Claude's Record-a-Skill for legal research: useful for the boring half, dangerous for the rest

What it is

Record a skill is a demonstration-to-automation workflow in Claude Cowork. You perform a task on your computer while recording the screen, narrate what you are doing and why, and Claude converts that evidence into a Skill it can apply again. It lives inside the Claude desktop app's Cowork surface, not the browser, and the artifact you get back is a reusable procedure — not a trained model.

The next-work-session test

Say your next session is a run of due-diligence redlines on five NDAs from the same counterparty. The manual loop: open the doc, jump to the indemnity clause, compare against your playbook's carve-outs and caps, flag anything that widens the indemnitor's exposure, note the governing-law clause, then repeat for the confidentiality and IP sections.

Record yourself doing one NDA end-to-end, narrating why you're rejecting an uncapped indemnity or accepting a mutual one. In theory, Claude replays that pattern across the other four and hands you a starting redline set. The ZDNet author reported the workflow cut research from several hours to about 30 minutes — treat that as one anecdote, not a benchmark. For a lawyer, the honest win is a first-pass clause triage, not a finished markup.

Pricing

There is no separate Record-a-Skill SKU. Access is bundled into Claude subscriptions. Claude Cowork was initially exclusive to Max subscribers, and as of January 2026, Anthropic opened Cowork access to Pro ($20/month) subscribers as well. Team Standard is $20 per seat per month with annual billing or $25 billed monthly. Team Premium is $100 annualized or $125 monthly. Max plans run higher — Max 5x is $100 and Max 20x is $200. Verify against Anthropic's own plans page before you expense it; the tiers move.

One thing that is not clearly priced: skill runs draw from the same usage pool as your chats and Claude Code, so a firm running skills across a document set will burn quota faster than a user just chatting.

What we'd actually use it for

Narrow it. Good candidates:

• Clause extraction runs. Pull every indemnity, LoL, and governing-law clause out of a folder of vendor MSAs into a structured table.
• Playbook checks. Compare a counterparty's redline against your firm's standard positions and flag deviations for human review.
• Precedent hunts inside your own DMS. Record how you search NetDocuments or iManage for prior deals with a specific carve-out; rerun on a new matter.
• Cite pulls. Narrate how you verify a case is still good law across a set of citations.

Everything downstream of that — the actual redline judgment, the client memo, the negotiation call — stays with the lawyer.

Limits

The privilege and confidentiality problems are the ones to sit with.

Anthropic has not published a dedicated retention and privacy page for Record a Skill, so treat anything visible during a recording as potentially reviewable. Claude asks permission to record your screen and watch what you do. The warning: do not type passwords or secrets, or show private conversations, while it is rolling. For a lawyer that warning extends to client names in tab titles, matter numbers in DMS breadcrumbs, and any privileged doc open in another window.

Other constraints:

• Recording quality drives skill quality. A rushed, silent walkthrough produces a weaker skill than a deliberate, narrated one. Legal reasoning is exactly the kind of tacit work that gets lost when you don't narrate.
• Availability is a phased rollout rather than a simultaneous flip for all eligible users — your Pro seat may not have it yet.
• Skills are brittle. If counsel's DMS UI changes, or a counterparty's contract structure differs, the skill drifts. You still verify every output.
• No jurisdiction awareness by default. A skill that flags indemnity carve-outs against a New York playbook will happily do the wrong thing on an English-law contract.
• Bar rules on generative AI, supervision, and client confidentiality haven't caught up. Any use on client matter needs partner sign-off and, arguably, engagement-letter language.

Try it if

• You're in-house and run the same clause-triage loop across vendor paper every week.
• You have a Team or Enterprise seat with a clear data-handling posture from Anthropic and IT approval to use Cowork on business data.
• You're willing to record against sanitized documents first and treat the skill as a first-pass tool, not a work product.
• Your DMS and browser tabs can be cleared of unrelated matters before you hit record.

Skip it if

• You'd be recording on live client matter without written client consent and a defensible confidentiality analysis.
• Your work is bespoke — bet-the-company litigation, novel M&A structures — where every clause is negotiated fresh.
• You can't reliably narrate why you make a call. The skill will only be as good as the reasoning you speak aloud.
• You need an audit trail for regulators; retention specifics for these recordings aren't publicly nailed down yet.

Bottom line: Record-a-Skill is a plausible way to automate the mechanical half of legal research — extraction, comparison, checklist runs. It is not a way to automate judgment, and the screen-recording surface makes it a confidentiality question before it's a productivity one. Read the source walkthrough on ZDNet for the workflow, then take it to your GC before you take it to a matter.

---

This content is for informational purposes only and is not legal advice. Confirm confidentiality, privilege, and jurisdictional rules before using any AI tool with client matters.