From saurun
Use when user wants to ask Codex, get GPT's opinion, get a second opinion from OpenAI, compare perspectives, have Codex review code or docs, or let Codex fix/refactor/implement something.
How this skill is triggered — by the user, by Claude, or both
Slash command
/saurun:codex-bridgehaikuThis skill is limited to the following tools:
The summary Claude sees in its skill listing — used to decide when to auto-load this skill
Call OpenAI's Codex CLI to get responses, reviews, or perform tasks.
Call OpenAI's Codex CLI to get responses, reviews, or perform tasks.
Activate when user says:
| Flag | When | Example |
|---|---|---|
| (none) | Opinions, reviews, questions, analysis | node "$CODEX_BRIDGE" "Review this code" --working-dir "<project>" |
--full-auto | Fix, refactor, implement, update files | node "$CODEX_BRIDGE" "Fix errors" --full-auto --working-dir "<project>" |
--model MODEL | Override model (default: Codex default) | node "$CODEX_BRIDGE" "Explain" --model gpt-4o --working-dir "<project>" |
--json | Bridge outputs raw JSON events (debugging) | node "$CODEX_BRIDGE" "Analyze" --json --working-dir "<project>" |
--timeout MS | Override 20-minute default timeout | node "$CODEX_BRIDGE" "Long task" --timeout 2400000 --working-dir "<project>" |
# Resolve bridge path (one command):
CODEX_BRIDGE=$(ls -1 "$HOME/.claude/plugins/cache/saurun-marketplace/saurun/"*/skills/codex-bridge/codex-bridge.mjs 2>/dev/null | head -1)
CRITICAL: Always pass --working-dir with the project directory. Without it, Codex runs in its install location and can't access project files.
| Mode | When |
|---|---|
| Read-only (no flags) | Opinion, explanation, review, comparison, questions |
--full-auto | Modify files. Trigger words: "fix", "refactor", "implement", "update", "change" |
When user wants Codex to review a file:
node "$CODEX_BRIDGE" "Review this document for feasibility, risks, and suggestions:
---
[FILE CONTENT HERE]
---" --working-dir "/path/to/project"
For opinions/reviews:
Codex (OpenAI) says: [response]
For comparisons: Provide your own analysis AND Codex's response, noting differences.
For file modifications: After Codex completes, show changes with git diff.
| Mistake | Fix |
|---|---|
Omitting --working-dir | Codex can't read project files. Always pass it. |
Using --full-auto for reviews | Read-only is safer for opinions. Only use --full-auto when user wants file changes. |
| Sending huge files as CLI args | Bridge handles this automatically (writes to temp file for prompts >7000 chars). |
| Forgetting prerequisites | Codex CLI must be installed (npm install -g @openai/codex) and authenticated (codex login). |
If the bridge fails (exit code non-zero), check:
which codex (install: npm install -g @openai/codex)codex login--timeout MS (default: 1,200,000ms / 20 min)npx claudepluginhub fiatkongen/saurun-marketplace --plugin saurunRuns OpenAI Codex CLI as a subagent for second opinions, code reviews, and questions. Useful when you want a different AI model's perspective.
Consults OpenAI Codex (GPT-5) via CLI for code investigation, debugging, or review. Runs read-only with full project access; activates on 'ask codex' phrases or /ask-codex.
Creates, edits, and optimizes skills for Claude Code, including drafting, evaluating with test prompts, iterating on performance, and improving skill descriptions for better triggering accuracy.