirements.
Sanitization: No explicit sanitization or validation logic is defined for the external data before it is processed by the model.
[EXTERNAL_DOWNLOADS]: The skill utilizes the composio-mcp framework to connect with well-known third-party services (Notion, Slack, Linear) to retrieve project-related information.

pipa-stakeholder-map