Anthropic Cowork Legal Plugin Review Contract Clause by Clause

Anthropic's `/review-contract` skill — shipped February 2026 inside the open-source Cowork legal plugin — does clause-by-clause contract review against a configured negotiation playbook, returning GREEN, YELLOW, or RED flags plus suggested redlines. The market read in February was severe: Thomson Reuters dropped 16%, RELX dropped 14%, Wolters Kluwer dropped 13%, and roughly $285 billion in combined market cap was wiped in a single trading session per Canadian Lawyer's coverage. This is the operator's read on what `/review-contract` actually does, how it compares to Spellbook and Harvey on the same task, and where the open-source plugin fits in a real firm's contract review stack. Most coverage stayed on the market reaction. Here's the workflow.

What /review-contract actually does, clause by clause

Per the Anthropic plugin documentation and the GitHub repo, the `/review-contract` workflow runs in three stages:

Stage 1 — Configure the playbook. The firm or in-house team writes a YAML playbook defining acceptable indemnity caps, preferred forum selection, IP assignment defaults, payment terms, limitation of liability ceilings, termination triggers, and any deal-specific red lines. The playbook is the firm's negotiation doctrine in machine-readable form. A typical playbook for a SaaS reseller might be 15-30 lines; for an M&A indemnity package, 100-200.

Stage 2 — Feed Claude the contract. Drop the document into Claude (via Cowork, claude.ai, or Claude For Word inside Microsoft Word). Invoke `/review-contract`. Claude reads the contract, walks each substantive clause, and compares it against the playbook.

Stage 3 — Get the flagged output. Each clause comes back with one of three flags: - GREEN — clause matches playbook, no negotiation needed - YELLOW — clause deviates from playbook in a negotiable way; Claude proposes redlines - RED — clause violates a hard playbook rule; Claude flags for counsel review and proposes a counter-clause

The output ships with the original clause text, the playbook rule that triggered the flag, the suggested redline, and a short rationale. For a typical SaaS MSA with 40-60 clauses, the review runs in 60-180 seconds depending on Claude tier and effort level.

The second-order effect: a contract reviewer's first pass — typically 45-90 minutes for a senior associate on a mid-complexity contract — gets compressed to a 5-10 minute review of Claude's flagged output. That's the time-saving math that drove the $285B market reaction in February.

How the playbook configuration actually works

The killer feature is configurability, not the workflow. Most clause-by-clause review tools ship with vendor-defined defaults that don't match firm-specific risk tolerance. `/review-contract` ships with no defaults — the firm writes the playbook.

Playbook structure is straightforward YAML:

```yaml indemnity: max_cap: 12_months_fees carve_outs: [gross_negligence, willful_misconduct, ip_infringement] flag_red_if: cap_exceeds_24_months forum: preferred: New York or Delaware flag_red_if: jurisdiction_other_than_us ip_assignment: default: work_for_hire flag_yellow_if: license_only flag_red_if: contractor_retains_ownership ```

The firm checks the YAML into git, version-controls negotiation doctrine across the team, and updates it when a partner revises a position. New associates inherit the senior partner's playbook automatically — the document is the institutional knowledge.

For in-house legal ops teams that have spent 18 months trying to build internal contract review tools through engineering tickets, this is an end-run around the build cycle. The playbook lives in YAML, not in a vendor's proprietary database. Migration cost between Claude tiers or even between LLM providers is the cost of re-pointing a YAML file.

The third-order effect: firms that use `/review-contract` build a tangible asset (the playbook) over time. After 12-18 months of refinement, the YAML captures the firm's negotiation IP — the kind of institutional knowledge senior partners traditionally carried in their heads. That's a knowledge management win, not just a workflow win.

How /review-contract compares to Spellbook and Harvey clause review

All three tools do contract clause review. The structural differences:

Spellbook. Native Word add-in, ships with a precedent library and the Spellbook Library precedent learning feature. Pricing is quote-only per Spellbook's site — industry estimates per Artificial Lawyer reporting suggest $180-$300/seat/month with a $199/seat/month enterprise minimum at 10 seats, not vendor-confirmed. Strong on transactional contracts (NDAs, MSAs, M&A docs). Configurability is good but works through Spellbook's UI, not raw YAML.

Harvey. Quote-only enterprise pricing — industry estimates per Artificial Lawyer's June 2025 reporting suggest $1,200-$2,000+/seat/month at AmLaw scale, not vendor-confirmed. Sized for AmLaw 100 with deep workflow customization, multi-tool integration (DMS, deal rooms, contract repositories), and dedicated customer success. The fit is BigLaw-grade matter management more than solo clause review.

Anthropic /review-contract via Cowork. Open source, hosted on GitHub, free. Cost is the underlying Claude tier ($17-$20/user/month for Pro at annual billing or $25/seat/month for Team at monthly billing per Anthropic pricing). Configurability is YAML-based. Works inside Claude For Word, claude.ai, or Cowork directly.

Where /review-contract wins on FIT: solo and small-firm transactional work, in-house legal ops teams under 50 contracts/month, and any team that values configurable doctrine over vendor-defined defaults. Where it doesn't win on FIT: AmLaw 100 deal teams running multi-billion-dollar M&A closings with dedicated deal-room integration and 24/7 customer success — that's still Harvey or CoCounsel territory. (detailed comparison spoke)

The pickable side: at sub-$25/seat/month with playbook configurability the user owns, `/review-contract` is the cheapest competent contract review tool in the market for solo through mid-market firms in 2026. Spellbook and Harvey ship more pre-built workflows but at 8-80x the cost depending on tier.

Privilege and the 'assistance not advice' disclaimer

The Cowork plugin's outputs ship with explicit "assistance not advice" disclaimer language. That's not boilerplate — it's the integrity guardrail Anthropic built in deliberately. The plugin is a workflow tool, not counsel.

The context that matters: *United States v. Heppner* (SDNY, February 17, 2026) ruled that written exchanges between criminal defendant Bradley Heppner and consumer Claude were not protected by attorney-client privilege or work-product doctrine. The court reasoned that Claude isn't an attorney, so privilege doesn't attach, and Heppner generated the materials independently of counsel direction, so work-product doesn't either. (Heppner explainer)

Applying Heppner to `/review-contract`: the contract being reviewed is a client deliverable. The flagged output is Claude's analysis. If the analysis is shared between attorney and client at attorney's direction, work-product privilege likely attaches under standard doctrine. If the client uses `/review-contract` independently and shares the output with the attorney, the analytical content is closer to client-prepared work product — protection is fact-specific.

The operational fix: firms running `/review-contract` should run it on the Claude Team or Enterprise tier (Anthropic does not train on Team/Enterprise/API inputs per Anthropic's data handling page) and document the deployment surface in the engagement letter. For privileged matters, the Team tier ($20-$25/seat/month) is the minimum. The free or Pro consumer tier carries the Heppner-style risk that defaults to no privilege protection.

The second-order effect: most firms haven't updated their AI use policies to name model versions and deployment tiers since 2024. Cowork plugin deployment forces the policy update — the policy now needs to specify Claude Team or higher, the playbook governance, and the documentation expectation. (firm AI policy template spoke)

Real-world deployment: a 5-step rollout for a 25-attorney firm

Translation of the Cowork plugin into a deployable workflow for a typical mid-market firm:

Step 1 — Procure Claude Team. 25 attorneys at $20/seat/month annual = $500/month or $6,000/year (per Anthropic pricing). The Team tier includes admin controls and explicit data-protection guarantees Anthropic doesn't extend to Pro or consumer accounts. Setup time: 2 hours including SSO configuration.

Step 2 — Write the firm playbook. Senior partner or contracts-focused partner spends 4-6 hours documenting the firm's negotiation doctrine in YAML. Cover the top 8-10 clause categories the firm encounters most: indemnity, limitation of liability, IP, payment terms, termination, forum selection, change of control, confidentiality. The first version is rough; iterate over 6-8 weeks of real reviews.

Step 3 — Pilot on inbound NDAs and MSAs. Pick the 2-3 attorneys handling the most contract review work. Run `/review-contract` against the next 20-30 inbound contracts in parallel with traditional review. Track time savings and flag-accuracy errors. Iterate the playbook.

Step 4 — Document privilege and policy. Update the firm's AI use policy to specify Claude Team as the deployment surface, document the playbook governance process, and add a notation in matter-opening checklists about Claude usage on the matter. Engagement letters reference AI use generally per the firm's standard language.

Step 5 — Roll out firm-wide. After 6-8 weeks of pilot, expand to the full 25-attorney roster. Rollout time: 1 week including training. Total deployment cost: ~$6,000 in Claude licensing, ~12 hours of partner time on playbook configuration, plus a 6-8 week pilot window.

Compare to Spellbook ($180-$300/seat/month industry estimate per Artificial Lawyer = $54,000-$90,000/year for 25 attorneys, not vendor-confirmed) or Harvey ($1,200-$2,000+/seat/month industry estimate = $360,000-$600,000/year, not vendor-confirmed). The Cowork-on-Claude-Team stack runs at roughly 1-15% the cost depending on which incumbent you compare against. The tradeoff is the partner-time investment in playbook configuration. For most mid-market firms that's a multi-year ROI. (read the procurement checklist for mid-market firms)

The Bottom Line: My take: `/review-contract` is the cheapest competent clause-by-clause review tool in legal AI right now for solos through mid-market firms. The playbook configurability means the firm owns its negotiation doctrine in YAML rather than renting it from a vendor's proprietary database. For AmLaw 100 deal teams running multi-billion-dollar closings with dedicated customer success expectations, Spellbook or Harvey still win on workflow integration. For everyone else under that ceiling, the Cowork-on-Claude-Team stack at $6,000/year-for-25-attorneys is the procurement conversation that actually closes.