Guide · Agent skill
How to use the Tabular Review agent skill (tabular-review)
Updated · written with AI from the project's own documentation and checked against our catalogue data
The tabular-review skill is an agent skill for Claude Code or another SKILL.md-compatible agent. It reads a batch of documents, answers the same set of questions for each one in a fixed typed format, and writes the results to a spreadsheet with one row per document and one column per data point. This guide explains what the skill does, how to install it, how to trigger it with plain requests, and what to check before using it on confidential material.
Tabular Document Reviewanthropics/claude-for-legal · Apache-2.0 Download the skill Quick start
- 1Download the ZIP. Download the tabular-review skill ZIP from the directory page.
- 2Unzip into a skills folder. Unzip into ~/.claude/skills/ for all projects, or into a project's .claude/skills/ folder for one project.
- 3Or upload in the Claude apps. In the Claude apps, go to Settings → Capabilities → Skills and upload the skill.
- 4Make your documents available. Provide a local folder, a Google Drive folder, or a VDR path through the relevant connector.
- 5Ask in plain language. Say something like 'review these contracts for change of control and assignment clauses' to start the workflow.
- 6Approve the schema and sample. Check the generated .review-schema.yaml and the sample run results, then adjust before the full batch.
- 7Verify flagged cells. Open the output, check the quotes and locations, and work through the needs_review and unclear cells.
What the tabular-review skill makes the agent do
When you have many documents and the same questions to ask of each, the skill has the agent read every document and answer each question in a consistent typed format. Column C means the same thing in row 1 as in row 200, which is the main reason to use it over free-form prompting.
Every answer comes with the exact quoted sentence and its location, so a person can check it quickly. Cells that are missing, ambiguous, or need a human judgment are marked with explicit states (not_present, unclear, needs_review) instead of being left blank.
- Column types include verbatim, classify, date, duration, currency, number and free.
- Each non-verbatim cell has a companion source quote.
- The skill is built for M&A diligence, such as checking change-of-control, assignment or MAC clauses across many target contracts, but works for any batch review that needs a spreadsheet.
Install the tabular-review skill
The skill is distributed as a ZIP from the directory page. Unzip it into one of the following locations, depending on whether you want it for all projects or one project.
You can also upload it in the Claude apps under Settings → Capabilities → Skills.
- Personal install: unzip into ~/.claude/skills/ so the skill is available across projects.
- Project install: unzip into the project's .claude/skills/ folder so it is available only in that project.
- Claude apps: upload the skill under Settings → Capabilities → Skills.
Trigger the skill with natural requests
You do not need to type a command. The skill's description lists phrases that activate it, so ordinary requests work. You can also point the agent at a folder of documents and ask it to compare them.
The skill also has an explicit form, /corporate-legal:tabular-review, with optional flags for a schema file, a document path, a template, an output format and a sample size.
- "Tabular review of these contracts"
- "Build a grid for these leases"
- "Extract these fields from these contracts: counterparty, effective date, assignment clause"
- "Review these documents for change of control, assignment and MAC clauses"
- "Give me a spreadsheet of the termination terms in this folder"
- "Batch review the vendor agreements in ./vdr/02-Contracts/"
/corporate-legal:tabular-review
/corporate-legal:tabular-review --schema .review-schema.yaml --docs ./vdr/02-Contracts/
/corporate-legal:tabular-review --template ma-diligenceFollow the workflow: schema, sample run, then full run
The skill works in stages and asks you to confirm along the way. First it confirms which documents, which columns and where the output goes. It then writes a reusable schema file called .review-schema.yaml and asks you to approve it.
Before processing everything, it runs a sample of 3 to 5 documents by default, so you can adjust the schema. After the full run, it normalises the answers, flags outliers and writes the output.
- Confirm documents, columns and output location.
- Build and approve the typed schema in .review-schema.yaml.
- Run a sample (default 5 documents, set with --sample) and adjust.
- Run the full batch, one sub-agent per document, in parallel.
- Normalise, flag inconsistencies, and write the outputs with a summary.
Read the output and verify the cells
The skill writes an .xlsx file or a Google Sheets file (it asks which), plus .csv, _sources.csv and markdown files. The Excel layout puts the source quote and location next to each answer, and a cell comment on each data cell carries the same quote.
The summary reports the verification workload: counts of not_present, unclear and needs_review per column, plus flagged columns. Treat every cell as a lead to check against the source, not as a finding.
- Review sheet: one row per document, one column per schema column, with a hidden _source column after each data column.
- Cell comments hold the quote and location for hover checking.
- The summary lists which columns need a human to check.
Safety notes: scripts, network and confidential documents
This skill runs scripts and uses the network. The skill can write files with openpyxl (pip3 install openpyxl) if Excel tooling is not already available, and it can use Google Sheets or connectors to Google Drive or a VDR (Box, Datasite, iManage). Read the SKILL.md and reference files before installing, and check which connectors are enabled.
Do not send confidential documents to a Drive or VDR unless your organisation has approved the agent's data handling for that matter.
- The skill runs scripts and uses the network.
- Google Sheets output needs access to Google Sheets; the skill checks write access and falls back to CSV if it is not verified.
- Optional practice-level config lives at ~/.claude/plugins/config/claude-for-legal/corporate-legal/CLAUDE.md.
Tips for better results
Start with the ma-diligence template or a schema you already have, then add or cut columns. The template is a starting point, not a checklist. Use the sample run to catch vague prompts before you spend time on a large batch.
Use classify columns with fixed options wherever answers should be comparable, and keep free-text columns to a minimum because they drift.
- Write prompts as the question a reviewer would ask of the document.
- Keep the number of documents in a single batch manageable; if there are more than 200, the skill suggests starting with a materiality-filtered subset.
- Use --schema to rerun the same questions against new documents.
What people use it for
- M&A diligence across many target contracts, such as checking change-of-control or assignment clauses.
- Building a consistent spreadsheet from a folder of similar documents such as leases or vendor agreements.
- Producing a verification workload summary showing which cells need a human to check.
- Comparing termination, notice or renewal terms across a supplier contract portfolio.
Frequently asked questions
How do I install the tabular-review agent 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 for one project. You can also upload it in the Claude apps under Settings → Capabilities → Skills.
What do I need to run tabular-review?
You need Claude Code or another SKILL.md-compatible agent and access to your documents, either as a local folder, a Google Drive folder, or a VDR through connectors. Google Sheets access is needed only if you choose Sheets output, and the skill checks write access first.
Is the tabular-review skill free?
The source material does not state pricing or licence terms for the skill beyond its licence, which is Apache-2.0. Check the licence file in the repository for the terms that apply.
Does tabular-review give legal conclusions?
No. The skill describes every cell as a lead that needs verification, not a finding. It is risky to rely on any cell without reading the source.
When should I not use tabular-review?
Avoid it for final legal conclusions and for issue spotting across a small set of documents, where a single careful read is better. Also avoid sending confidential documents to a Drive or VDR unless your organisation has approved the agent's data handling.
Why does the skill output needs_review or unclear instead of blank cells?
Blank cells hide information. The skill uses not_present when the clause is absent, unclear when it is present but cannot be classified confidently, and needs_review when a human must make the call.
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 tabular-review if you have many documents and the same questions for each, such as M&A diligence or lease and vendor contract portfolios, and you want cited, consistently typed answers in a spreadsheet. Pick something else if you need a final legal opinion, a careful issue spot across a handful of documents, or you cannot approve the data handling for confidential files.
Download the skill