The Agent Skills Directory

[SAFE]: The skill provides a legitimate design pattern for orchestrating multi-session tasks. It includes comprehensive instructions for orientation, documentation, and handoff protocols that prioritize human oversight and clear termination boundaries.
[PROMPT_INJECTION]: The skill is designed to ingest and act upon instructions stored in external markdown files (charter, status, and log). While this creates a potential surface for indirect prompt injection if those files were to be modified by an untrusted source, the skill contains explicit mitigating logic.
Ingestion points: The agent is instructed to read task definitions and current state from charter.md and status.md within the .pi-web/relays/ directory (SKILL.md).
Boundary markers: Absent; the framework relies on the agent's interpretation of structured markdown files as separate 'batons' of state.
Capability inventory: The skill utilizes spawn_session to launch independent agent legs and standard filesystem tools to write state updates (SKILL.md).
Sanitization: No explicit content filtering or escaping is implemented; the system relies on human intervention signals to resolve ambiguity or goal drift.

relay