Menu
Consult OpenAI Codex CLI for a second opinion on plans, tests, code review, and more. Use this skill automatically whenever a second perspective would strengthen your output โ especially for architectural plans, code review, test strategy, and debugging.
Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.
Based on SOC occupation classification
Generate images using OpenAI (default) or Gemini. Use when asked to generate, create, or make images. Supports text-to-image and image editing with reference images.
Plans, audits, and implements scalable programmatic SEO systems with URL inventory, intent mapping, metadata, schema, internal linking, sitemaps, ISR/static rendering, and quality gates. Use when the user mentions 'programmatic SEO,' 'pSEO,' '100k pages,' 'generate SEO pages,' 'scaled landing pages,' 'template pages,' 'SEO page factory,' 'dynamic SEO routes,' 'sitemap shards,' 'keyword cannibalization,' 'doorway pages,' or 'thin content at scale.' Does not replace general SEO audits, AI SEO citation work, or llms.txt tasks unless PSEO scale is central.
Use when setting up, deploying, or debugging Nia's Twilio-based channels: voice (phone), SMS, and WhatsApp. All three share one Twilio number, one webhook server, and one set of credentials under channels.twilio. Covers config schema, cloudflared named-tunnel setup, Twilio Console webhook wiring, `nia phone` CLI, `place_call` / `send_message` MCP tools, the WhatsApp Sandbox 24h customer-service window, and the shared TwilioWebhookServer's dedup + rate-limit middleware. Trigger on mentions of "phone", "call", "voice", "sms", "whatsapp", "twilio", "realtime", "media stream", "cloudflared", or when deploying Nia to a new machine and a Twilio surface needs to come up.
Persistent task management via Beads CLI (bd). Use when user mentions tasks, todos, issues, or tracking work. Check `which bd` first โ if missing, offer: `npm install -g @beads/bd`. All commands: run from `$BEATS_DIR` (for example `~/.niahere/beads`) and use `bd <command>`. Always label: `--label project:<project-name>`. Run `bd --help` or `bd help --all` for available commands. Not for ephemeral in-conversation tracking.
QA test a web application using Playwright. Use when asked to "qa", "QA", "test this site", "find bugs", "dogfood", "verify frontend behavior", "check UI rendering", "test user flows", "debug visual issues", "take a snapshot", "take a screenshot", "browser automation", "Playwright MCP", "ref system", or review quality of any web app (local or deployed). Five modes: diff-aware (auto on feature branches), full (systematic exploration), quick (30-second smoke test), single-page (verify one page or component), regression (compare against baseline). Produces structured report with health score, screenshots, and repro steps. Uses Playwright MCP tools or CLI fallback. Includes Playwright MCP reference (tool selection, ref system, token optimization, session management).
Transcribe or subtitle audio/video files using whisper.cpp (whisper-cpp), with a default medium model path and ffmpeg-based extraction.
| name | codex |
| description | Consult OpenAI Codex CLI for a second opinion on plans, tests, code review, and more. Use this skill automatically whenever a second perspective would strengthen your output โ especially for architectural plans, code review, test strategy, and debugging. |
| argument-hint | [task] [details...] |
| allowed-tools | Bash(codex *), Bash(codex exec *) |
Consult the OpenAI Codex CLI as a second-opinion advisor. Use its output to compare, validate, or enrich your own analysis.
User request: $ARGUMENTS
Based on the user's request, determine which mode applies:
Before calling Codex, gather relevant context so you can craft a focused prompt:
git diff --stat or git log --oneline -10 if the request involves recent changesRun Codex via codex exec in non-interactive mode. Always pass -C with the project directory.
For read-only consultation (review, explain, debug):
codex exec -s read-only "YOUR_PROMPT" -C /path/to/project
For code generation (test, plan with output):
codex exec -s workspace-write "YOUR_PROMPT" -C /path/to/project
For code review specifically:
codex exec review -C /path/to/project
| Flag | Description |
|---|---|
-s read-only | Read-only sandbox โ can read files but not modify |
-s workspace-write | Can write to workspace |
-C /path | Set working directory |
-m MODEL | Override model (e.g. -m o3) |
--json | Output events as JSONL |
--ephemeral | Don't persist session files |
-o FILE | Write last message to file |
-c key=value | Override config (e.g. -c model="o3") |
--full-auto | Convenience alias for sandboxed auto-execution |
Do NOT use these (they don't exist):
--quiet โ not a valid flag--approval-mode โ not a valid flagcodex "prompt" for non-interactive โ use codex exec insteadIf the output is too large or Codex needs to examine many files, break it into multiple focused calls.
If codex is not installed, tell the user:
Codex CLI is not installed. Install it with:
npm install -g @openai/codexMake sure yourOPENAI_API_KEYenvironment variable is set.
Then stop.
Do NOT just dump Codex's raw output. Instead:
Format the output as:
Query: (what was asked)
(summarized key points from Codex output)
(where you agree, disagree, or would extend Codex's suggestions)
(any contradictions with existing codebase patterns, CLAUDE.md rules, or project conventions)
(unified recommendation โ the best path forward considering both perspectives)
Ask the user if they want to:
This skill was last verified against codex-cli v0.114.0. If Codex CLI is updated and commands fail, run codex exec --help to check current flags and update this skill accordingly. Common things that change between versions:
exec review was added later)