Second opinion

Consult an OpenAI model through the codex CLI and relay what it says. The value of a second opinion is independence: let the other model reach its own conclusion, verify what it claims, then report it faithfully -- especially where it disagrees with you.

Preflight

Check auth once per session: codex login status should print Logged in using ChatGPT (or the API-key equivalent). Anything else: stop and ask the user to authenticate -- suggest they type ! codex login (ChatGPT account, usage billed to their plan) or use codex login --api-key <key>.

Core command

Set up a working directory, write the prompt to a file (avoids shell-quoting pain and keeps a record of what was asked), then run codex in the background:

DIR=<your scratchpad>/second-opinion
mkdir -p "$DIR"
# write $DIR/prompt.md, then, as a background Bash call:
codex exec -C <repo root> -s read-only -o "$DIR/answer.md" - < "$DIR/prompt.md" > "$DIR/codex.log" 2>&1