nda review
Use when reviewing one-way (unilateral) commercial NDAs, analyzing key clauses for risk, producing clause-by-clause issue logs with preferred redlines, fallbacks, and negotiation guidance.
Der Skill
# NDA Review Playbook (Commercial, Jurisdiction-Agnostic) Version 1.0 — December 2025 > This skill is a structured review playbook. It is not legal advice. When the NDA is high-risk, high-value, cross-border, or otherwise sensitive, escalate to qualified counsel. > **DRAFT — qualified counsel review required before signing.** > Reviewer of record: __________________________ (named lawyer, required before send) > Every output produced by this skill MUST carry this header verbatim and MUST leave the reviewer-of-record line in place until a named qualified lawyer has signed off. Do not send any redline, issue log, or summary to a counterparty until this line is filled. ## Overview | What this skill does | What it does not do | |---|---| | Reviews an NDA and outputs issues, risks, and suggested redlines | Provide jurisdiction-specific legal conclusions | | Supports *Recipient* or *Discloser* perspectives (user-chosen) | Guarantee enforceability | | Produces an executive summary + clause-by-clause markup guidance | Replace counsel for complex deals | **Scope:** supports **one-way (unilateral) commercial NDAs only**. If mutual, this playbook is out of scope. > **Variation callouts** appear throughout: M&A/Due diligence, Employment/contractor, Investor/VC ## Inputs to Collect (Ask Before Reviewing) ### A. Role and deal context (required) - Are we reviewing as **Recipient** (we receive confidential info) or **Discloser** (we disclose confidential info)? - Confirm the NDA is **one-way (unilateral)** — if mutual, stop: out of scope - What is the **purpose** / permitted use? - What are the **parties** (legal names) and any **affiliates**? - What information types are expected (tech, pricing, customer data, product roadmap, source code)? - Desired **timeline**: when do we need to sign? ### B. Practical constraints (recommended) - Do we need to share with affiliates, advisors, contractors, auditors, or potential acquirers? - Will we **export** data across borders or store in cloud tools? - Will any **personal data** be shared? > **Jurisdiction-agnostic note:** avoid asserting "this clause is invalid" without governing law; focus on *commercial risk*, *operational feasibility*, and *market norms*. ## Deliverables ### A. Executive Summary (1 page) - Party role (Recipient or Discloser) and confirmation it is one-way - Top 5 negotiation points (ranked) - "Sign as-is" / "Sign with changes" / "Escalate" recommendation ### B. Clause-by-Clause Issue Log | Clause | Issue (1 line) | Risk (H/M/L) | Preferred redline | Fallback | Rationale (1-2 sentences) | Owner | Deadline | |---|---|---:|---|---|---|---|---|---| | Definition | Overbroad; includes unmarked info with no reasonableness | | | | | | | | Term & survival | Perpetual confidentiality for all information | | | | | | | | Use restriction | Purpose too broad; blocks internal evaluation | | | | | | | | Disclosures | Representatives undefined; strict liability | | | | | | | | Return/destruction | No backup carve-out | | | | | | | | Remedies | One-way fees + automatic injunction | | | | | | | | Liability | Indemnity + unlimited consequential damages | | | | | | | | Boilerplate | Assignment prohibits change of control | | | | | | | ### C. Risk Band Rubric (How to Score H / M / L) Every row in the Issue Log MUST carry one of the three bands below. Use the most-severe band that applies; do not average. | Band | Criteria (any one triggers the band) | Typical examples | |---|---|---| | **High (H)** | Forced or one-way arbitration imposed on Recipient; unlimited liability or uncapped consequential damages; non-mutual indemnity that survives termination; perpetual confidentiality on *all* information with no trade-secret distinction; one-way attorneys' fees + automatic injunction against Recipient; standstill / no-hire / no-contact bundled into the NDA; cross-border personal-data flow with no carve-out; assignment clause that triggers on change of control. | "Recipient indemnifies Discloser for any and all claims"; "Recipient waives right to a jury and consents to arbitration in [foreign seat]"; "obligations survive in perpetuity". | | **Medium (M)** | Ambiguity in a defined term that could be read against us in negotiation but is not catastrophic; onerous-but-bounded scope (e.g., 5-year confidentiality on non-trade-secret info); missing standard carve-outs (compelled disclosure, prior knowledge, independent development) where the omission is recoverable in redline; representative-liability language without a written-confidentiality limiter; return/destruction language with no backup carve-out. | "Confidential Information includes any information disclosed in connection with the Purpose" (no marking standard); "Recipient shall be liable for any breach by its Representatives". | | **Low (L)** | Minor drafting cleanup — typos, defined-term capitalisation, cross-reference errors, stylistic inconsistencies; boilerplate that is non-standard but commercially harmless; clarifying tweaks that improve readability without shifting risk. | Inconsistent defined-term casing; redundant recitals; severability/notice-address mechanics. | > **Scoring discipline:** if you can credibly explain to a deal lead in one sentence why a clause could materially damage the business or expose the firm, it is at least Medium. If the explanation requires "and then if X, and then if Y…", it is Low. ## 5-Step Workflow ### Step 1 — Identify Stance (Recipient vs Discloser) - Confirm which side we are on for *this specific NDA* (titles are often misleading) - Confirm the NDA is **one-way (unilateral)**. If mutual, stop: out of scope. **Quick heuristic:** - If asked to keep *their* info secret → **Recipient** - If sharing *our* sensitive info → **Discloser** ### Step 2 — Triage the NDA (Fast Risk Scan) Flag immediately: - [ ] **Perpetual** confidentiality for *all* information (no trade secret distinction) - [ ] **Residuals clause** allowing use of "memory" or generalized knowledge - [ ] **Injunctive relief** + **attorneys' fees** one-way against Recipient - [ ] **Indemnity** for breach or broad third-party claims - [ ] **No carve-outs** for compelled disclosure or prior knowledge - [ ] **Overbroad definition**: "all information, whether marked or not" with no reasonableness - [ ] **Affiliate coverage** missing when we must share internally ### Step 3 — Clause-by-Clause Review Use reference modules: - `references/KEY_CLAUSES.md` — Common NDA clauses and implications - `references/PARTY_OBLIGATIONS.md` — Analysis of party obligations - `references/DURATION_SCOPE.md` — Duration and scope considerations - `references/REMEDIES_LIABILITY.md` — Remedies and liability provisions - `references/STANDARD_EXCEPTIONS.md` — Standard exceptions ### Step 4 — Draft Redlines and Negotiation Positions For each issue: - **Preferred redline** (best risk outcome) - **Fallback position** (acceptable compromise) - **Rationale** (1-2 sentences: business + operational feasibility) - **Owner** (Legal, Sales, Security, Product) - **Deadline** > **Negotiation discipline:** do not propose 20 changes. Focus on 5-10 that materially change risk. ### Step 5 — Finalize the Package - [ ] Ensure consistency across definitions - [ ] Confirm operational feasibility - [ ] Re-scan Step 2 triage list; ensure each flagged item is in the issue log - [ ] Provide "what we changed and why" summary ## Perspective-Specific Checklists ### A. Recipient Checklist (Incoming NDA) | Topic | Red Flags | Typical Ask | |---|---|---| | Definition of Confidential Information | Overbroad; includes independently developed info; no marking standard | Add reasonableness + identification standard | | Purpose / Permitted Use | Any use restriction beyond evaluation; bans on internal sharing | Tie to stated purpose; allow internal need-to-know | | Representatives | Liable for any representative breach without control | Limit to written confidentiality; commercially reasonable care | | Term & survival | Perpetual for everything; unclear start date | Fixed term; longer only for trade secrets | | Return / destruction | Requires immediate deletion of backups | Add backup carve-out | | Remedies | One-way fees + broad injunction language | Mutuality or reasonableness | | Liability / indemnity | Indemnity; unlimited damages; consequential damages | Cap or exclude categories; remove indemnity | | Residuals | Allows use of "retained in memory" | Delete or narrow heavily | > **M&A / Due diligence:** ensure diligence sharing (advisors, financing, affiliates) is permitted and data room exports/notes are covered. ### B. Discloser Checklist (When Sharing Sensitive Info) | Topic | Red Flags | Typical Ask | |---|---|---| | Definition | Too narrow; requires marking only; excludes oral disclosures | Add oral confirmation mechanism | | Security standard | Only "reasonable" with no baseline | Add minimum safeguards | | Exclusions | Too broad (e.g., "independently developed" with no proof) | Require written evidence | | Term & survival | Too short | Extend for sensitive categories | | Remedies | No equitable relief, no fees | Add equitable relief carefully | > **Investor / VC:** watch for standstill, solicitation, and "no contact" provisions. ## Limitations - This skill provides a structured framework, not legal advice - Jurisdiction-specific law not covered; always verify with qualified counsel - High-risk, high-value, or cross-border deals require escalation - Does not cover mutual NDAs — those require separate review approach - All outputs must be reviewed by a qualified legal professional before use ## QA Remediation (LegalQuants, 2026-05) This skill was QA'd by LegalQuants against the Legal Skill Design Framework on 2026-05-11 (verdict: SOME CONCERN) and remediated on 2026-05-12. The remediations target the two open gaps from that report — undefined H/M/L risk bands and an accountability gap not structurally enforced by the output shape — while leaving the technical content intact. **What changed** - **H/M/L risk bands now operationalised.** A new section "Risk Band Rubric (How to Score H / M / L)" was added immediately after the Clause-by-Clause Issue Log table, with explicit criteria and worked examples for each band. The rubric is mandatory for every Issue Log row. - High = forced arbitration, unlimited or uncapped liability, non-mutual indemnity, perpetual confidentiality on all information, one-way fees + injunction, bundled standstill/no-hire, cross-border personal-data flow without carve-out, change-of-control assignment trigger. - Medium = ambiguity in defined terms that is recoverable in redline, onerous-but-bounded scope, missing standard carve-outs, representative-liability without a written-confidentiality limiter, return/destruction without backup carve-out. - Low = minor drafting cleanup (typos, defined-term casing, cross-reference errors, harmless boilerplate, readability tweaks that do not shift risk). - **Accountability gap closed by output shape.** A "DRAFT — qualified counsel review required before signing" banner with a named-reviewer placeholder is now part of the file header and is required on every output produced by the skill (Executive Summary, Issue Log, Redline package, Step 5 finalisation). The banner cannot be removed until a named qualified lawyer is written into the reviewer-of-record line; nothing leaves for the counterparty until that line is filled. This makes the lawyer-review requirement structurally enforced rather than purely a disclaimer the reader can skim past. - **Versioning metadata refreshed.** Frontmatter now carries `version: 1.0.0`, `last_reviewed: 2026-05`, and `last_reviewed_by: LegalQuants (QA remediation)`. Authorship remains with Jamie Tso. **What did not change** - Scope (one-way commercial NDAs only), jurisdiction-agnostic posture, inputs to collect, 5-step workflow, reference modules, perspective-specific checklists, variation callouts, and the existing limitations section are all preserved verbatim. The remediation is additive. **Open items deferred to a later pass** - Audience declaration, work-shape declaration, consolidated escalation section with named-role routing, and a "common failure modes" list (all flagged as ⚠️ in the QA report) are not addressed in this remediation. They are non-blocking for the two priority gaps and can be folded into the next minor version.
So wird es eingerichtet
Claude Code / Claude Desktop 1. Create a folder named after the skill inside your skills directory: - Claude Code: `~/.claude/skills/<skill-name>/` - A project-local skill: `<project>/.claude/skills/<skill-name>/` 2. Save the text below as `SKILL.md` in that folder, keeping the `---` frontmatter block at the top — the `name` and `description` fields are what make the skill discoverable. 3. Restart Claude Code, or run `/skills` to confirm it is listed. The skill loads only when its `description` matches what you are doing, so a precise description matters more than a long body.
Hat es bei Ihnen geklappt?
Zum Schreiben anmelden.Woher es stammt
2026 Legal Quants