Claude Dynamic Workflows on Codex skill
The /codex-workflows skill: Claude Code writes a script, dozens of Codex agents run it, and the run shows on a map
- 327 ★ on the repository
- Last commit
- MIT
Broad access
Why this level
- Launches dozens of Codex agents that write to the project folder by default
- Spends tokens on your Codex account, and you have to set the ceiling yourself
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add skills/claude-dynamic-workflows-on-codexDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- Launches dozens of Codex agents that write to the project folder by default
- Spends tokens on your Codex account, and you have to set the ceiling yourself
This is third-party code. Review the repository files before installing.
What it does
Codex Workflows is a Claude Code skill that hands a large task to many Codex agents. You type `/codex-workflows` and a sentence or two about the task, for example auditing every route under src/ for missing authorization. Claude writes a dynamic-workflow style script itself, with agent(), parallel() and phase() calls and a budget. It saves the script into the project as a name.workflow.js file so it can be read and rerun. The script runs on the local Codex app server, and every agent() call becomes a separate Codex thread in place of a Claude subagent. Only the result enters the conversation context, while the run shows up on a map in the browser or as a text tree in the terminal. Workers can last longer than one turn: you can give them new instructions in the same thread, race several and cancel the ones behind. The `--multi` flag launches several scripts at once, and Claude supervises them: it answers questions at stop points, redirects agents and closes dead ends. To look at a result without Codex there is an offline viewer with a bundled demo incident.
Who it is for. For people who work in Claude Code, have a Codex login and want to spread a code audit, a large migration or a cross-checked research job across dozens of parallel agents.
Good fit when
- You want an audit of every route under src/ for missing access checks, with each finding verified by a skeptic in parallel
- A large migration that splits into phases and runs as a batch of agents under a hard token ceiling set with `--budget`
- Research where every claim must be re-checked independently and only confirmed ones kept
- You want to size the run first: `--plan` counts agents per phase and estimates a budget without spending tokens
- You need a gate for your decision inside the run: the script stops at an answer card and waits for you
Not a fit when
- Codex CLI is not installed or you have not logged in with `codex login`
- The task is small and fits one or two subagent calls: the script overhead will not pay off
- You want Claude to pick the skill up without your command: it runs only manually through `/codex-workflows`
- You want to mix different models across the stages of a run
Example request
/codex-workflows Audit every route under src/ for missing authorization, read-only, and show me the run on a mapLimitations
Needs Node 18 or newer and the Codex CLI on PATH with a completed login; the project has no npm dependencies. The OpenAI API is unavailable from Russia. The project is unofficial and not affiliated with OpenAI or Anthropic. Agents may write to the working folder by default, and `--sandbox read-only` leaves read access only. Tokens are billed to your Codex account. The whole run uses one flagship model, and only the reasoning effort changes. The skill runs only manually, auto-invocation is disabled.
How to disable. If you installed the plugin, remove it with `/plugin uninstall codex-workflows@codex-workflows`. If you cloned it, delete the ~/.claude/skills/codex-workflows folder. Saved .workflow.js scripts stay in your projects, so delete them by hand.
FAQ
How do I install Codex Workflows in Claude Code?
Run `/plugin marketplace add scasella/claude-dynamic-workflows-codex` and `/plugin install codex-workflows@codex-workflows`. The other route is cloning the repository into ~/.claude/skills/codex-workflows. Either way the Codex CLI must be installed and logged in.
How do I check that Codex is reachable?
Run `npx github:scasella/claude-dynamic-workflows-codex doctor`; a healthy answer shows state ready. The same entry point gives run, view, map and other commands without installing.
How do I look at an example without Codex?
Clone the repository and run `node runner/bin/view-run.js examples/incident-demo --open`. It opens an offline map of a demo incident with slow checkout: a parallel triage, a race of three workers and one gate.
Can I limit token spend?
Yes. The `--budget` flag sets a hard ceiling, and `--plan` counts agents per phase and estimates a budget without spending tokens. If the ceiling trips, you can continue with `--resume`: finished agents replay from the journal for free.
How is this different from native Claude Code dynamic workflows?
The script syntax is the same, except each agent() runs as a Codex thread where native workflows use Claude subagents. On top come long-lived workers, races that cancel the ones behind and a `--multi` mode that supervises several scripts.
Discussion
No questions or reviews yet.
Related
SkillsSkills for cross-provider work between Claude Code and Codex: one agent drafts the plan, the other reviews it, then a build and an independent code inspection f
Workflows for Claude Code: /do, /omo, /bmad-pilot and more commands, with Codex, Gemini or OpenCode writing the code through codeagent-wrapper
A visual workflow builder for agents: draw a flow on a canvas and export it as skills and commands for Claude Code, Codex, Cursor and others
A desktop app where Claude Code, Codex, Gemini and other agents work as a team of specialists in a shared project
Multi-agent orchestration for Claude Code: teams, roles, autopilot and task routing across skills and subagents
A skill that turns several Claude Code, Codex and Cursor windows into a team with a coordinator, shared registry and cross-review
Comparison with similar tools
Entries that share the main search query, and how they differ.
| Entry | Official | Repository stars | Main difference |
|---|---|---|---|
| Claude Dynamic Workflows on Codex (this entry) | no | 327 | |
| Codex Astra + Luna Orchestrator | no | 1,730 | Codex profiles where one model orchestrates and another runs exploration |
| Citadel | no | 927 | a shared operating layer for Claude Code and Codex with a single /do command and project memory |
| Hive | no | 580 | a local browser workbench where agents work as a team under your supervision |
README in short
The project has two parts: a skill for Claude Code and a standalone runner with a viewer that works without Claude Code. The repository ships example scripts, including bug hunts, deep research and tournament sort.
README badge
Original SKILL.md
--- name: codex-workflows description: >- Run a dynamic-workflow script on a local Codex App Server — orchestrate many Codex / GPT agents (the agent / parallel / pipeline / phase / budget DSL) instead of Claude subagents, for codebase audits, large migrations, and multi-agent review or research. Give it one or two rough sentences and it compiles the right harness for you; add --multi for a supervised fleet of concurrent workflows. Manual-invoke only via /codex-workflows. disable-model-invocation: true --- # Codex Workflows Run a Claude Code dynamic-workflow script against a local **Codex App Server**. The authoring surface is identical to native dynamic workflows — `export const meta` plus a body using `agent()`, `parallel()`, `pipeline()`, `phase()`, `log()`, `args`, `budget`, `workflow()` — but every `agent()` call runs as one Codex (GPT) thread+turn instead of a Claude subagent.