Guide · Agent skill
How to use the Document Review agent skill (ce-doc-review)
Updated · written with AI from the project's own documentation and checked against our catalogue data
ce-doc-review is an agent skill that makes your AI agent review a planning document, such as a requirements file, a plan or a spec, the way a small panel of specialist reviewers would. It is for product managers, founders and engineers who want gaps and feasibility problems found before work starts. After this guide you will know what the skill does, how to install it, how to ask for a review and how to read what comes back.
Document ReviewEveryInc/compound-engineering-plugin · MIT Download the skill Quick start
- 1Download and unzip the skill. Download the ce-doc-review ZIP from the directory page and unzip it so the folder contains SKILL.md and references.
- 2Place it where your agent looks. Copy the folder into ~/.claude/skills/ for personal use or into a project's .claude/skills/ folder. In the Claude apps, upload the ZIP under Settings, then Capabilities, then Skills.
- 3Confirm the document is readable. Make sure the agent can read the file at the path you will give it. Unreadable paths stop the review before any reviewer runs.
- 4Ask for a review. Describe the goal in plain language and include the path to the .md or .html document.
- 5Answer the agent's questions. In interactive mode, approve or decline each grouped finding and routing choice. Questions cover only changes the agent is not allowed to make alone.
- 6Check the result. Read the findings and the coverage section, then confirm the fixes that were applied.
What the Document Review skill makes your agent do
When you ask your agent to review a planning document, the skill sends the document to several reviewer sub-agents. Coherence and feasibility reviewers always run. Depending on the document, product, design, security or adversarial reviewers can be added too.
The skill then checks the reviewers' claims against the whole document and combines their findings into a short list of real problems. It applies only fixes it is already allowed to make and asks you about the rest. A document that is adequate needs no changes, and the skill is designed not to invent work for it.
- Coherence and feasibility reviewers always run
- Product, design, security and adversarial reviewers run only when the document calls for them
- Findings are combined into a short list rather than a long report
- Fixes outside the skill's permission are raised as questions for you
Install the Document Review skill
Download the ZIP for ce-doc-review from the directory page and unzip it. The skill is a folder that contains SKILL.md and a references folder.
For a personal install, place the folder in ~/.claude/skills/. To limit it to one project, place it in the project's .claude/skills/ folder instead.
In the Claude apps you can also upload the ZIP under Settings, then Capabilities, then Skills.
- Personal install: ~/.claude/skills/
- Project install: .claude/skills/ inside your project
- Claude apps: Settings → Capabilities → Skills, then upload the ZIP
Check what the skill needs before you run it
The skill needs an agent runtime that supports SKILL.md skills and subagent dispatch. The agent also needs readable filesystem access to the document's path, so keep the document in a location the agent can read.
A config file is optional. If your project uses a custom docs_root, add .compound-engineering/config.yaml to the repository. Other model providers are optional and only matter for the cross-model review pass described below.
- An agent runtime with SKILL.md skill support and subagent dispatch
- Readable filesystem access to the document
- Optional: .compound-engineering/config.yaml when a custom docs_root is used
- Optional: other model providers for the cross-model review pass
Ask for a review with a natural request
You do not need a special command. Describe what you want in plain language and include the path to the document. The skill accepts Markdown (.md) and HTML (.html) files.
Requests that work well name the document and the goal, for example asking the agent to review a spec to find gaps before building, or to check a technical plan for feasibility before work starts.
The skill's argument hint is [mode:non-interactive] [path/to/document]. Add mode:non-interactive only when you want a structured result with no questions.
- "Review docs/plans/billing-redesign.md for gaps before we start building."
- "Check this technical plan for feasibility problems."
- "Improve the requirements in specs/onboarding.html."
- "Review this spec, mode:non-interactive, and return the findings as text."
[mode:non-interactive] [path/to/document.{md,html}]Read the review and decide what happens next
Before reviewers run, the agent announces the team and why each reviewer was chosen. When they finish, the findings are grouped and checked against the document. In interactive mode you see grouped findings and a routing question, then walk through them one at a time or approve a bulk action after a preview.
If a reviewer fails, the review result names it in its coverage section so you know what was and was not checked.
- Each finding has a confidence level and a fix class
- Fixes within the agent's authority are applied and checked
- Other items are presented as questions, deferred to Open Questions, or noted as observations
- Nothing is applied in a bulk action until you approve the preview
Know the safety and privacy limits
The skill runs no scripts. It does use the network, because the optional cross-model review pass can send document content to other models. The agent discloses this before anything leaves the machine, and the pass can be skipped. A checkout setting named cross_model_review_mode can skip it with a stated reason.
Do not send documents with sensitive personal or confidential data through the cross-model pass unless you have confirmed where that data may go. The skill is not a substitute for legal, security or compliance sign-off.
- No scripts are run by the skill
- Network use comes from the optional cross-model pass only
- Disclosure happens before any content leaves the machine
- The cross-model pass is additive: if it fails or times out, the review continues and the failure is named in coverage
What people use it for
- A product manager checks a feature spec for gaps and unclear requirements before engineering starts.
- An engineer asks for a feasibility check on a technical plan before committing to a design.
- A team using compound-engineering workflows reviews the requirements and plan files those workflows produce.
- A founder runs a quick review on a planning document to catch contradictions between sections.
Frequently asked questions
Is the Document Review skill free to use?
The source material for this guide does not state pricing. The skill is published under the MIT licence, which is named on this page. Check the licence file in the source repository for the full terms.
What do I need to run ce-doc-review?
You need an agent runtime that supports SKILL.md skills and subagent dispatch, plus readable filesystem access to the document. A config file and other model providers are optional.
How do I install the ce-doc-review skill?
Download the ZIP from the directory page and unzip it into ~/.claude/skills/ for personal use or into a project's .claude/skills/ folder. In the Claude apps, upload the ZIP under Settings, then Capabilities, then Skills.
Does the skill send my document to other AI models?
Only the optional cross-model review pass can send content to other models. The agent discloses this before anything leaves the machine, and the pass can be skipped. Avoid it for documents with sensitive personal or confidential data.
Why did the review not run on my document?
The skill checks that the document path is readable before dispatching any reviewer. If the file is not on disk where the agent can read it, for example on a branch that is not checked out, no reviewers run.
How is this different from asking a chatbot to review my plan?
This skill uses several reviewer sub-agents with different lenses, checks their claims against the whole document, and applies only fixes it is allowed to make. It is still AI-generated review, not expert certification.
New agent skills, repos and AI tools — every week
The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.
Our verdict
Use ce-doc-review if you want a spec or technical plan checked for gaps and feasibility problems before building, and you work in an agent runtime that supports SKILL.md skills. Pick something else for final legal, security or compliance sign-off, for very short notes where a full multi-reviewer pass adds cost without value, or for sensitive documents you cannot safely send to other models.
Download the skill