How do I install the Codex Pipeline skill?

Run npx skillmds@latest add v-songbird/codex-pipeline in your terminal (requires Node.js), paste this page's agent-chat prompt into Claude, Cursor, or any MCP-connected agent, or download the SKILL.md file and copy it into your agent's skills directory.

Is Codex Pipeline safe to use?

This skill has not completed SkillMD's automated safety review yet. SkillMD never runs a skill's scripts for you; review the SKILL.md before installing.

Which AI agents work with Codex Pipeline?

This skill is tagged as working with Claude Code, Claude.ai, OpenAI Codex. SKILL.md is an open format, so most agents that read a skills directory can load it too.

Is Codex Pipeline free to use?

Yes. Installing skills from SkillMD is free, and the skill stays under its author's original license.

Who published Codex Pipeline?

V-Songbird (@v-songbird) published this skill. Their other Agent Skills are listed on their SkillMD profile.

Codex Pipeline by v-songbird — Agent Skill

Name: Codex Pipeline
Author: V-Songbird

Codex Pipeline

Runs a dev task through the cross-model codex pipeline — spec and held-out tests written here, implementation by `mcp__codex__codex`, fresh-context diff review, fixes via `mcp__codex__codex-reply`. Use when the user asks to run the codex pipeline or have Codex or GPT implement something — e.g. "run this through codex", "use the cross-model pipeline on this" — or when an implementation task carries real defect risk (a new module with a subtle contract, parsing or validation logic, edge-case-heavy algorithms, multi-file changes). Do NOT use for small mechanical edits, docs, config, or plain review of existing code — implement those directly or use /code-review instead.

V-Songbird 1beeda5 2.4 KB Updated Sep 17, 2026

File contents

Codex pipeline

Cross-model build-and-review. This session architects, Codex implements, a fresh context reviews. The fresh review is the point: a second model with different blind spots catches what the implementer and the tests both miss.

1. Architect

Read the code the task touches.
Write a self-contained spec file: the contract, required error types, and an "Adversarial notes" section listing the edge cases you'd attack.
Write a held-out verification suite yourself, named distinctly (e.g. verify.test.js). Codex must never see it.

2. Implement

Call mcp__codex__codex with:

prompt: implement from the spec file, write and run its own selftest, and do NOT open the held-out suite (name it explicitly).
cwd set to the working directory, sandbox: workspace-write, approval-policy: never.

Save the returned threadId. Codex returns only a one-paragraph summary — trust the files on disk, never the summary.

3. Verify

Run the held-out suite plus the project's real gates (tests, types, lint) yourself, in this session.

4. Fresh review

Spawn a general-purpose Agent with no pipeline context: give it only the spec path and the implementation path, read-only intent. Require it to reproduce every suspected defect before reporting, and to return APPROVE or REQUEST-CHANGES with repro input and actual-vs-expected output per defect.

5. Fix

Send confirmed defects — with their repros — to mcp__codex__codex-reply on the saved threadId. Then re-run step 3.

6. Close

At most two review→fix rounds; hand anything still open to the user.
Report: what was built, verification results, what the fresh review caught.
Never commit without the user's explicit go.

---name: codex-pipelinedescription: Runs a dev task through the cross-model codex pipeline — spec and held-out tests written here, implementation by `mcp__codex__codex`, fresh-context diff review, fixes via `mcp__codex__codex-reply`. Use when the user asks to run the codex pipeline or have Codex or GPT implement something — e.g. "run this through codex", "use the cross-model pipeline on this" — or when an implementation task carries real defect risk (a new module with a subtle contract, parsing or validation logic, edge-case-heavy algorithms, multi-file changes). Do NOT use for small mechanical edits, docs, config, or plain review of existing code — implement those directly or use /code-review instead.---# Codex pipelineCross-model build-and-review. This session architects, Codex implements, afresh context reviews. The fresh review is the point: a second model withdifferent blind spots catches what the implementer and the tests both miss.## 1. Architect- Read the code the task touches.- Write a self-contained spec file: the contract, required error types, and an "Adversarial notes" section listing the edge cases you'd attack.- Write a held-out verification suite yourself, named distinctly (e.g. `verify.test.js`). Codex must never see it.## 2. ImplementCall `mcp__codex__codex` with:- `prompt`: implement from the spec file, write and run its own selftest, and do NOT open the held-out suite (name it explicitly).- `cwd` set to the working directory, `sandbox: workspace-write`, `approval-policy: never`.Save the returned `threadId`. Codex returns only a one-paragraph summary —trust the files on disk, never the summary.## 3. VerifyRun the held-out suite plus the project's real gates (tests, types, lint)yourself, in this session.## 4. Fresh reviewSpawn a `general-purpose` Agent with no pipeline context: give it only thespec path and the implementation path, read-only intent. Require it toreproduce every suspected defect before reporting, and to return APPROVE orREQUEST-CHANGES with repro input and actual-vs-expected output per defect.## 5. FixSend confirmed defects — with their repros — to `mcp__codex__codex-reply` onthe saved `threadId`. Then re-run step 3.## 6. Close- At most two review→fix rounds; hand anything still open to the user.- Report: what was built, verification results, what the fresh review caught.- Never commit without the user's explicit go.

Frequently asked questions