Prompt catalogue

Every piece of prompt text lives in api/src/prompts/catalogue.ts with a code default, and nowhere else. The Prompt Lab can override any of these per browser; the override travels with each request (see Prompt Lab). This page lists ids and purposes. The wording itself is in the catalogue, where it is reviewed like code.

Generated from the code

This page is generated by docusaurus/scripts/generate-reference.ts from the source it describes. Edit the source, then run npm run docs:reference.

System prompts

Assembled in a fixed order on every turn (see The chat pipeline).

Id	Label	Purpose	Words
system.base	System — identity & working style	Who the assistant is and how it works. Prepended to every turn, text and voice. {{assistant}} and {{product}} are filled in with the names the caller runs under.	299
system.screen_actions	System — pointing at the screen	Appended only when the host can carry out screen actions. Without it the assistant is never told it can point at anything, because it cannot.	48
system.safety	System — safety & boundaries	The guardrails: scope, prompt-injection resistance, no persona changes, no configuration leaks. Always applied, before any claim data; overrides everything else.	496
system.locale_uk	System — UK style	Appended when the handler is on the UK product: British English, UK dates and terminology.	44
system.locale_us	System — US style	Appended when the handler is on the US product: US English, US dates and terminology.	60
system.voice_greeting	System — voice greeting	The opening line when a voice call connects. Sent as the first turn, so the handler is greeted rather than met with silence.	124
system.voice	System — voice mode	Appended when the handler is talking to the claim by voice.	180

Capability playbooks

Id	Capability	Purpose	Words
capability.claim_qa	claim_qa	Guidance for free-form questions (no quick action selected).	93
capability.claim_summary	claim_summary	The playbook for a handler-ready claim summary.	145
capability.document_summary	document_summary	The playbook for summarising the documents on the file.	153
capability.timeline	timeline	The playbook for a dated chronology.	107
capability.data_insights	data_insights	The playbook for what the claim’s numbers say.	159
capability.policy_validation	policy_validation	The playbook for checking the loss against the policy.	265
capability.reserve_estimation	reserve_estimation	The playbook for estimating the likely cost and judging reserve adequacy.	189
capability.next_actions	next_actions	The playbook for recommending what the handler should do next.	125
capability.settlement_guidance	settlement_guidance	The playbook for a settlement range and negotiation strategy.	199
capability.litigation_risk	litigation_risk	The playbook for assessing the likelihood and cost of litigation.	167
capability.recovery_opportunities	recovery_opportunities	The playbook for finding money the claim could recover.	132
capability.fraud_check	fraud_check	The playbook for screening the claim for fraud and leakage.	238
capability.vulnerable_customer	vulnerable_customer	The playbook for spotting customer vulnerability and adapting handling (FCA Consumer Duty).	209
capability.complaints_guidance	complaints_guidance	The playbook for identifying complaints and escalating correctly.	190
capability.qa_review	qa_review	The playbook for auditing how the claim has been handled.	153
capability.document_production	document_production	The playbook for drafting letters, emails, file notes, reports and strategy documents.	207
capability.call_assist	call_assist	The playbook for live support before and during a phone call.	208
capability.field_visit_briefing	field_visit_briefing	The playbook for briefing an adjuster or investigator before a site or claimant visit.	172
capability.handover_brief	handover_brief	The playbook for briefing a handler taking over the claim.	150
capability.capabilities_briefing	capabilities_briefing	The playbook for explaining the assistant’s capabilities.	97