Anthropic Cowork Legal Plugin Triage NDA Rapid Review

Anthropic's `/triage-nda` skill is the high-volume sibling to `/review-contract` inside the open-source Cowork legal plugin released in February 2026. It does rapid pre-screening on inbound NDAs — categorizing each document into standard-approval, counsel-review, or full-review buckets based on configured firm risk tolerance. For in-house legal teams getting 50-300 inbound NDAs per month, that's the difference between a paralegal working through a Friday-night queue and Claude returning sorted output in 30-60 seconds per document. The skill is open source, configurable via YAML, runs inside any Claude tier from Free up, and ships with the same "assistance not advice" disclaimer as `/review-contract`. This is the operator's read on what `/triage-nda` actually does, when it's the right tool, and where the configuration tradeoffs sit.

What /triage-nda does and why volume changes the math

Per Anthropic's plugin page and the GitHub repo, the `/triage-nda` skill targets one specific bottleneck: the NDA pipeline that floods most in-house legal teams. The workflow:

Configure the risk tolerance. YAML playbook defines what kinds of NDAs auto-approve (mutual NDA, fixed term under 3 years, standard carve-outs, US jurisdiction), what kinds escalate to counsel review (one-way NDAs from large counterparties, jurisdictional outliers, perpetual confidentiality periods), and what kinds require full attorney review (M&A-adjacent NDAs, NDAs with non-standard injunctive relief language, anything tied to active litigation).

Feed the NDA. Drop the document into Claude. Invoke `/triage-nda`. Claude returns a categorization in 30-60 seconds: STANDARD APPROVAL, COUNSEL REVIEW, or FULL REVIEW, plus the specific clauses or terms that triggered the categorization.

Route accordingly. STANDARD APPROVAL goes to a paralegal for execution. COUNSEL REVIEW goes to an associate for a 15-30 minute targeted review. FULL REVIEW goes to a senior attorney for the traditional comprehensive read.

The volume math: an in-house legal team handling 100 inbound NDAs/month spends roughly 20-30 hours doing first-pass categorization manually. `/triage-nda` compresses that to 1-2 hours of human review on the COUNSEL REVIEW and FULL REVIEW outputs. That's 18-28 hours/month of legal ops time recovered. At a $150/hour blended in-house rate, that's $2,700-$4,200/month in recovered capacity per legal ops headcount. The plugin pays for the underlying Claude Team license ($20-$25/seat/month) in the first hour of the first month.

How to configure the playbook for inbound NDAs

The playbook YAML for `/triage-nda` is shorter than `/review-contract` because NDAs are narrower in clause variety. A typical 20-30 line playbook covers:

```yaml auto_approve: - type: mutual_nda term_max: 3_years jurisdiction: [US, UK, Canada, EU member state] carve_outs_present: standard_set counterparty_type: [vendor, customer, partner] counsel_review: - type: one_way_nda counterparty_size: greater_than_500m_revenue - jurisdiction_outside: [US, UK, Canada, EU] - confidentiality_term: perpetual - injunctive_relief_language: non_standard full_review: - ma_adjacent: true - litigation_active: true - confidentiality_term_with_specific_carveout: trade_secret_definition_modified - special_provisions: [non_solicitation, non_compete, residual_clause] ```

The playbook is checked into git and version-controlled. When the GC or senior in-house counsel updates risk tolerance — say, after a year of seeing counterparty pushback patterns — the YAML updates and every subsequent triage runs against the new doctrine.

The second-order effect: in-house teams that build out their `/triage-nda` playbook over 6-12 months end up with a documented, machine-readable representation of their actual NDA risk tolerance. That's an institutional knowledge asset most legal departments have only ever held in senior counsel's head. The third-order effect: when the senior counsel leaves or rotates, the playbook stays. Onboarding the replacement takes hours instead of months because the doctrine is documented in YAML, not tribal knowledge.

When /triage-nda is the right tool — and when it isn't

Where `/triage-nda` wins on FIT:

- In-house legal teams handling 30+ inbound NDAs/month. Volume is the use case. Below 30/month, the playbook configuration overhead (4-8 hours of senior counsel time) doesn't pay back fast. - Procurement-driven NDA pipelines. Vendors, customers, partners, and channel-partner NDAs follow patterns. Patterns are what playbooks encode well. - Legal ops teams with paralegal capacity. The output is most useful when a paralegal or contract specialist handles the STANDARD APPROVAL bucket and an associate handles COUNSEL REVIEW. Without that downstream capacity, the categorization saves time but not workflow. - Teams already running on Claude Team or Enterprise. Free and Pro tiers work technically but carry the Heppner-style privilege risk for any sensitive matter. (Heppner explainer)

Where `/triage-nda` doesn't win on FIT:

- High-stakes M&A diligence NDAs. Single-document, transaction-specific NDAs with complex carve-outs and jurisdictional provisions need partner-grade review, not triage. Use `/review-contract` for the clause-by-clause work, but the triage step adds little value here. - Litigation-adjacent confidentiality agreements. Protective orders, settlement-related NDAs, and matter-specific confidentiality stipulations are case-specific. Playbook categorization can miscategorize these. - Firms without a documented NDA risk tolerance. If the senior counsel can't articulate the risk thresholds, the playbook can't either. Time-spent on configuration goes to waste until the doctrine is clear.

The pickable side on FIT: for in-house legal ops teams handling vendor/customer NDA volume, `/triage-nda` plus a Claude Team license is the cheapest competent solution in the market. For BigLaw transactional teams handling deal-specific NDAs, the value is closer to a 2x time-saver, not a 10x — `/review-contract` is the better skill for that workflow. (deeper read in the Cowork vs vendor plugins comparison)

Cost comparison: /triage-nda vs LinkSquares vs Ironclad vs Spellbook

Inbound NDA triage is a category vendors have built into broader contract lifecycle management (CLM) platforms. The pricing reality:

LinkSquares. Quote-only enterprise pricing per LinkSquares site. Industry estimates suggest $30,000-$80,000/year for mid-market deployments depending on contract volume and CLM module set. The platform includes broader CLM (storage, version control, e-signature integration) plus AI-driven triage and review.

Ironclad. Quote-only enterprise pricing per Ironclad site. Industry estimates suggest similar to LinkSquares — $30,000-$100,000+/year for mid-market through enterprise. Strong on workflow automation across the contract lifecycle, not NDA-specific.

Spellbook. Quote-only per Spellbook pricing. Industry estimates per Artificial Lawyer reporting suggest $180-$300/seat/month with a $199/seat/month enterprise minimum at 10 seats, not vendor-confirmed. For a 5-person in-house team, that's roughly $12,000-$18,000/year.

Anthropic /triage-nda via Cowork. Open source, free. Cost is the underlying Claude tier — Team at $20/seat/month annual = $1,200/year for a 5-person in-house team (per Anthropic pricing).

The pickable side on FIT: for pure NDA triage, `/triage-nda` is 10-60x cheaper than the CLM alternatives. The tradeoff is that LinkSquares and Ironclad ship a full contract lifecycle platform — storage, repository, audit trail, e-signature, vendor management — in addition to triage. If the firm needs the broader CLM capability, the platform pricing makes sense. If the firm needs NDA triage and is fine routing storage/audit-trail through existing tools (Word + SharePoint + DocuSign), `/triage-nda` is the procurement decision that closes.

Privilege, training data, and the deployment tier decision

The `/triage-nda` plugin runs the same data-handling rules as the rest of Cowork. Per Anthropic's data handling page, Anthropic does not train on Team, Enterprise, or API inputs. That guarantee does not extend to Free or Pro consumer accounts.

For an in-house legal team running `/triage-nda` on inbound vendor NDAs, the privilege calculus is generally lower-stakes than a litigation context because the documents are pre-execution and not yet privileged. But the inputs to Claude reveal counterparty identities, contract terms, and risk-tolerance thresholds — operationally sensitive information the firm doesn't want flowing into model training data.

The operational rule: run `/triage-nda` on Claude Team or higher. The cost differential between Pro ($17-$20/user/month) and Team ($20-$25/seat/month) is small enough that the data-handling guarantee is worth it on its own. For larger teams or firms with stricter procurement requirements, Enterprise ($20/seat/month plus usage at API rates per Anthropic Enterprise pricing) adds advanced security and compliance controls.

The second-order effect: most firm AI use policies don't yet name Cowork plugin deployment specifically. After deploying `/triage-nda`, the policy needs an update — naming the deployment tier (Team or higher), the playbook governance process, and the documentation expectation for any NDA categorized as FULL REVIEW. (firm AI policy template spoke)

The third-order effect: in-house teams using `/triage-nda` build a documented decision trail (Claude flagged this NDA as STANDARD APPROVAL, paralegal executed). For audit purposes — internal SOC 2, external customer security questionnaires — that's a positive signal. The triage tool is generating its own compliance evidence.

The Bottom Line: My take: `/triage-nda` is the cheapest competent NDA triage tool for in-house legal teams handling 30+ inbound NDAs per month. At $1,200-$3,000/year in Claude Team licensing for a 5-person team versus $12,000-$80,000+/year for Spellbook, LinkSquares, or Ironclad, the procurement math closes fast. Use it where volume justifies playbook configuration. Skip it for low-volume teams under 30 NDAs/month and for deal-specific M&A NDAs where partner-grade review is the requirement.