debate-speech-writer
The AI prompt library behind the FIAT speech and flow features, plus the batch quote-analysis helper that scores parsed cards.
import {
speechToFlowPrompt, // speech document -> a new flow column
speechToResponsePrompt, // opponent speech -> response options
judgeDecisionPrompt, // flow -> aff-wins and neg-wins decision outlines
findFlawsPrompt, // quote -> warrants, gaps and overstatements
textToHighlightedPrompt, // card text -> highlight/underline spans
topicToResearchOutlinePrompt, // topic -> research outline of keyphrases
} from "debate-speech-writer"analyzeQuotes() walks an outline of parsed cards, sends each card's HTML through
findFlawsPrompt, and writes the summaries, warrants, scores and flaws back onto the
outline entries.
Also exports JudgeProfilesPanel, JudgeParadigmPickerPanel, OpponentPersonaPickerPanel,
and CoachMaterialsPanel.
Package layout
Logic lives under src/, grouped by role; tests live under test/.
debate-speech-writer/
├── src/
│ ├── analysis/ # batch LLM analysis over parsed cards
│ ├── coach/ # team coach-material library, grounded prompt, real AI Q&A call
│ ├── judge/ # judge-paradigm registry, judge-profile aggregation
│ ├── opponent/ # AI practice-opponent persona registry
│ ├── panels/ # JudgeProfilesPanel, JudgeParadigmPickerPanel, OpponentPersonaPickerPanel, CoachMaterialsPanel
│ ├── prompts/ # the prompt library
│ ├── state/ # localStorage-backed persistence stores
│ └── index.ts # public entry point
└── test/ # Vitest suites asserting the prompt contracts and state helpersTests
bun run test # or: npx vitest run
bun run coverage # writes ./coverage for this package alone
