The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill provides instructions to install the @openai/codex package via NPM or Homebrew. This is a reference to an official tool provided by a well-known service provider.
[PROMPT_INJECTION]: The skill identifies an attack surface for indirect prompt injection as it passes user-provided task prompts to an autonomous subagent. 1. Ingestion points: The task prompt string passed as an argument to codex exec in SKILL.md. 2. Boundary markers: Absent; the input is provided as a raw command-line argument. 3. Capability inventory: The skill specifies the workspace-write sandbox, which permits file system modifications. 4. Sanitization: Absent; the skill relies on the user and the underlying model's internal safety filters.
[COMMAND_EXECUTION]: The skill utilizes shell commands to manage the Codex CLI. It promotes operational security by directing the agent to use </dev/null for non-interactive execution and providing instructions to monitor background processes.

codex-subagent