Document analysis

A structured first pass over any document — the playbook's composability example names this skill explicitly as the one a compliance-review skill calls before extraction happens.

Workflow

1. Detect the document type first. Contract, financial statement, policy, form, correspondence, filing — the type determines what structure to expect and what matters. Don't skip straight to content.
2. Map the structure before reading for content. Identify sections, headers, and any numbering/clause scheme. A document's structure often tells you where the important parts live (definitions up front, obligations in the middle, signatures/dates at the end).
3. Identify entities. Who are the parties, dates, amounts, and defined terms? List them before analyzing what they mean — this catches inconsistencies (a party named three different ways, a date that doesn't match the header).
4. Flag ambiguity and missing sections explicitly. If a document type normally has a section (e.g. a termination clause) and this one doesn't, say so — absence is often as important as presence.
5. Summarize by section, not just overall. A one-line summary per section is more useful downstream (to extraction or compliance skills) than one paragraph for the whole document.

Composition

This is typically the first skill applied to a raw document. Extraction pulls specific fields once structure is known; compliance-review and legal-contract-review both build on this skill's output rather than re-parsing the document themselves.

Quality bar

• Document type is stated explicitly, not implied.
• Every major section has at least a one-line summary.
• Entities (parties, dates, amounts, defined terms) are listed, not just mentioned in prose.
• Missing expected sections are called out, not silently skipped.