The Agent Skills Directory

[COMMAND_EXECUTION]: The helper script scripts/relay.mjs executes system commands including codex, git, and taskkill. On Windows systems, it uses the shell: true option to properly resolve npm command shims. The script includes regex validation for critical arguments like session IDs and reasoning effort levels.
[EXTERNAL_DOWNLOADS]: The skill's documentation and prerequisites require the installation of the @openai/codex CLI from the npm registry. This is a standard dependency from a well-known service and is necessary for the skill's core functionality.
[PROMPT_INJECTION]: The skill processes repository data to provide context for the delegated task, creating a surface for indirect prompt injection.
Ingestion points: SKILL.md and references/writing-the-brief.md instruct the agent to read repository configuration files (e.g., Makefile, package.json) to define task boundaries and gate commands.
Boundary markers: The documentation recommends using XML-style delimiters such as <task> and <verification_loop> to structure information for the implementer.
Capability inventory: scripts/relay.mjs executes the codex CLI, which can modify the local filesystem. The skill mitigates risk by requiring orchestrator review and manual commits.
Sanitization: No explicit sanitization of ingested repository content is performed by the script; responsibility for brief construction lies with the orchestrator.

codex-delegate