otaip: Instructions file for Claude Code

OTAIP — Claude Code Constitution

Core Rule

DO NOT INVENT DOMAIN LOGIC. All travel domain knowledge comes from files in docs/knowledge-base/. If something is ambiguous or missing, write a // TODO: DOMAIN_QUESTION: {question} comment and move on. Do not guess. Do not fill gaps from "general knowledge." Hotel and airline domains have edge cases that seem obvious but aren't.

When Blocked

If you lack domain input to proceed: refactor, clean, document existing code, then surface the blocking question. Never invent travel industry behavior.

Tech Stack

• TypeScript (strict: true, plus noUncheckedIndexedAccess, noImplicitOverride, noPropertyAccessFromIndexSignature; exactOptionalPropertyTypes is off — enabling it would require widespread call-site changes, tracked as a separate cleanup)
• Node.js >=24
• pnpm 10+ (workspace monorepo)
• Vitest for testing
• tsup for building
• ESLint + Prettier for linting/formatting
• ESM modules (type: "module")

Agent Contract

Every agent implements:

interface Agent<TInput, TOutput> {
  readonly id: string;
  readonly name: string;
  readonly version: string;
  initialize(): Promise<void>;
  execute(input: AgentInput<TInput>): Promise<AgentOutput<TOutput>>;
  health(): Promise<AgentHealthStatus>;
}

Imported from @otaip/core.

Error Handling

Use existing error classes from @otaip/core:

• AgentNotInitializedError — execute() before initialize()
• AgentInputValidationError — bad input (with field + reason)
• AgentDataUnavailableError — external source down
• AgentError — base class for custom errors

Agent File Structure

Follow the pattern in packages/agents/reference/src/airport-code-resolver/:

{agent-name}/
  types.ts          — all input/output/internal types
  {logic}.ts        — core business logic (pure functions)
  index.ts          — Agent interface implementation
  __tests__/
    {agent}.test.ts  — vitest tests

Rules

• No any types without explicit justification comment
• No secrets or API keys in code
• No hardcoded credentials
• Every agent needs tests
• Agents are stateless
• Use existing error classes, don't create new ones unless necessary
• Mock external APIs in tests — never call real APIs from tests