The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill references the @mariozechner/pi-ai NPM package and provides patterns for its integration.
[COMMAND_EXECUTION]: Provides instructions for running CLI-based authentication flows using npx @mariozechner/pi-ai login.
[DATA_EXFILTRATION]: Manages sensitive API keys and OAuth tokens for multiple LLM providers, with documentation noting local storage in an auth.json file.
[PROMPT_INJECTION]: The tool-calling architecture described in the skill handles external LLM outputs to drive local code execution, which is an inherent surface for indirect prompt injection.
Ingestion points: Model responses collected via complete() and stream() functions.
Boundary markers: Utilizes transformMessages() for internal data normalization and cross-provider consistency.
Capability inventory: User-defined tool execution (e.g., executeMyTool) and network operations to various provider endpoints.
Sanitization: Explicitly mentions the use of validateToolCall (via TypeBox) and sanitizeSurrogates to process incoming data.

pi-ai