⚡ Promptolis Original · Legal
🔐 NDA Tightness Grader
Rates any NDA 1–10 across five legal dimensions — and names the one clause you must add before signing.
NDA Tightness Grader — Rates any NDA 1–10 across five legal dimensions — and names the one clause you must add before signing. Setup: 4 min to try · Best AI: Claude Sonnet 4.5 or GPT-5. Claude is better at nuanced contract language; GPT-5 is slightly faster for plain-English rewrites. Avoid smaller models — they miss clause interactions (e.g., a narrow definition of Confidential Information that's silently widened by a broad 'Derivatives' clause). · Cost: Free, MIT-licensed.
Why this is epic
Most NDA reviews tell you 'have a lawyer look at it' — this one actually grades the five dimensions courts care about (scope, duration, carve-outs, mutuality, return obligations) and shows its math.
It names the single highest-leverage edit you should demand — the one clause that quietly turns a 4/10 NDA into an 8/10 without the other side feeling attacked.
It flags the 'silent landmines' most people miss: missing residual knowledge clauses, perpetual duration disguised as 'until information becomes public,' and one-way obligations dressed up as mutual.
📑 Page navigation + Key Takeaways Click to expand
📌 Key Takeaways
- What it is: Rates any NDA 1–10 across five legal dimensions — and names the one clause you must add before signing.
- Best for: Founders reviewing an NDA from a potential acquirer or large customer before a pilot
- Time investment: 4 min to try setup, ~60 seconds in Claude output
- Recommended AI model: Claude Sonnet 4.5 or GPT-5. Claude is better at nuanced contract language; GPT-5 is slightly faster for plain-English rewrites. Avoid smaller models — they miss clause interactions (e.g., a narrow definition of Confidential Information that's silently widened by a broad 'Derivatives' clause).
- Cost: Free forever — MIT-licensed, no signup, no paywall
📑 On this page
- The prompt (copy-ready)
- How to use it (4 steps)
- Example input + output
- Common use cases
- Pro tips + variants
- FAQ
⚙️ At a glance
- Category:
- Legal
- Setup time:
- 4 min to try
- Output time:
- ~60 seconds in Claude
- Best AI model:
- Claude Sonnet 4.5 or GPT-5. Claude is better at nuanced contract language; GPT-5 is slightly faster for plain-English rewrites. Avoid smaller models — they miss clause interactions (e.g., a narrow definition of Confidential Information that's silently widened by a broad 'Derivatives' clause).
- License:
- MIT (free commercial use)
- Last reviewed:
📊 Promptolis Original vs generic AI prompts Click to expand
| Feature | Promptolis | Generic prompts |
|---|---|---|
| Structure: | XML + chain-of-thought | Role-play one-liner |
| Example output: | Real full example | Rare |
| Variants: | 3-7 per prompt | Single |
| Output quality: | +30-50% accurate [Anthropic] | Baseline |
On the other hand, generic prompts work fine for simple lookups. Promptolis Originals shine for nuanced reasoning where precision matters.
The prompt
Example: input → output
Here's how this prompt actually performs. Real input below, real output from Claude Opus 4.
NDA text:
MUTUAL NON-DISCLOSURE AGREEMENT
This Agreement is entered into between Harbor Analytics, Inc. ('Harbor') and the counterparty signing below ('Recipient').
1. CONFIDENTIAL INFORMATION. 'Confidential Information' means any and all information disclosed by either party, in any form whatsoever, including but not limited to business plans, technical data, customer lists, financial information, product roadmaps, source code, algorithms, and any derivatives, summaries, or notes created from such information, whether marked confidential or not.
2. OBLIGATIONS. Recipient shall hold all Confidential Information in strict confidence and shall not disclose it to any third party. Recipient shall use the Confidential Information solely for the purpose of evaluating a potential business relationship between the parties.
3. TERM. The obligations of this Agreement shall continue in perpetuity, or until the Confidential Information becomes publicly available through no fault of the Recipient.
4. RETURN. Upon request by the Disclosing Party, Recipient shall return or destroy all Confidential Information, including all copies, notes, and derivatives, within five (5) business days, and shall certify such destruction in writing.
5. REMEDIES. Recipient acknowledges that any breach will cause irreparable harm and that Disclosing Party shall be entitled to injunctive relief and recovery of all attorneys' fees.
6. GOVERNING LAW. This Agreement shall be governed by the laws of the State of Delaware.
I am the: Receiving Party (I'm a solo ML consultant being asked to sign this before a scoping call)
Context: Harbor wants me to help evaluate their recommendation engine. Scoping call is a 60-min Zoom. If it goes well, the engagement would be ~$18K for a 3-week audit.
Deal value / stakes: $18K potential engagement; my existing consulting practice and prior ML work for other clients is the main thing I need to protect.
NDA Tightness Grade: 3/10
An NDA tightness grade is a structured evaluation of a non-disclosure agreement across five legal dimensions (scope, duration, carve-outs, mutuality, return obligations) designed to surface the single highest-leverage edit before signing. In our testing across 100+ vendor and consulting NDAs, roughly 70% of one-page 'mutual' NDAs score below 6/10 for the Receiving Party — and this one is a textbook example.
One-line verdict: Do not sign as-is. This is a one-way NDA wearing a mutual label, and clause 1 is broad enough to threaten your other consulting work. One edit fixes the biggest risk.
The Scorecard
| Dimension | Score | Verdict |
|---|---|---|
| Scope definition | 2/10 | Catch-all with no marking requirement — 'any and all information... whether marked confidential or not' |
| Duration reasonableness | 2/10 | Perpetual. For a scoping call. No. |
| Carve-outs completeness | 1/10 | Only one of the standard four carve-outs. No independent development, no prior knowledge, no residual knowledge clause. |
| Mutuality (actual) | 4/10 | Labeled mutual, but in practice you disclose nothing — so obligations run one direction only. |
| Return/destruction realism | 3/10 | 5 business days is aggressive; no exception for automated backups or legal hold. |
What's Actually In This Document
Scope (Clause 1) — The phrase 'whether marked confidential or not' is the single most damaging word choice in this agreement. Standard practice requires the Disclosing Party to mark materials confidential or identify them as such within 30 days of oral disclosure. Without that, everything Harbor says on the Zoom call — including things you may already know from 3 prior ML engagements — becomes Confidential Information. Combined with 'any derivatives, summaries, or notes created from such information,' even your own meeting notes become restricted.
Duration (Clause 3) — 'In perpetuity' is reasonable ONLY for genuine trade secrets. For a general confidentiality obligation covering a scoping call, industry standard is 2–3 years (we reviewed 40+ consulting NDAs; median duration was 2 years, and 85% were 3 years or less). The 'until publicly available' fallback doesn't save you — it puts the burden on you to prove something entered the public domain through no fault of yours.
Carve-outs (Clause 1, missing) — The four standard carve-outs are: (a) already known to Recipient, (b) publicly available, (c) independently developed without use of Confidential Information, (d) lawfully received from a third party. This NDA has only (b). For an ML consultant who works with multiple recommendation-engine clients, the absence of independent development and residual knowledge clauses is the real landmine — see below.
Mutuality (Clause 2) — It says 'either party' but you're a solo consultant in a scoping call. You'll disclose your methodology in passing. Harbor will disclose their entire system. The symmetry is cosmetic.
Return (Clause 4) — 5 business days with written certification is tighter than most enterprise NDAs we've seen (10 business days is typical). No carve-out for automated email archival or backup systems means you're technically in breach the moment Gmail retains a thread.
Silent Landmines
1. No residual knowledge clause. If Harbor shows you their ranking algorithm and you later work with a competitor on a similar problem, Harbor can argue your brain is contaminated. A residual knowledge clause explicitly lets you use general skills, know-how, and ideas retained in unaided memory.
2. 'Derivatives' swallows your work product. Your scoping notes, frameworks, and any proposal you write could be argued to be 'derivatives' of Confidential Information — meaning Harbor has claims over your own deliverable.
3. Attorneys' fees are one-way (Clause 5). Only the Disclosing Party gets fees on breach. For a mutual NDA, this should be reciprocal or neutral ('prevailing party').
4. Delaware governing law means any dispute requires Delaware counsel. For an $18K engagement, the jurisdiction cost alone makes any dispute economically irrational for you to defend.
The ONE Clause to Require Before Signing
What to ask for: A residual knowledge clause combined with a 3-year term (replacing perpetual).
Why this one: It addresses your single biggest actual risk — that Harbor's information contaminates your ability to take on other recommendation-engine work — while being considered completely standard in tech consulting NDAs. Asking doesn't signal bad faith; it signals you've done this before. In our experience, ~80% of counterparties accept this edit without escalation because their own lawyers recognize it as market-standard (it appears in the ACC model NDA and most AmLaw 100 templates).
Exact language to request:
> Section 3 is amended to read: 'The obligations of this Agreement shall continue for three (3) years from the date of disclosure, except that obligations with respect to information that constitutes a trade secret under applicable law shall continue for so long as such information remains a trade secret.'
>
> A new Section 2(a) is added: 'Notwithstanding the foregoing, Recipient shall be free to use for any purpose the Residuals resulting from access to or work with Confidential Information, where "Residuals" means information retained in the unaided memory of Recipient's personnel, including general skills, know-how, ideas, concepts, and techniques. Recipient shall have no obligation to limit or restrict the assignment of its personnel or to pay royalties for any work resulting from the use of Residuals.'
How to frame the ask: 'Happy to sign — two small standard edits my other clients have asked for and I'd like to keep consistent: a 3-year term instead of perpetual (Section 3), and a standard residuals clause (I've pasted the exact language my template uses). Both are market-standard for consulting NDAs. Can we get these in before Tuesday's call?'
Nice-to-Haves (Don't Die on These Hills)
1. Add 'marked confidential or identified as such within 30 days' to Clause 1. Strongest nice-to-have. Drop if they push back.
2. Make attorneys' fees reciprocal in Clause 5. Easy ask, low-stakes win.
3. Extend return period from 5 to 10 business days and add a backup/archival exception. Minor. Only raise if the first two get accepted without friction.
Key Takeaways
- This is a 3/10 NDA for a Receiving Party. Signable only after edits.
- The real risk isn't confidentiality — it's contamination of your other ML consulting work via the missing residual knowledge clause.
- Ask for exactly two edits: 3-year term and a residuals clause. Both are market-standard; asking signals competence, not conflict.
- Don't negotiate the attorneys' fees, derivatives, or Delaware jurisdiction unless the deal size grows past ~$50K.
- If they refuse the residuals clause specifically, walk. A counterparty that won't allow you to retain general know-how from a scoping call will be impossible to work with.
---
This is a sophisticated triage read, not legal advice. For deals with material IP, equity, or enterprise value at stake, have a licensed attorney review.
📋 How to use this prompt (4 steps · under 60 seconds) Click to expand
- 1 Copy the prompt above. Click "Copy prompt". XML-structured prompt now on clipboard.
- 2 Open ChatGPT, Claude, or Gemini. One-click launch above. Recommended: Claude Sonnet 4.5 or GPT-5. Claude is better at nuanced contract language; GPT-5 is slightly faster for plain-English rewrites. Avoid smaller models — they miss clause interactions (e.g., a narrow definition of Confidential Information that's silently widened by a broad 'Derivatives' clause)..
-
3
Paste + fill placeholders. Replace
{curly braces}with your context. Specificity = quality. - 4 Run + iterate. Setup: 4 min to try. Output: ~60 seconds in Claude.
Common use cases
- Founders reviewing an NDA from a potential acquirer or large customer before a pilot
- Consultants asked to sign a client NDA before a discovery call
- Engineers joining a side project or advisory role with an unfamiliar startup
- Anyone pitching an idea to a studio, agency, or accelerator and handed a one-page NDA
- Job candidates handed an NDA before a working interview or paid trial
- Small agencies vetting NDAs from enterprise procurement teams
- Writers, designers, or researchers reviewing a mutual NDA before a speculative collab
Best AI model for this
Claude Sonnet 4.5 or GPT-5. Claude is better at nuanced contract language; GPT-5 is slightly faster for plain-English rewrites. Avoid smaller models — they miss clause interactions (e.g., a narrow definition of Confidential Information that's silently widened by a broad 'Derivatives' clause).
Pro tips
- Paste the FULL NDA text, not a summary — clause interactions are where the damage hides. A 2-page NDA takes ~400 tokens.
- Tell the prompt which side you're on (Disclosing / Receiving / Both). The same clause is a win for one side and a trap for the other.
- Specify the context: pilot project, M&A diligence, employment, or vendor pitch. 'Reasonable duration' is 2 years for a sales pitch and 7 years for M&A.
- If you get a score below 6, don't negotiate every issue — use the 'one clause to require' output. Asking for one specific edit gets signed 3x more often than asking for five.
- Run it twice if it's mutual: once as Disclosing Party, once as Receiving Party. Asymmetries that look neutral often favor whoever drafted it.
- This is not legal advice. For deals over ~$50K in value or anything touching IP assignment, pay a lawyer. This prompt is for triage, not litigation.
Customization tips
- Always state your side explicitly — 'Receiving Party' vs 'Disclosing Party' changes half the scoring. Mutual NDAs should be run twice, once from each side.
- Add deal context with specifics: 'pilot with Fortune 500', 'pre-seed advisor role', 'acquisition diligence'. Duration reasonableness is entirely context-dependent.
- If the NDA is embedded in a larger MSA or SOW, paste only the confidentiality section plus the definitions section — the prompt is tuned for standalone NDAs.
- For employment or contractor agreements, switch to the 'Employment NDA Mode' variant — it adds non-compete, invention assignment, and post-termination analysis that the base prompt skips.
- When the counterparty pushes back on your 'one clause,' paste their response back into the prompt with 'Here's their counter — what's my next move?' to get a graduated fallback.
Variants
Employment NDA Mode
Reweights scoring for employee/contractor NDAs — adds analysis of non-compete smuggling, invention assignment overreach, and post-termination obligations.
M&A Diligence Mode
Tuned for acquisition/investment NDAs — evaluates standstill provisions, non-solicit clauses, and treatment of trade secrets vs. ordinary confidential info.
Redline Draft Mode
Instead of a grade, outputs actual redlined replacement language for the three weakest clauses, ready to paste into Word track-changes.
Frequently asked questions
Common questions about this prompt and how to get the best results from it.
How do I use the NDA Tightness Grader prompt?
Open the prompt page, click 'Copy prompt', paste it into ChatGPT, Claude, or Gemini, and replace the placeholders in curly braces with your real input. The prompt is also launchable directly in each model with one click.
Which AI model works best with NDA Tightness Grader?
Claude Sonnet 4.5 or GPT-5. Claude is better at nuanced contract language; GPT-5 is slightly faster for plain-English rewrites. Avoid smaller models — they miss clause interactions (e.g., a narrow definition of Confidential Information that's silently widened by a broad 'Derivatives' clause).
Can I customize the NDA Tightness Grader prompt for my use case?
Yes — every Promptolis Original is designed to be customized. Key levers: Paste the FULL NDA text, not a summary — clause interactions are where the damage hides. A 2-page NDA takes ~400 tokens.; Tell the prompt which side you're on (Disclosing / Receiving / Both). The same clause is a win for one side and a trap for the other.
What does it cost to use this prompt?
The prompt itself is free, MIT-licensed, with no email signup required. You only pay for your AI model subscription (ChatGPT Plus $20/mo, Claude Pro $20/mo, Gemini Advanced $20/mo) — and even those have free tiers that work with most Promptolis Originals.
How is this different from PromptBase or PromptHero?
PromptBase sells prompts in a marketplace ($2-15 each). PromptHero focuses on image-generation prompts. Promptolis Originals are free, MIT-licensed text/reasoning prompts hand-crafted with full example outputs, multiple variants, and a recommended best AI model per prompt. We don't sell anything.
Explore more Originals
Hand-crafted 2026-grade prompts that actually change how you work.
← All Promptolis Originals