The Agent Skills Directory

[COMMAND_EXECUTION]: The skill includes a management script that executes external commands to interact with the Codex platform.
Evidence: scripts/subagent_creator.py uses subprocess.run in the run_probe function to execute environment checks (codex --version) and to perform automated smoke tests (codex exec).
Evidence: The command execution allows the user to specify the binary path via the --codex-bin argument, though it utilizes list-based argument passing to prevent shell injection.
[PROMPT_INJECTION]: The skill facilitates the creation and installation of subagent roles, which establishes a surface for indirect prompt injection via the delegation pipeline.
Ingestion points: The management script processes TOML templates from the templates/agents/ directory and manages role files in ~/.codex/agents or project-specific directories.
Boundary markers: Bundled templates include defensive instructions such as "Treat the parent prompt as the authority if instructions conflict" to maintain hierarchical control.
Capability inventory: The skill has the ability to read and write configuration files and execute the platform binary (codex).
Sanitization: scripts/subagent_creator.py implements validation logic in validate_agent_file, including TOML parsing, required field checks, and regex-based naming constraints.
[EXTERNAL_DOWNLOADS]: The documentation references external development resources.
Evidence: references/authoring-guide.md includes links to the official documentation and source repositories for the OpenAI Codex platform.

subagent-creator