The Agent Skills Directory

[COMMAND_EXECUTION]: The scripts/relay.mjs script uses the node:child_process module to spawn and interact with the pi CLI and the system git binary. These tools are used to execute delegated tasks, manage sessions, and verify changes in the repository.
[EXTERNAL_DOWNLOADS]: The skill's functionality depends on the @earendil-works/pi-coding-agent package, which must be installed from the NPM registry as a prerequisite.
[PROMPT_INJECTION]: The skill is subject to indirect prompt injection risks because the delegated agent automatically ingests context from files within the workspace (such as AGENTS.md or CLAUDE.md). If these files are part of an untrusted repository, they could contain instructions that influence the agent's behavior.
Ingestion points: Files in the target repository specified by the --cd flag, including specific context files like AGENTS.md and CLAUDE.md which are auto-loaded by the agent.
Boundary markers: The documentation recommends using XML-style tags to structure the task brief, though these markers do not apply to the workspace files the agent inspects independently.
Capability inventory: The delegated agent has the authority to read, write, and edit files, and to execute shell commands within the host environment.
Sanitization: The relay script validates CLI arguments (provider, model, session) against a strict alphanumeric character set to prevent command injection, but does not sanitize the content of the task brief or the workspace files.

pi-delegate