What problem does it solve? Delegating coding tasks to an autonomous CLI agent without input validation or output verification risks executing unverified code and silent failures. This Skill enforces a disciplined dispatch workflow for the codex CLI: select a backend protocol first, validate inputs before starting, and only deliver code changes after five verification stages pass. ## Core Features & Use Cases - Protocol-First Dispatch: Requires explicit selection among OpenAI Responses, Anthropic Messages, or OpenAI Chat Completions before invoking the codex CLI. - Input Gating: Refuses to start the agent when task descriptions, file paths, or backend configurations are incomplete or invalid, returning errors with context and recovery guidance. - Five-Stage Verification: Validates input, process, output, boundary, and error handling before delivering any code changes, with a Golden Set of test cases as the single source of truth. - Use Case: Ask the agent to write an F1 evaluation script for a dataset; the Skill selects the protocol, verifies the dataset path exists, runs the codex CLI, and delivers the patch only after all five checks pass. ## Quick Start Use the codex CLI to write an F1 evaluation script for the pima dataset, selecting the OpenAI Responses protocol and verifying all five checks before delivering the patch.
