Tools Reference

Documentation tools

These tools search and read the CL-SDK documentation. They require no model configuration.

list_doc_sections

List all documentation sections and their pages.

Input: none

Output: JSON array of sections with title, slug, and pages.

[
  { "title": "Getting Started", "slug": "getting-started", "pages": ["quickstart", "architecture", "models"] },
  { "title": "Document Extraction", "slug": "extraction", "pages": ["pipeline", "classification", "applying-results"] }
]

search_docs

Full-text search across all MDX documentation pages. Returns top 5 matches ranked by relevance.

Input:

Parameter	Type	Required	Description
query	string	yes	Search query
section	string	no	Limit to a section slug (e.g. extraction)

Output: Ranked results with title, slug, score, and a text excerpt around the match.

read_doc_page

Read a specific documentation page by its slug.

Input:

Parameter	Type	Required	Description
slug	string	yes	Page slug (e.g. getting-started/quickstart)

Output: The page title and full MDX content.

---

SDK prompt builders

Pure functions that return prompt strings. No model or API key needed.

build_agent_system_prompt

Generate an insurance-aware agent system prompt. Wraps buildAgentSystemPrompt(ctx).

Input:

Parameter	Type	Required	Description
platform	"email" | "chat" | "sms" | "slack" | "discord"	yes	Communication platform
intent	"direct" | "mediated" | "observed"	yes	Communication intent
siteUrl	string	yes	Company website URL
companyName	string	no	Company name
companyContext	string	no	Company context details
userName	string	no	Recipient name
coiHandling	"broker" | "user" | "member" | "ignore"	no	COI routing strategy
brokerName	string	no	Broker company name
brokerContactName	string	no	Broker contact name
brokerContactEmail	string	no	Broker contact email

build_field_extraction_prompt

Get the application field extraction prompt. Wraps buildFieldExtractionPrompt().

Input: none

build_auto_fill_prompt

Generate an auto-fill prompt for application fields. Wraps buildAutoFillPrompt(fields, orgContext).

Input:

Parameter	Type	Required	Description
fields	{ id, label, fieldType, section }[]	yes	Application fields
orgContext	{ key, value, category }[]	yes	Organization context

build_question_batch_prompt

Generate a batched question prompt for unfilled fields. Wraps buildQuestionBatchPrompt(unfilledFields).

Input:

Parameter	Type	Required	Description
unfilledFields	array	yes	Fields with id, fieldType, section, required, and optional label, text, condition

---

Data transformation tools

Map raw extraction output to structured fields. No model needed.

apply_extracted

Map raw policy extraction JSON to structured fields. Wraps applyExtracted(extracted).

Input:

Parameter	Type	Required	Description
extracted	any	yes	Raw JSON from extractFromPdf

Output: Structured object with carrier, policyNumber, coverages, premium, effectiveDate, etc.

apply_extracted_quote

Map raw quote extraction JSON to structured fields. Wraps applyExtractedQuote(extracted).

Input:

Parameter	Type	Required	Description
extracted	any	yes	Raw JSON from extractQuoteFromPdf

Output: Structured object with carrier, quoteNumber, premiumBreakdown, subjectivities, etc.

---

Extraction tools

These call the LLM-powered extraction pipeline. Requires a configured model and API key.

classify_document

Classify a PDF as a policy or quote. Wraps classifyDocumentType(pdfBase64, options).

Input:

Parameter	Type	Required	Description
pdfBase64	string	yes	Base64-encoded PDF

Output: { documentType: "policy" | "quote", confidence: number, signals: string[] }

extract_policy

Full multi-pass policy extraction. Wraps extractFromPdf(pdfBase64, options).

Input:

Parameter	Type	Required	Description
pdfBase64	string	yes	Base64-encoded PDF

Output: { rawText: string, extracted: any } — pass extracted to apply_extracted for structured fields.

extract_quote

Full multi-pass quote extraction. Wraps extractQuoteFromPdf(pdfBase64, options).

Input:

Parameter	Type	Required	Description
pdfBase64	string	yes	Base64-encoded PDF

Output: { rawText: string, extracted: any } — pass extracted to apply_extracted_quote for structured fields.

---

PDF tools

Direct PDF operations. No model needed.

get_acro_form_fields

List all fillable AcroForm fields in a PDF. Wraps getAcroFormFields(pdfDoc).

Input:

Parameter	Type	Required	Description
pdfBase64	string	yes	Base64-encoded PDF

Output: Array of { name, type, options? } where type is "text", "checkbox", "dropdown", or "radio".

fill_acro_form

Fill AcroForm fields and return the flattened PDF. Wraps fillAcroForm(pdfBytes, mappings).

Input:

Parameter	Type	Required	Description
pdfBase64	string	yes	Base64-encoded PDF
mappings	{ acroFormName, value }[]	yes	Field name to value mappings

Output: Base64-encoded PDF with fields filled and flattened.

overlay_text_on_pdf

Overlay text at specified coordinates on a flat PDF. Wraps overlayTextOnPdf(pdfBytes, overlays).

Input:

Parameter	Type	Required	Description
pdfBase64	string	yes	Base64-encoded PDF
overlays	array	yes	Each with page (0-indexed), x, y, text, optional fontSize and isCheckmark

Output: Base64-encoded PDF with text overlaid.