scenario-test
Skill0 starsChanged 32 days agoMITView on GitHub
Stress-test a contract, deal structure or piece of legal advice by running it through realistic future events (late payment, scope creep, termination, insolvency, a founder leaving, a data breach, the business being sold, deadlock, a law change) and tracing which clauses fire, where the document is silent, what default law fills the gap, and what the client actually ends up with. Verdict per scenario (Handled, Partly handled, Gap, Conflict), ranked fixes with exact wording, and "fix it" to apply them as tracked changes. Use when reviewing or drafting any contract, shareholders agreement, lease, employment or supply arrangement, and whenever the user asks "what happens if", "how protected am I", "stress test", "war-game", "worst case" or "scenario". Also sanity-checks advice by flipping the facts it depends on, and war-games a negotiation before redlines go out.
In this repository27 other files it gives its agents.
- adversarial-reviewSkill
- business-saleSkill
- business-setupSkill
- cite-checkSkill
- contract-reviewSkill
- disputesSkill
- doc-auditSkill
- draft-contractSkill
- employment-checkSkill
- fundraisingSkill
- ip-brandSkill
- jurisdiction-adaptSkill
- law-auSkill
- law-change-impactSkill
- law-nzSkill
- law-ukSkill
- law-usSkill
- lease-reviewSkill
- legal-adviceSkill
- legal-health-checkSkill
- legal-researchSkill
- marketing-complianceSkill
- nda-triageSkill
- privacy-complianceSkill
- redlineSkill
- sprintlaw-modeSkill
- website-complianceSkill
The file
We split this skill into the rules it gives an agent
4 rules. About 0 are common, 3 are distinctive, and 1 is very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈3
Distinctive · estimate
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 15
Run it after (or alongside) contract-review, inside draft-contract before a draft is final, and inside legal-advice when the answer turns …
- Distinctive: line 26
Before imagining anything, turn the document into a machine you can run.
- Distinctive: line 141
Do not invent facts to make a scenario dramatic.
- Distinctive: line 142
Do not invent law to fill silence.
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Scenario Test
A checklist review finds clauses that are missing. A scenario test finds clauses that exist but fail when real life happens: the indemnity the liability cap quietly swallows, the termination right that leaves the client unpaid for work already done, the notice clause nobody will comply with because everyone actually uses WhatsApp. Good lawyers run these simulations in their heads. This skill makes them explicit, repeatable and visible to the client.
Run it after (or alongside) contract-review, inside draft-contract before a draft is final, and inside legal-advice when the answer turns on facts.
Inputs
- The document(s). Read every word, including schedules, definitions and anything incorporated by reference. If something is incorporated but not supplied (a "Master Terms", a policy on a website, an award), list it as a gap.
- Whose side you are on. If unknown, ask once; if still unknown, run the test from both sides and label it.
- The client's business context: what they sell, rough deal value, who has leverage, what they fear. One or two questions at most; otherwise state assumptions.
- Governing law. Load the matching jurisdiction skill (
law-au,law-uk,law-nz,law-us) because silence in a contract is filled by statute, implied terms and case law, and that is where most scenario outcomes are decided.
Step 1. Build the contract model
Before imagining anything, turn the document into a machine you can run. Produce this table internally (show it to the user only if asked or if the document is long and the map helps):
| Component | What to capture |
| Parties and roles | Legal names, entity types, who is the client, guarantors, related entities |
| Term | Start, end, renewal (automatic?), minimum term, lock-ins |
| Obligations | Per party: what, by when, to what standard, measured how (clause refs) |
| Money | Price, invoicing trigger, payment terms, late-payment interest, set-off, price changes, deposits, refunds |
| Conditions and triggers | Conditions precedent, acceptance testing, milestones, deemed acceptance |
| Discretions | Anything one party may decide alone ("acting reasonably" or not) |
| Termination | Who can terminate, for what, notice, cure periods, termination for convenience |
| After termination | Accrued rights, payment for work done, return of data and property, transition, survival |
| Risk allocation | Liability caps (amount, what is inside and outside), exclusions, indemnities, insurance |
| IP and data | Ownership, when it vests, licences back, confidentiality, privacy obligations |
| Disputes | Escalation, mediation, arbitration or courts, venue, governing law |
| Boilerplate that bites | Notices, assignment and change of control, variation, entire agreement, waiver, severance, counterparts |
Note every definition that changes the plain meaning of a clause. Defined terms are where scenarios most often go wrong.
Step 2. Choose the scenarios
Pick 8 to 15 scenarios. Fewer for a simple NDA, more for a shareholders agreement or major supply deal. Draw from three decks:
- Core deck (always consider): the client is not paid on time; the client wants out early; the other side wants out early; the work or goods are disputed as defective; something goes wrong and a third party sues; the counterparty becomes insolvent; the client's business is sold or restructured; the relationship sours and nobody is talking.
- Type deck: the scenarios for this kind of document in references/scenario-library.md.
- Bespoke deck: two to four scenarios specific to this client's business. A café lease needs "the landlord redevelops the centre"; a SaaS reseller needs "the vendor discontinues the product"; a founders agreement needs "one founder stops working but keeps their shares".
Every run must include at least one of each: a money scenario, an exit scenario, a blame scenario, an external-shock scenario, a counterparty-failure scenario, and a success scenario (the client grows tenfold, wins a huge customer, gets acquired, raises money). Contracts drafted for a small business routinely break on success: exclusivity that blocks the big deal, a change-of-control clause that kills the sale, a cap that was fine at $5k a month and absurd at $500k.
Rank scenarios by likelihood x impact and run the top ones in full. List the rest in one line each.
Step 3. Run each scenario as a timeline
For each scenario, write the timeline the way it would actually unfold:
- Trigger (Day 0). What happens, concretely. Use a realistic fact pattern with numbers.
- Moves. What each party may do and must do next, in order, citing clause numbers. Include practical moves (stop work, withhold payment, call the bank guarantee) not just legal ones.
- Silence and ambiguity. Where the contract does not answer the question, or answers it two ways. Quote the words that create the ambiguity.
- Default law. What fills the silence: statute, implied terms, equitable doctrines, consumer or employment protections that override the contract. Take this from the jurisdiction skill and verify anything you rely on (P4 in
sprintlaw-mode, andcite-check). If you cannot verify, say so. - End state for the client. Money in or out, time lost, control kept or lost, relationship, reputation. Give an exposure range with the assumptions that drive it. Rough is fine; false precision is not.
- Verdict and risk. Verdict: Handled (the document deals with it well), Partly handled (deals with it, but with cost, delay or doubt), Gap (the document is silent or the default law leaves the client exposed), Conflict (two clauses point different ways, or the outcome turns on a contested reading). Risk to the client: High, Medium or Low.
Step 4. Hunt for interaction failures
The highest-value findings come from two clauses that are each fine alone but fail together. Check every one of these against the model:
- The liability cap applies to the indemnity, so the indemnity is worth no more than the cap. Or the reverse: the indemnity is carved out of the cap, so the cap means nothing.
- Termination (either kind) is silent on payment for work done or goods delivered before termination.
- Termination for convenience sits next to a minimum commitment, exclusivity or upfront investment, so one side can walk away after the other has sunk costs.
- Invoicing depends on acceptance, and acceptance has no deadline and no deemed acceptance, so payment can be delayed forever.
- IP vests "on payment", and the contract ends before payment, so nobody clearly owns the work.
- Survival clause omits confidentiality, IP, indemnities, limitation of liability or payment obligations.
- Cure periods are longer than the time the client can survive the breach (for example, 30 days to fix a payroll outage).
- Notice clause requires post or a specific address, while the parties actually communicate by email or chat. Valid notice may be impossible in a crisis.
- Assignment is restricted but change of control is not (or the reverse), so a share sale or asset sale has an unexpected result.
- Exclusivity, non-compete or most-favoured-customer clauses with no minimum volume or no end date.
- Automatic renewal combined with a narrow cancellation window, plus consumer or small-business protections that may make the renewal or the window unenforceable.
- Liquidated damages or fees on termination that may be unenforceable as a penalty (test under the governing law).
- Clauses that let a party terminate or change terms because the other is insolvent or restructuring, where statute may stay or void that right.
- Set-off rights (or no-set-off clauses) that change who holds the money during a dispute.
- Dispute clause forces arbitration, a foreign forum or expensive litigation for a claim worth less than the cost of running it.
- Personal guarantees, security interests or registration requirements that were never completed, so the protection does not exist when needed.
- "Best endeavours", "reasonable endeavours", "commercially reasonable efforts" and "material breach" used without definition where the scenario turns on them.
Step 5. Fix
For every scenario that is not Handled, give the fix. Two kinds:
- Drafting fix: the exact words to add, delete or replace, in the style of the document, with the clause number. Hand off to
redlineif the user wants tracked changes. - Commercial or operational fix: insurance, a deposit, a personal guarantee, a PPSR or UCC registration, a calendar reminder for a renewal window, a process for giving notice. Many scenarios are better solved outside the document.
Rank fixes by value to the client divided by negotiation cost. Flag which ones the counterparty will resist and why.
Step 6. Report
Lead with the answer. Then:
## Scenario test: <document> (acting for <client>)
Governing law: <jurisdiction> <n> scenarios: <n> handled, <n> partly handled, <n> gaps, <n> conflicts
### The three that matter
1. <Scenario name>: <one-paragraph story of what happens, with clause refs and exposure>. Fix: <one line>.
2. ...
3. ...
### Scenario matrix
| # | Scenario | What happens now (clauses) | Silence or ambiguity | Default law | Client end state | Verdict | Risk | Fix |
|---|---|---|---|---|---|---|---|---|
### Interaction failures found
- ...
### Fixes, ranked
| Priority | Fix | Type (drafting / commercial) | Counterparty likely to resist? |
|---|---|---|---|
### Assumptions and what would change the result
- ...Finish with: "Say fix it to apply all of these as tracked changes, or fix 1 and 3 to apply some." Do not change the document during the scenario test itself. When the user says fix it, hand the chosen drafting fixes to redline, apply them to the latest version, update cross-references, and say which scenarios are now handled.
Keep the war stories vivid and concrete. A client reads "if your biggest customer goes into administration owing you three months of fees, clause 14.2 lets them keep the software and you become an unsecured creditor for $45,000" and acts on it. They skim "consider strengthening termination provisions".
Mode: advice sensitivity test
When the output is advice rather than a document, list the facts the conclusion depends on, then flip each one: "If the worker set their own hours, the answer changes from employee to likely contractor." Return a short table of fact, current assumption, answer if flipped, and how to confirm. Those flips become the questions to ask the client before relying on the advice.
Mode: negotiation war-game
Before sending redlines, play the counterparty. For each requested change, predict their response (accept, resist, counter) from their incentives, their standard paper and market practice, and draft the likely counter and the client's fallback. Output a give/get plan: what to trade, what to hold, what to drop. Mark predictions as predictions.
Guardrails and limits
- Do not invent facts to make a scenario dramatic. If the scenario needs a fact you do not have, state it as an assumption.
- Do not invent law to fill silence. If you are not sure what the default rule is in this jurisdiction, say "unclear, check" and keep going.
- Exposure figures are estimates. Show the arithmetic and the assumptions.
- A scenario test is not a prediction that the scenario will happen. Say so once, briefly, if the client seems alarmed.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

