The Agent Skills Directory

[COMMAND_EXECUTION]: The relay.mjs script executes the pi and git command-line tools. On Windows systems, pi is executed through a shell wrapper (shell:true) to resolve the .cmd shim.
[INDIRECT_PROMPT_INJECTION]: The pi agent is designed to automatically ingest context files from the repository (such as AGENTS.md or CLAUDE.md). This creates a surface where malicious instructions embedded in the codebase could influence the agent's behavior.
Ingestion points: Auto-loading of AGENTS.md and CLAUDE.md from the workspace and parent directories.
Boundary markers: The skill emphasizes using task-specific briefs but does not implement delimiters for the auto-loaded context files.
Capability inventory: The delegated agent (pi) has the capability to modify the filesystem and execute shell commands with host permissions.
Sanitization: No sanitization is performed on the ingested repository content before it is processed by the agent.
[EXTERNAL_DOWNLOADS]: The skill instructions require the manual installation of the @earendil-works/pi-coding-agent package from the NPM registry.
[DYNAMIC_EXECUTION]: The relay.mjs script dynamically locates and executes a resolution script (lane.mjs) from a relative path if the --lane feature is utilized.

pi-delegate