The AgentivUse cases

Sort incoming NDAs before counsel sees them

An agent that pre-screens every NDA that reaches the sales inbox against the company's own NDA playbook, sorts it GREEN, YELLOW or RED with the clause behind each flag, pulls the counterparty's earlier agreements from SharePoint, and routes it for signature, a reviewer or counsel, with Anthropic's Legal plugin in Claude Cowork

Function: OperationsEffort: An afternoonPublished: 2026-10-25

The problem

Every new supplier, partner or prospect sends an NDA before the real conversation starts, and most are fine, but the few with a hidden non-compete, a perpetual term or an IP grant look exactly like the rest. Without a lawyer on staff the owner either signs without reading closely or lets the deal wait days for an outside review of a routine document.

How the agent works

  1. It starts

    The triage-nda skill in Anthropic's Legal plugin for Claude Cowork runs when a new NDA arrives as an uploaded PDF or Word file, a link to the document, or pasted text; through the plugin's Microsoft 365 connector the agent can also search Outlook for the thread and SharePoint for earlier agreements with the same counterparty.

  2. The agent works

    It reads the company's NDA rules from a local playbook file and, if none exists, uses market standard defaults and says so on the report. It checks ten areas (agreement structure, definition of confidential information, receiving party obligations, the five standard carveouts, permitted disclosures, term and survival, return or destruction, remedies, problem clauses such as non-solicits, non-competes, exclusivity, residuals and IP grants, and governing law) and marks each PASS, FLAG or FAIL. The result is GREEN (approve under standard delegation, typically the same day), YELLOW (minor issues a designated reviewer can fix in one pass, typically one to two business days) or RED (do not sign, counsel prepares a counterproposal, typically three to five business days), with the parties, term, governing law, the list of issues and a suggested fix for each. A person acts on every result: approves and routes GREEN for signature, sends the YELLOW flags to the reviewer, holds RED.

  3. The result

    The plugin states that its output is a screening aid to be checked by a qualified lawyer, and its defaults are US based, so the playbook must be adapted to the law where you operate.

Worked example

Anthropic's triage-nda skill ships with market standard defaults: a 2 to 3 year term (up to 5 years for trade secrets), confidentiality surviving 2 to 5 years after the end, 30 days to confirm an oral disclosure in writing, five standard carveouts, and certification of destruction accepted while a sworn affidavit is treated as onerous. In its own examples a 5 year term with 7 year survival is YELLOW; a one sided agreement where mutual was needed, a non-compete, liquidated damages, a term of 10 years or more without a trade secret reason, or commercial terms hidden inside a document called an NDA are all RED. The /vendor-check command in the same plugin reports the existing NDAs, MSAs and DPAs with a vendor and their expiry dates, which is the next question every NDA raises.

Data protection. This use case touches personal data, so check the rules where you operate (for example GDPR, PDPA). An NDA carries the names, titles and contact details of the signatories, and connecting Outlook and SharePoint lets the agent see other mail and documents. Limit the connector to the contracts folder and the mailbox that receives NDAs, store the screening report with that contract, and keep counterparty NDA text out of chats people outside the team can read. Check the data protection rules where you operate.

Join the Agent OS Club

8 weeks to the AI team a 50 person company would hire. Instant access to our agent blueprints. USD 15 a month, cancel any time.

Join the Agent OS Club

Building this for an organization? Learn more at theagentiv.com

Sources

Related use cases

Reorder before you run out, not after

An operations agent that reads sales velocity and stock on hand from Shopify, Square or a CSV, projects the stockout date per SKU against supplier lead time, sizes a 60 day reorder, drafts one purchase order and vendor email per supplier, and stages the coded cost for the bookkeeper, with three separate approvals before anything is sent or recorded

OperationsEffort: About a weekClaude Cowork · Shopify · Square

A small shop usually learns an item is out when a customer asks for it, and learns it over ordered when cash sits in stock that does not move. Reordering runs on the owner's memory of what sells, even though the rhythm shifts every month and nobody counts what is already on its way from the supplier.

Answer data access requests on time

An agent that logs each personal data request (access, correction, deletion), works out the deadline under the law that applies, searches Outlook, SharePoint, OneDrive and Teams for the person's data, and drafts the acknowledgement and the response for a person to review before anything is sent

OperationsEffort: About a weekClaude Cowork · Microsoft 365 · Outlook

When a customer or former employee writes 'send me all the data you hold on me' or 'delete my data', most small businesses have no process: nobody owns it, the deadline is unclear, and the data is spread across mailboxes, shared folders and chats. Missing the deadline or sending the wrong person's data is a bigger risk than the request itself.

Every monday.com board clean by Friday

An operations agent that audits the monday.com boards every Friday for items with no owner, no due date, a stale status or a likely duplicate, posts the findings with a proposed fix for each, and applies only the fixes the project lead ticks, in bulk, through the official monday.com MCP

OperationsEffort: An afternoonmonday.com · Claude Code · Slack

A monday.com account that started with one clean board now has forty: items with no owner, due dates in the past, statuses nobody has touched since the kickoff, three columns that mean the same thing. The weekly report built on those boards is wrong before anyone reads it, and nobody has an afternoon free to tidy up.