← ClaudeAtlas

codexlisted

Use when Claude wants a read-only second opinion from OpenAI Codex CLI on: exploring an unfamiliar codebase, reviewing a plan/design .md, or reviewing a PR diff. Codex runs sandboxed read-only (no writes, no prompts, no network).
YoniChechik/claude-code-config · ★ 0 · AI & Automation · score 65
Install: claude install-skill YoniChechik/claude-code-config
# Codex: Read-Only Second Opinion via OpenAI Codex CLI Delegate read-only review work to the `codex` CLI. Codex gets its own look at the code/plan/diff and reports back. Claude stays in charge of any writes. ## Invocation Template (use verbatim) Every codex invocation MUST write output to a temp file via `-o`. Run the command using `run_in_background=true` on the Bash tool -- Claude gets auto-notified when the process finishes. ```bash CODEX_OUT="/tmp/codex-out-$(date +%s%N).md" codex exec \ --sandbox read-only \ -c approval_policy="never" \ -C "<repo-or-dir>" \ --skip-git-repo-check \ -o "$CODEX_OUT" \ "<review prompt>" ``` Run the above with `run_in_background=true`. When notified of completion, read `$CODEX_OUT` with the Read tool. `-m <model>` overrides the default model -- do NOT set unless the user asks. ## Rules (non-negotiable) - **Always use `-o <file>`** to write output to a file. Never consume codex stdout directly. - **Always use `run_in_background=true`** on the Bash tool call. Never run codex in the foreground. - Sandbox MUST be `read-only`. Never use `--full-auto` (implies workspace-write) or `--dangerously-bypass-approvals-and-sandbox`. - Approval policy goes via `-c approval_policy="never"` -- `codex exec` has no `--ask-for-approval` flag. - `read-only` sandbox **blocks network egress**. Any `gh`/`git fetch`/`curl` must be run by Claude OUTSIDE codex and piped into codex via stdin (or written to a file codex reads). - `--skip-git-repo-che