Inkwell -- Professional Contracts in Seconds - Vertical Agent Spec

One-line definition

An agent that drafts, customizes, and delivers service agreements for freelancers and small agencies based on structured intake of project scope, deliverables, and payment terms.

The workflow it owns end-to-end

Intake: Receives a brief description of the engagement (client name, project type, deliverables, timeline, rate) via form, Slack message, or plain-text prompt.
Clarification loop: Asks 3-5 targeted follow-up questions when critical fields are missing: kill fee percentage, revision limits, late payment penalty, IP ownership clause preference.
Draft generation: Produces a jurisdiction-aware contract draft using clause libraries tuned to the user's state or country and service category (design, development, copywriting, consulting).
Review pass: Flags any clauses that deviate from the user's previously saved defaults and presents a diff summary before sending.
Delivery: Exports to PDF or pushes to DocuSign or HelloSign for client signature, then archives the signed copy in a connected folder.

What it knows that a generic LLM doesn't

Clause libraries by trade: A web developer's contract needs different IP assignment language than a photographer's; a marketing consultant needs different confidentiality scope than a copywriter. Generic prompts collapse these distinctions.
Kill fee norms by industry: Standard kill fees range from 25-50% for creative work depending on project stage. The agent applies sensible defaults and flags when a user's stated rate falls outside that range.
Jurisdiction-specific provisions: Late payment interest caps, mandatory cancellation notice periods, and lien rights vary by state and country. This requires structured rule tables, not general knowledge.
Scope creep clause patterns: Freelancers systematically underestimate scope creep costs. The agent defaults to a defined revision count and an hourly rate for overages without being prompted.
The user's own contract history: After several contracts, it knows this user's standard payment terms, preferred governing law, and which clients have previously triggered dispute clauses.
Red-flag intake signals: If the intake describes vague deliverables, no stated deadline, or a client demanding unlimited revisions, the agent surfaces that as a structural risk before generating anything.

What it explicitly declines

Legal advice: It will not tell a user whether their contract is enforceable, whether they have a winning dispute, or how a clause will be interpreted by a court. It generates documents; it does not practice law.
Disputes and collections: Once a contract is signed and a problem arises, the agent stops. Debt collection, demand letters, and litigation strategy are outside its scope.
Regulated industries: Medical, financial, real estate, and employment contracts carry compliance requirements that clause libraries cannot reliably satisfy. The agent declines those categories entirely rather than producing a confident-sounding draft that will fail.
Acting as counsel to the counterparty: It works for the freelancer generating the contract, not the client reviewing it.

Tools and integrations required

DocuSign or HelloSign: For signature collection and a verifiable audit trail on delivered contracts.
Google Drive or Dropbox: For archiving signed contracts and retrieving prior contracts as context for repeat clients.
Stripe or QuickBooks: To pull project value and payment schedule data directly, and to verify that rate formatting matches what was actually invoiced.
Calendly or Google Calendar: To auto-populate project start dates and milestone dates from scheduled kickoff calls rather than having the user re-enter them.
A versioned jurisdiction clause API: Something like Gavel or a custom-built clause library tagged by state and contract type. This is the highest-cost build component and the part most likely to go stale as statutes change.
Slack or email: For delivering draft links and signature-complete notifications without requiring the user to log into a separate dashboard.

Trust escalation: when it pings a human

Novel contract type: If the user's intake does not match a known category, the agent pauses and asks for classification rather than hallucinating a clause set that sounds plausible but is stitched together incorrectly.
Large contract value: For any engagement over a user-configured threshold (default $25,000), the agent generates the draft but requires manual confirmation before delivery. The risk of a bad clause scales with contract size.
Client with a prior dispute flag: If stored history shows this client name was previously involved in a dispute, the agent surfaces that before generating a new contract for them.
Low-confidence jurisdiction: Contracts for jurisdictions outside the US, UK, Canada, and Australia receive a clear warning that clause library coverage is limited, with a recommendation to have an attorney review before sending.

Pricing model

Per-contract pricing at $4 per generated and delivered contract is the honest model here. At that rate, a freelancer who signs 10 clients per month pays $40, which is defensible if the tool saves 30-45 minutes of template hunting and editing per contract. The problem is that the median freelancer signs 2-4 new clients per month, putting their monthly cost at $8-16. That is so low it makes the business unviable as a standalone product or forces the operator to bundle adjacent features (invoicing, client CRM, proposal generation) to justify a $15-25 monthly floor. The per-contract model is honest about value delivered. The subscription model requires shipping a broader product. This spec, as written for contract generation alone, does not produce a viable standalone business at the $4 per-contract rate unless volume is extremely high or the customer segment shifts toward agencies processing 50 or more contracts per month, which is a different product with a different sales motion.

Differentiation from a generic LLM wrapper

A user who pastes "write me a freelance contract for a $5,000 web design project" into Claude gets a generic, jurisdiction-agnostic, clause-poor document that will look plausible and fail under scrutiny. The agent described here differs in three specific ways: it maintains the user's contract history and defaults so that each new contract builds on what the user actually does rather than inferring from scratch; it applies a structured, versioned clause library that can be updated when a state changes its late payment statute, rather than relying on training data that is already months stale; and it handles the delivery workflow, meaning signature collection, archival, and client notification, so the output is a completed process rather than a text block the user still has to format, export, and track manually. Whether those three differences justify $4 per contract depends entirely on contract frequency and how much the user cares about clause defensibility versus document appearance. For a high-volume agency, the case is real. For a solo freelancer signing three clients per year, it probably is not, and no amount of positioning will change that math.