claude-flow
Version:
Ruflo - Enterprise AI agent orchestration for Claude Code. Deploy 60+ specialized agents in coordinated swarms with self-learning, fault-tolerant consensus, vector memory, and MCP integration
45 lines • 1.48 kB
TypeScript
/**
* Smoke tests for extractFinalAnswer + buildUserMessage (iter 53a T2 narrowed).
*
* Iter 53a changes vs iter 52 T2:
* - Stage 2/3 (prose fallback) REMOVED — extractFinalAnswer is Stage 1 only.
* - Commitment prompt no longer has "FINAL_ANSWER: unknown" surrender instruction.
* - Reversed-text preprocessor PRESERVED (not the source of regressions).
*
* Anti-regression suite (7 cases):
* These are the exact failure modes from iter 52 T2 that must not regress.
* They test that the extraction logic is stable and correct for the 7 questions
* where iter 52 broke but iter 51 was correct.
*
* Run (after build):
* node dist/benchmarks/gaia-extract.smoke.js
*
* Exit 0 on all pass, 1 on any failure.
*
* Refs: ADR-133, ADR-135, iter 52 T2, iter 53a, #2156
*/
interface AnthropicResponse {
id: string;
model: string;
stop_reason: string;
content: Array<{
type: string;
text?: string;
}>;
usage: {
input_tokens: number;
output_tokens: number;
};
}
declare let _extractFinalAnswer: (resp: AnthropicResponse) => string | null;
declare let _buildUserMessage: (question: string) => string;
declare function makeResp(text: string): AnthropicResponse;
interface TestCase {
label: string;
input: string;
expected: string | null;
isUserMsg?: boolean;
}
declare const CASES: TestCase[];
declare function main(): Promise<void>;
//# sourceMappingURL=gaia-extract.smoke.d.ts.map