DeepWork plugin
Claude Code plugin that runs complex tasks as step-by-step workflows with quality gates and reviews changes against project rules.
- 51 ★ on the repository
- Last commit
Extended access
Why this level
- The plugin runs review agents and quality hooks that read project files and write reports.
Install
- Works in
Manual install
claude plugin marketplace add Unsupervisedcom/deepwork && claude plugin install deepwork@deepwork-plugins && claude "/deepwork:new_user"Quickstart command from the README; onboarding starts afterwards.
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- The plugin runs review agents and quality hooks that read project files and write reports.
This is third-party code. Review the repository files before installing.
What it does
DeepWork, a Claude Code plugin, drives the agent through a multi-step task on a strict script: every step has defined inputs, outputs and quality checks, and no step can be skipped. A script is built from work already done. Claude performs the task, you correct the result, and `/deepwork` turns that run into a repeatable process. The second part, `/review`, runs the rules from `.deepreview` files in parallel, each reviewer seeing only the files it needs. It catches version drift, stale docs and unpinned CI actions. The third part, DeepSchemas, validates a file against a schema at write time. `/deepplan` builds a plan and turns it into an executable process.
Who it is for. Teams and solo developers who hand Claude Code multi-step jobs such as research or a codebase audit and are tired of re-checking the agent by hand.
Good fit when
- A monthly competitor review goes differently every time because the agent skips steps
- A repository needs checks a linter lacks: doc freshness, version consistency, pinned CI actions
- Config or API description formats should be validated when a file is written, not at review
- A complex task is planned first and then executed against an approved plan with checkpoints
Not a fit when
- The task is a one-off and short, so a workflow will not pay off
- Your agent runs outside Claude Code, where full support is claimed
- You use the VS Code extension: the author recommends the terminal for full features
Example request
Research our top three competitors, write a SWOT for each, then turn what we did into a workflow called `update_swot`.Limitations
Claude Code is fully supported. For Gemini CLI the skill is connected manually in TOML format, other agents are planned. The author recommends the terminal because some features are missing in the VS Code extension. The agent sometimes bypasses the workflow and does the task itself, so run the process explicitly with `/deepwork` and its name. Describing a process takes about ten minutes of questions. GitHub did not recognise the repository license, so read the license and contributor agreement before commercial use. Adding the marketplace in Claude Desktop on Windows currently fails.
How to disable. Disable or remove the `deepwork` plugin through the `/plugin` menu. Delete `.deepreview` files, the `.deepwork/` directory and schemas in your project separately.
FAQ
How do I install DeepWork in Claude Code?
Run `claude plugin marketplace add Unsupervisedcom/deepwork`, then `claude plugin install deepwork@deepwork-plugins`. The `/deepwork:new_user` command starts onboarding.
How do I turn a finished task into a workflow?
Do the task with Claude and refine the result. Then call `/deepwork` and ask it to create a job and workflow from what was done. The agent asks clarifying questions and generates the steps; after that you run it as `/deepwork <name>`.
What does `/review` check?
The rules in `.deepreview` files against changed files on the branch. The README lists code style, stale docs, versions in `pyproject.toml` and `CHANGELOG.md`, unpinned GitHub actions and migration conflicts.
Does DeepWork work with Cursor or Codex?
The support table lists only Claude Code and, partially, Gemini CLI. OpenCode and GitHub Copilot CLI are planned, and Cursor and Codex are not mentioned.
What are DeepSchemas?
File-level schemas with RFC 2119 requirements (MUST, SHOULD, MAY). A file is checked on write through JSON Schema and bash commands, and the same requirements feed reviews.
Discussion
No questions or reviews yet.
Related
PluginsA Claude Code plugin that hands coding tasks to Codex agents, monitors them and verifies the results independently
A Claude Code plugin with correction memory: every fix becomes a rule, plus topic wikis, hooks and dozens of skills
Workflows for Claude Code: /do, /omo, /bmad-pilot and more commands, with Codex, Gemini or OpenCode writing the code through codeagent-wrapper
Spacedock splits agent work into stages with decision gates, so a human approves or returns each result together with evidence.
A Claude Code plugin that switches session modes with /work-mode: autonomous work, delegation and converging wrap-up.
Skills 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
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Repository stars | Main difference |
|---|---|---|---|
| DeepWork (this card) | no | 51 | |
| Pro Workflow | no | 2,954 | accumulates corrections as rules and a wiki; DeepWork builds your own processes with step checks |
| Engineering Workflow Skills | no | 1,629 | nine ready skills for development phases; DeepWork builds custom workflows and review rules |
README in short
The project targets unreliable agents on long tasks. Workflows fix the order of steps, reviews check every change, schemas define file shape. All three parts run on one engine.