The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes multiple shell commands including uuidgen, tr, head, rm, and codex exec. These are used to manage local temporary files and interface with the external review CLI tool.
[EXTERNAL_DOWNLOADS]: The skill requires the @openai/codex Node.js package. This is a dependency from a well-known technology organization.
[DATA_EXFILTRATION]: Implementation plans are captured in temporary files (/tmp/claude-plan-...) and passed to the external codex tool. This tool transmits the plan content to external servers for analysis, which is the intended functionality of the skill.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection (Category 8) because it uses the output of an external model to drive its internal logic.
Ingestion points: The skill reads strings such as "VERDICT: APPROVED" and "VERDICT: REVISE" from the /tmp/codex-review-... file generated by the external CLI.
Boundary markers: There are no boundary markers or instructions to ignore embedded commands when the agent reads the external review feedback.
Capability inventory: The agent has the capability to enter plan mode, revise implementation plans, and execute shell commands.
Sanitization: No sanitization or validation is performed on the feedback content before it is used to determine whether the agent should proceed or revise its plan.

codex-review