Skip to content
Agentmart
Submit a tool

codex-consult

Consult or delegate to Codex (GPT) via codex exec.

Skill by rcedwards

IndexedMIT
Write a review

Install

Works with Claude Code, Claude apps, Codex, Cursor, Gemini CLI, GitHub Copilot, VS Code, OpenCode, Goose, Windsurf. Read the repository before you give any tool access to your data.

npx agentmarket install skill/rcedwards-codex-consult

Asks which of your tools to add it to (Claude Code, Codex, Cursor, VS Code, Windsurf, Gemini CLI, Claude Desktop) and shows the change first. See https://agentmart.dev/cli.

Get the weekly diff

The weekly diff: one email a week with new agent tools, notable updates and grade changes, plus security alerts for tools you follow. Double opt-in, one-click unsubscribe.

We use your email only for this newsletter. Privacy

About

Consult or delegate to Codex (GPT) via codex exec. Two modes: (1) Consult -- get a second opinion on a plan, solution, architecture decision, or problem you're stuck on. (2) Delegate -- hand off independent work (writing tests, implementing a component, code review) to Codex in the background while you continue working. Use when the user says "ask Codex", "second opinion", "what does GPT think", "validate this plan", "cross-check this", "consult Codex", "delegate to Codex", "have Codex write tests", "get another take", or similar. Consider self-triggering when you've been going back and forth on an approach without progress, when a decision has significant consequences and an independent review would reduce risk, or when you have independent work items that Codex could handle in parallel.

Use cases: frontend, writing, agent-building

No reviews yet. Installed it? Reviews come from verified installers only. How reviews work

Permissions

What this skill can do and what it asks for, from its SKILL.md. Checked Oct 25, 2026.

Tools
No tools found
Secrets
None asked for
Install scripts
None found

Network hosts it talks to 1

  • codeload.github.com · found in install command

Files it touches

  • ~/.claude/skills/codex-consult

What's inside

Skill name
codex-consult
Description
Consult or delegate to Codex (GPT) via codex exec. Two modes: (1) Consult -- get a second opinion on a plan, solution, architecture decision, or problem you're stuck on. (2) Delegate -- hand off independent work (writing tests, implementing a component, code review) to Codex in the background while you continue working. Use when the user says "ask Codex", "second opinion", "what does GPT think", "validate this plan", "cross-check this", "consult Codex", "delegate to Codex", "have Codex write tests", "get another take", or similar. Consider self-triggering when you've been going back and forth on an approach without progress, when a decision has significant consequences and an independent review would reduce risk, or when you have independent work items that Codex could handle in parallel.
License
MIT
Instructions
388 lines in SKILL.md · view on GitHub

Files (1)

  • SKILL.md

SKILL.md (first lines)

# Codex Consult

Call Codex (GPT) non-interactively from Claude Code. Two modes of operation:

- **Consult**: Get a second opinion. Codex runs in explicit read-only mode (`-s read-only`). You wait for
  the response, synthesize it with your own perspective, and present both to the user.
- **Delegate**: Hand off independent work. Codex runs with `-s workspace-write`. You
  run it in the background and continue working. You review the output before reporting it done.

## Choosing the right mode

### Consult when:

- You want to validate an implementation plan before executing
- You want an independent perspective on a solution or architecture decision
- You're stuck on a problem and want a fresh approach from a different model
- You're weighing multiple approaches and want another perspective

### Delegate when:

- Writing tests for code you just wrote
- Implementing a component while you work on another
- Code review of completed work items
- Any independent task that doesn't depend on your in-progress work

### Do NOT delegate when:

- The task depends on something you're currently building (it will work against stale state)
- Multiple agents would edit the same file (merge conflicts are inevitable)
- The task requires your current conversation context (Codex starts cold)

## Codex basics

Codex reads `AGENTS.md` files automatically for project instructions (`~/.codex/AGENTS.md` for
global defaults, `<repo>/AGENTS.md` for project-specific). This is similar to CLAUDE.md for
Claude Code.

Sandbox modes in non-interactive (`codex exec`) mode:

| Mode | Flag | Behavior |
|------|------|----------|
| Read-only | `-s read-only` | Reads and read-only commands; writes blocked |
| Workspace-write | `-s workspace-write` | Allows workspace edits; approvals follow configuration |

Use explicit read-only for consultations, `-s workspace-write` for delegations.

Progress output goes to stderr, final result to stdout. The `-o` flag captures the final message
to a file for clean reading.

---

## Consult Mode

### Constructing the prompt

Every consultation prompt needs three parts:

**1. Role framing** -- Tell Codex it's acting as a consultant/reviewer.

**2. Context** -- Include relevant context directly in the prompt. Codex has filesystem access,
but providing key information upfront saves time and focuses the analysis. Include:
- The plan, solution, or problem description
- Relevant file paths to examine (Codex cannot use `@` file references in exec mode)
- Constraints or requirements the user has mentioned
- What you've already tried (if getting unstuck)

**3. Specific ask** -- Be precise about what feedback you want. Vague prompts get vague answers.

**Good**: "Review this migration plan. Specifically: (1) will the schema change be safe under
concurrent writes? (2) should any steps be reordered? (3) what's the rollback path if step 3 fails?"

**Bad**: "What do you think of this plan?"

### Running the consultation

Create a unique temp file, then run Codex in read-only mode (`-s read-only`):

```bash
out="$(mktemp /tmp/codex-consult.XXXXXX)"
cat <<'PROMPT' | codex exec -s read-only --ephemeral -o "$out" -
Your prompt here.
PROMPT
```

If you are NOT inside a git repository (e.g., reviewing files in ~/.claude/), add
`--skip-git-repo-check`.

To set a specific working directory (useful when running from a different context):

```bash
cat <<'PROMPT' | codex exec -s read-only --ephemeral -C /path/to/project -o "$out" -
Your prompt here.
PROMPT
```

To attach screenshots or images for UI-related consultations:

```bash
codex exec -s read-only --ephemeral -i screenshot.png -o "$out" "Review this UI and suggest improvements"
```

Flags:
- `-s read-only`: Explicitly block writes while allowing read-only commands. Do not rely on the
  configured default sandbox mode for consultations.
- `--ephemeral`: Don't persist the session (one-shot consultation).
- `-C <path>`: Set working directory explicitly.
- `-i <path>`: Attach image(s) for visual context.
- `--skip-git-repo-check`: Required when running outside a git repository.
- `-o "$out"`: Capture Codex's final response for clean reading.

Set the Bash tool timeout to 300000ms. Consultations typically take 30-120 seconds, but complex
codebase questions where Codex needs to read several files can take longer.

### Presenting consultation results

After Codex responds:

1. Read the output from the temp file, then delete it
2. Synthesize both perspectives. Don't just parrot Codex's response. Add your own analysis.

From rcedwards/blaude-bode, MIT.

More from rcedwards/blaude-bode

Signals

Stars
0
Forks
0
Open issues
0
Last commit (this folder)
Oct 16, 2026 (13 days ago)

From GitHub (fetched Oct 24, 2026).

Details

License
MIT
First seen
Oct 24, 2026
Last checked
Oct 24, 2026

More from rcedwards

Cite this page

codex-consult skill. Agent Mart. https://agentmart.dev/skills/rcedwards-codex-consult (last updated 2026-10-24)