The Agent Skills Directory

[COMMAND_EXECUTION]: The relay.mjs script executes the zcode CLI and git commands via node:child_process. This is the core functionality used to dispatch tasks and monitor the working tree state.
[INDIRECT_PROMPT_INJECTION]: The skill acts as a bridge to an external LLM implementer by passing it task instructions ('briefs'). This presents an attack surface for indirect prompt injection.
Ingestion points: The readBrief function in relay.mjs reads instructions from files or stdin.
Boundary markers: The relay script isolates the brief in a dedicated file and uses a fixed prompt to guide the implementer.
Capability inventory: The implementer (ZCode) can write to the filesystem and execute shell commands. The relay script itself can execute ZCode and Git.
Sanitization: No content sanitization is performed on the brief; instead, the skill mandates a strict human-in-the-loop review of all outputs before they are committed.
[DYNAMIC_EXECUTION]: The relay script dynamically resolves the path to the ZCode CLI using environment variables, flags, or local app installation paths, and executes it using spawn or execSync.

zcode-delegate