End-to-End Guide

The full journey: from a document you trust to a skill that runs in your todo.law apps and in LegalQuants-community runtimes. Fifteen minutes the first time, less after that.

Authoring: document to published skill

1. Upload a document

Start from your best precedent, not a blank template. Contracts (NDA, MSA, term sheets, DPAs) become contract skills. Compliance documents (DPIA questionnaires, vendor audits, checklists) become assessment skills. PDF, DOCX, and TXT up to 10MB.

2. Review the analysis

The pipeline classifies the document, extracts every clause with a plain-language summary and bias assessment, and flags issues by severity. This is also where the document category is decided: contract or assessment. Check it before generating a skill, because it determines the skill type and destination.

3. Generate the skill draft

One click starts generation. Contract drafts get a clause library where each clause carries 3 to 5 negotiation options with pros, cons, and full legal text, plus contract boilerplate. Assessment drafts get scored criteria organized in categories, with per-criterion guidance and remediation. Both finish with evaluation cases grounded in the skill's own content.

4. Edit in review

Every draft stops in a review state before anything is published. Edit clauses, boilerplate, criteria, guidance, and metadata until the skill drafts the way you do. The generated evaluation cases are AI-drafted: have a lawyer confirm any legal claims in them before anyone relies on the skill.

5. Publish

Publishing exports the skill in both formats at once and runs a conformance check first, so a malformed skill fails at export with a clear message instead of failing at install time. The result is a folder with the engine data, the agent-facing SKILL.md, a README, parameters, evaluation cases, and the license you chose.

What a published skill contains

One publish produces a single folder that serves two worlds. Hosts read the parts they understand and ignore the rest.

Engine data (todo.law suite)

clauses.json and boilerplate.json for contract skills
template.json, assessment.json, and guidance.json for assessment skills
manifest.json and metadata.json

Agent skill (any skills-capable runtime)

SKILL.md with YAML frontmatter and declared inputs
README.md and LICENSE
parameters.json and evals/evals.json

The evaluation file carries three kinds of cases: a positive case naming the clauses or criteria the output must contain, a currency case probing the legal position most likely to go stale, and a gating case the skill must refuse or warn on. They are written from the skill's own content so they can be checked mechanically, and they are AI-drafted, so review them before relying on them.

Running your skill in todo.law apps

Dealroom

Contract skills install as signed packages. Your clause options become the negotiation surface: each party picks positions, the engine assembles the contract from your legal text.

DPO Central

Assessment skills install as scored templates. Your criteria, risk levels, and guidance drive the assessment wizard and its scoring.

AI Sentinel

AI governance assessments are supported at authoring time. The installer on the AI Sentinel side is in development, so these skills cannot be installed yet.

Installation is verified: packages are signed, integrity-checked, and validated against the installer's schema before anything is activated.

Running your skill in LegalQuants runtimes and agent hosts

The same published folder is a standard agent skill. Any LegalQuants-community runtime loads it as data: the runtime reads SKILL.md, surfaces the declared inputs as a form, and assembles the skill text into the model prompt. No conversion step, no separate export.

LQ.AI: place the skill folder in the runtime's skills directory, or contribute it to the community skills repository. The community gate requires the README, license, and evaluation cases your skill already ships with.
Claude and other agent-skill hosts: upload the folder (or a zip of it) wherever agent skills are accepted. Hosts that do not understand the engine JSON files simply ignore them.
Legal skill hubs: community hubs that accept agent skills can list your skill as-is. Check each hub's naming and review requirements before submitting.

When to use Clausemaster, and when to use a conversational builder

Conversational builders such as Anthropic's skill-creator write a skill from what you articulate in a chat. That fits knowledge that lives in your head: escalation rules, review discipline, negotiation judgment. Clausemaster starts from the other end: the knowledge already in your documents. A clause library refined over fifty deals cannot be dictated in a chat, but it can be uploaded.

The two combine well. Generate the skill from your document here, then refine the SKILL.md prose conversationally in any agent that supports skills. The structured engine data stays intact either way.

Go deeper

AI Analysis

How the three-step pipeline reads your document.

For Lawyers

Publisher onboarding, drafting, and the review queue.

For Startups

Analysis, reports, and lawyer reviews without authoring.