The Agent Skills Directory

[COMMAND_EXECUTION]: The skill leverages the codex exec command for automated code implementation and exploration. This includes instructions for using high-capability flags such as --dangerously-bypass-approvals-and-sandbox to allow for non-interactive execution and workspace modifications.
[EXTERNAL_DOWNLOADS]: The documentation instructs users to install the @openai/codex CLI tool from the npm registry. This dependency is a core component of the skill and is provided by a well-known service (OpenAI).
[PROMPT_INJECTION]: The skill's orchestration architecture creates an indirect prompt injection surface where a manager agent ingests and acts upon reports generated by subagents.
Ingestion points: Subagent reports (labeled with CHANGES, VERIFIED, GAPS, OBJECTIONS) and results stored in findings/*.md are incorporated into the primary session context.
Boundary markers: The skill employs structured reporting headers and explicit instructions for delegates to object to conflicting specifications rather than silently complying.
Capability inventory: The system possesses high-impact capabilities including filesystem write access through codex exec and shell command execution through the codex-worker agent.
Sanitization: The skill mitigates risks through a mandatory 'Verify' phase, requiring the orchestrator to perform independent diff reviews and manually re-run validation tests rather than trusting subagent reports as ground truth.
[SAFE]: The skill includes a local setup script (scripts/setup-roster.sh) that automates the installation of subagent definition files. The script is limited to standard, benign file system operations.

delegate