WorkflowX plugin
A plugin for Claude Code and Codex with xdo, xdel and xflow modes: a main agent plans, a coder builds, an evaluator checks.
- 54 ★ on the repository
- Last commit
- MIT
Extended access
Why this level
- The implementer and the main agent edit project files and run commands.
- The evaluator creates and runs tests in your environment.
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add plugins/workflowxDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- The implementer and the main agent edit project files and run commands.
- The evaluator creates and runs tests in your environment.
This is third-party code. Review the repository files before installing.
What it does
WorkflowX embeds into an AI coding tool a process where chatting with the agent becomes planning, acceptance and tracking. The user still talks to one main agent. It routes tasks, clarifies requirements, writes notes into `.agents/notes/`, dispatches work and verifies the result. The `coderX` implementer only builds a task against fixed criteria, and `evaluatorX` independently tests it and returns `PASS`, `NEEDS_FIX` or `UNEVALUABLE`. There are three modes. In `xdo` the main agent works alone with no new task. `xdel` hands one task to the implementer with no evaluator. `xflow` runs repository discovery, questions, a ready summary, tasks, the implementer and the evaluator, and suits new features and large refactors. The set includes the `orchestrateX`, `socratesX`, `engineeringX`, `specX`, `auditX`, `noteX` and `proseX` skills.
Who it is for. Developers running large tasks through Claude Code or Codex who want fixed acceptance criteria and independent verification.
Good fit when
- A feature touches several modules and the result must pass independent verification.
- A small fix in one place is done with `xdo` and no extra tasks.
- A fix is handed to the implementer with `xdel` and comes back with a note draft.
- Requirements get lost in chat history and should be kept as notes next to the code.
Not a fit when
- You need a quick one-off edit with no documentation: notes and tasks add overhead.
- The agent is neither Claude Code nor Codex: the README describes only these two platforms.
- The team is not ready to maintain module documents in `.agents/notes/`.
Example request
`xflow` implement email and password login with OAuth: explore the repository, ask clarifying questions and show a ready summary before tasks.Limitations
Node.js 18 or newer is required. In Claude Code the plugin installs by repository address, in Codex through `/plugins`, and manually you copy the `.claude/` or `.codex/` directory into the project root. Commands such as `/xdo` work in Claude Code, while Codex uses a text prefix at the start of a message. The `evaluatorX` evaluator takes part only in `xflow`, and parallel work must be requested explicitly. The repository holds many supporting files, including third-party skills under `.claude/skills/`, and the main descriptions are in Chinese with an English translation. Result examples in the README are teaching samples.
How to disable. In Claude Code remove the `workflowx` plugin through `/plugin`, in Codex through `/plugins`. For a manual install delete the copied directories. Notes in `.agents/notes/` stay in the project and are removed separately.
FAQ
How do I install WorkflowX in Claude Code?
Run `/plugin marketplace add https://github.com/TreeX-X/workflowX`, then `/plugin install workflowx`. For Codex open `/plugins`, search for `workflowx` and install.
Which mode should I choose?
`xdo` fits ordinary local work, `xdel` one task with a note, `xflow` a new feature or a refactor across modules. If unsure, describe the task and the main agent recommends a mode.
What does evaluatorX check?
It does not trust the implementer's report: from the fixed acceptance criteria it builds a minimal test set, runs it and returns `PASS`, `NEEDS_FIX` or `UNEVALUABLE`. It changes no project files.
Where are requirements and decisions stored?
In `.agents/notes/`, organized by module with notes, ideas, requirements, decisions and tasks. The main agent updates the shared task before each handoff.
Can I work in an isolated branch?
Yes, the `-box demo` flag isolates work in a sandbox branch. Parallel tasks are enabled only on explicit request.
Discussion
No questions or reviews yet.
Related
PluginsA 33-skill plugin with gates: question the request, verify facts against sources, test first, review by a second model
A workflow template for Claude Code and five more agents: a backlog task, a diff review and a commit.
Writ turns workflow rules into checks at tool-call time: Claude Code writes no code until you approve the plan and tests
A plugin for Codex and Claude Code that runs a task through a chain of roles from planning to an independent audit
A workflow for coding agents: a plan you approve, evidence-based verification and a handoff of state between sessions.
VBW
Vibe Better With Claude Code
A Claude Code plugin where a team of agents builds the work and a bash kernel accepts nothing until checks pass.
README in short
The README calls WorkflowX a file-driven multi-agent framework that turns a chat with an AI into a process with a plan, acceptance and tracking. It shows the modes, a table of common failures with mechanisms against them, the `xflow` chain and the note structure. An English version accompanies the Chinese text.