The Agent Skills Directory

[SAFE]: The skill consists of technical documentation and integration instructions for the Pi Agent Core library. No evidence of prompt injection, data exfiltration, obfuscation, or malicious command execution was found across the SKILL.md or reference files.\n- [EXTERNAL_DOWNLOADS]: The skill references the @earendil-works/pi-agent-core package from the NPM registry at a specific version (0.84.1). This is the primary subject of the skill and is documented as a standard library dependency.\n- [INDIRECT_PROMPT_INJECTION]: The documentation provides guidance on how to manage surfaces related to external data ingestion:\n * Ingestion points: Documented through Agent.prompt(), Agent.steer(), and Agent.followUp() methods in SKILL.md and references/api-surface.md.\n * Boundary markers: The instructions explicitly recommend preserving assistant message boundaries when forwarding multi-message output.\n * Capability inventory: The documented library supports tool execution (parallel/sequential), network proxying (streamProxy), and filesystem storage for sessions.\n * Sanitization: The skill recommends using transformContext and convertToLlm hooks to prune, filter, or map external messages before they are sent to the provider.

pi-agent-integration