| name | codex-verifier |
| description | Use OpenAI Codex CLI as a second-opinion agent for verification, exploration, and analysis tasks. Invoke when the user asks to verify assumptions, get a second opinion, cross-check analysis, explore alternatives, or use Codex for read-only analysis. Triggers on phrases like "verify with codex", "get a second opinion", "cross-check", "have codex analyze", "explore with codex". |
Codex Verifier
Delegates verification, analysis, and exploration tasks to OpenAI's Codex CLI as a second-opinion agent.
First: Get Current CLI Syntax
Run this at the start to ensure you have current options:
codex exec --help
Command Pattern
Remember that the prompt should contain all necessary context for the task at hand; filepaths, code snippets, previously run commands, the path to the plan you are currently working on, etc.
Use mktemp to avoid output file collisions. Always redirect stderr to avoid polluting context with intermediate reasoning. Run codex in the background with a 15-minute timeout. Do not stop waiting for codex if it takes a long time.
OUTPUT_FILE=$(mktemp /tmp/codex-XXXXXX)
codex exec --model gpt-5.2 --dangerously-bypass-approvals-and-sandbox -o "$OUTPUT_FILE" "Your verification prompt here" 2>/dev/null
cat "$OUTPUT_FILE"