The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes multiple shell commands to interact with the Codex CLI tool, including version checks (codex --version) and executing/resuming review sessions (codex exec, codex exec resume). These commands are gated by a 10-minute timeout and the shell process is restricted to a read-only sandbox to prevent unintended file modifications by the remote model.
[DATA_EXFILTRATION]: The skill reads project repository files and the implementation plan and transmits them to the OpenAI Codex API for analysis. This is a core functional requirement of the adversarial review process and targets a well-known service provider.
[DATA_EXFILTRATION]: The skill reads the user's local configuration file at ~/.codex/config.toml to identify and display the active model and configuration settings before starting the review loop.
[DATA_EXFILTRATION]: The skill ingests and processes external data in the form of critiques provided by the Codex API.
Ingestion points: The output of the codex exec command is captured into a temporary file (/tmp/codex-verdict.txt) and then read back into the agent's context.
Boundary markers: The agent provides a strict adversarial reviewer prompt to the external model, defining its role and the required verdict format (VERDICT: APPROVED or VERDICT: REVISE).
Capability inventory: The skill possesses shell execution capabilities (via the codex CLI) and file write access (PLAN.md, PLAN-REVIEW-LOG.md).
Sanitization: The skill explicitly instructs the agent to act as the final arbiter of the critique, deciding which points are valid and requiring human approval before proceeding to implementation.

codex-plan-review