Nylon

Read WITHIN a single document from the Nylon legal library, or read passages around a known extract. Use this after search_library has identified a relevant document (by documentId) or extract (by chunkId). Three modes, chosen by the field combination: - Whole document: pass only documentId. Returns the whole document, capped at ~50,000 tokens. If it is longer, only the first portion is returned — use search-within for the rest. - Search within: documentId + a "phrase". Returns the most relevant extracts inside that document. Pick searchType="keyword" for exact strings/citations or "semantic" (default) when you are unsure of the wording. - Adjacent: documentId + an "extractId" (chunkId). Returns the extracts immediately above and/or below that chunk — useful for reading the context around a passage or continuing past a cut-off. Choose direction via position (defaults to both). Returns extracts with their text, identifiers (chunkId / documentId), and a clickable resolver link. When citing a source, use ONLY the `link` field from the result. Never construct, guess, or recall a source URL yourself — in particular do not link to austlii.edu.au, nzlii.org, or other external case-law databases. If a result has no link, cite it by title without a URL.

read_document

Search the Nylon legal library (legislation, cases, and commentary across NZ, AU and the UK). WHEN TO USE: Call this tool whenever a question could possibly require tax or legal knowledge — including statutes, regulations, case law, legal procedure, compliance, contracts, tax rules/rates/filing, or any "is this allowed / what does the law say" question. Do NOT answer from memory for these; your training data may be outdated, jurisdiction-specific, or wrong. If there is any doubt about whether a question touches law or tax, search first. It is better to search unnecessarily than to give an unsourced legal or tax answer. Provide one or more `searches` (max 5) — they run in parallel and each returns one batch of fused results (starting from the first result by default). Pass `offset` to skip results and `limit` to change the batch size; when a search block reports "More results available", call again with the offset it suggests. Paging is stateless (a skip count only, no cursors). Filters are FACET SLUGS (jurisdictions: nz | au | gb; practiceAreas: slugs like "tax"; contentTypes: cases | legislation | commentary). Values within an array are OR-combined; fields are AND-combined. Unknown slugs are rejected with a hint, never silently ignored. JURISDICTION: If the user's question names or clearly implies a jurisdiction (New Zealand → nz, Australia → au, United Kingdom → gb), you MUST set the `jurisdictions` filter to match. Leaving it blank searches ALL jurisdictions and ranks purely by how much content each has, so a New Zealand question can come back full of Australian material. Only leave it blank when the user genuinely wants all jurisdictions, or none is implied. Returns ranked extracts with a clickable resolver link and identifiers (documentId / chunkId) that read_document can use to fetch more. Each result also carries its documentType facet slug, reusable in future filters. When citing a source, use ONLY the `link` field from the result. Never construct, guess, or recall a source URL yourself — in particular do not link to austlii.edu.au, nzlii.org, or other external case-law databases. If a result has no link, cite it by title without a URL.

search_library