@claude-flow/cli 3.10.4 → 3.10.6
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/dist/src/benchmarks/capability-tasks.d.ts +126 -0
- package/dist/src/benchmarks/capability-tasks.d.ts.map +1 -0
- package/dist/src/benchmarks/capability-tasks.js +147 -0
- package/dist/src/benchmarks/capability-tasks.js.map +1 -0
- package/dist/src/benchmarks/gaia-agent-contrastive.d.ts +53 -0
- package/dist/src/benchmarks/gaia-agent-contrastive.d.ts.map +1 -0
- package/dist/src/benchmarks/gaia-agent-contrastive.js +177 -0
- package/dist/src/benchmarks/gaia-agent-contrastive.js.map +1 -0
- package/dist/src/benchmarks/gaia-agent-gemini.d.ts +51 -0
- package/dist/src/benchmarks/gaia-agent-gemini.d.ts.map +1 -0
- package/dist/src/benchmarks/gaia-agent-gemini.js +352 -0
- package/dist/src/benchmarks/gaia-agent-gemini.js.map +1 -0
- package/dist/src/benchmarks/gaia-agent-planning.smoke.d.ts +18 -0
- package/dist/src/benchmarks/gaia-agent-planning.smoke.d.ts.map +1 -0
- package/dist/src/benchmarks/gaia-agent-planning.smoke.js +253 -0
- package/dist/src/benchmarks/gaia-agent-planning.smoke.js.map +1 -0
- package/dist/src/benchmarks/gaia-agent.d.ts +198 -0
- package/dist/src/benchmarks/gaia-agent.d.ts.map +1 -0
- package/dist/src/benchmarks/gaia-agent.js +651 -0
- package/dist/src/benchmarks/gaia-agent.js.map +1 -0
- package/dist/src/benchmarks/gaia-causal-memory.d.ts +133 -0
- package/dist/src/benchmarks/gaia-causal-memory.d.ts.map +1 -0
- package/dist/src/benchmarks/gaia-causal-memory.js +281 -0
- package/dist/src/benchmarks/gaia-causal-memory.js.map +1 -0
- package/dist/src/benchmarks/gaia-causal-memory.smoke.d.ts +22 -0
- package/dist/src/benchmarks/gaia-causal-memory.smoke.d.ts.map +1 -0
- package/dist/src/benchmarks/gaia-causal-memory.smoke.js +300 -0
- package/dist/src/benchmarks/gaia-causal-memory.smoke.js.map +1 -0
- package/dist/src/benchmarks/gaia-claude-p.d.ts +105 -0
- package/dist/src/benchmarks/gaia-claude-p.d.ts.map +1 -0
- package/dist/src/benchmarks/gaia-claude-p.js +269 -0
- package/dist/src/benchmarks/gaia-claude-p.js.map +1 -0
- package/dist/src/benchmarks/gaia-claude-p.smoke.d.ts +20 -0
- package/dist/src/benchmarks/gaia-claude-p.smoke.d.ts.map +1 -0
- package/dist/src/benchmarks/gaia-claude-p.smoke.js +168 -0
- package/dist/src/benchmarks/gaia-claude-p.smoke.js.map +1 -0
- package/dist/src/benchmarks/gaia-codeagent-runner.py +556 -0
- package/dist/src/benchmarks/gaia-codeagent.d.ts +129 -0
- package/dist/src/benchmarks/gaia-codeagent.d.ts.map +1 -0
- package/dist/src/benchmarks/gaia-codeagent.js +590 -0
- package/dist/src/benchmarks/gaia-codeagent.js.map +1 -0
- package/dist/src/benchmarks/gaia-codeagent.smoke.d.ts +33 -0
- package/dist/src/benchmarks/gaia-codeagent.smoke.d.ts.map +1 -0
- package/dist/src/benchmarks/gaia-codeagent.smoke.js +241 -0
- package/dist/src/benchmarks/gaia-codeagent.smoke.js.map +1 -0
- package/dist/src/benchmarks/gaia-convergence.d.ts +138 -0
- package/dist/src/benchmarks/gaia-convergence.d.ts.map +1 -0
- package/dist/src/benchmarks/gaia-convergence.js +260 -0
- package/dist/src/benchmarks/gaia-convergence.js.map +1 -0
- package/dist/src/benchmarks/gaia-convergence.smoke.d.ts +19 -0
- package/dist/src/benchmarks/gaia-convergence.smoke.d.ts.map +1 -0
- package/dist/src/benchmarks/gaia-convergence.smoke.js +246 -0
- package/dist/src/benchmarks/gaia-convergence.smoke.js.map +1 -0
- package/dist/src/benchmarks/gaia-critic.d.ts +123 -0
- package/dist/src/benchmarks/gaia-critic.d.ts.map +1 -0
- package/dist/src/benchmarks/gaia-critic.js +312 -0
- package/dist/src/benchmarks/gaia-critic.js.map +1 -0
- package/dist/src/benchmarks/gaia-critic.smoke.d.ts +21 -0
- package/dist/src/benchmarks/gaia-critic.smoke.d.ts.map +1 -0
- package/dist/src/benchmarks/gaia-critic.smoke.js +327 -0
- package/dist/src/benchmarks/gaia-critic.smoke.js.map +1 -0
- package/dist/src/benchmarks/gaia-dag-pilot.d.ts +22 -0
- package/dist/src/benchmarks/gaia-dag-pilot.d.ts.map +1 -0
- package/dist/src/benchmarks/gaia-dag-pilot.js +158 -0
- package/dist/src/benchmarks/gaia-dag-pilot.js.map +1 -0
- package/dist/src/benchmarks/gaia-dag.d.ts +110 -0
- package/dist/src/benchmarks/gaia-dag.d.ts.map +1 -0
- package/dist/src/benchmarks/gaia-dag.js +658 -0
- package/dist/src/benchmarks/gaia-dag.js.map +1 -0
- package/dist/src/benchmarks/gaia-dag.smoke.d.ts +21 -0
- package/dist/src/benchmarks/gaia-dag.smoke.d.ts.map +1 -0
- package/dist/src/benchmarks/gaia-dag.smoke.js +290 -0
- package/dist/src/benchmarks/gaia-dag.smoke.js.map +1 -0
- package/dist/src/benchmarks/gaia-decomposer.d.ts +125 -0
- package/dist/src/benchmarks/gaia-decomposer.d.ts.map +1 -0
- package/dist/src/benchmarks/gaia-decomposer.js +350 -0
- package/dist/src/benchmarks/gaia-decomposer.js.map +1 -0
- package/dist/src/benchmarks/gaia-decomposer.smoke.d.ts +21 -0
- package/dist/src/benchmarks/gaia-decomposer.smoke.d.ts.map +1 -0
- package/dist/src/benchmarks/gaia-decomposer.smoke.js +228 -0
- package/dist/src/benchmarks/gaia-decomposer.smoke.js.map +1 -0
- package/dist/src/benchmarks/gaia-e2e-smoke.d.ts +27 -0
- package/dist/src/benchmarks/gaia-e2e-smoke.d.ts.map +1 -0
- package/dist/src/benchmarks/gaia-e2e-smoke.js +136 -0
- package/dist/src/benchmarks/gaia-e2e-smoke.js.map +1 -0
- package/dist/src/benchmarks/gaia-ensemble-pilot.d.ts +17 -0
- package/dist/src/benchmarks/gaia-ensemble-pilot.d.ts.map +1 -0
- package/dist/src/benchmarks/gaia-ensemble-pilot.js +106 -0
- package/dist/src/benchmarks/gaia-ensemble-pilot.js.map +1 -0
- package/dist/src/benchmarks/gaia-ensemble.d.ts +103 -0
- package/dist/src/benchmarks/gaia-ensemble.d.ts.map +1 -0
- package/dist/src/benchmarks/gaia-ensemble.js +462 -0
- package/dist/src/benchmarks/gaia-ensemble.js.map +1 -0
- package/dist/src/benchmarks/gaia-extract.smoke.d.ts +45 -0
- package/dist/src/benchmarks/gaia-extract.smoke.d.ts.map +1 -0
- package/dist/src/benchmarks/gaia-extract.smoke.js +242 -0
- package/dist/src/benchmarks/gaia-extract.smoke.js.map +1 -0
- package/dist/src/benchmarks/gaia-gemini-pilot.d.ts +17 -0
- package/dist/src/benchmarks/gaia-gemini-pilot.d.ts.map +1 -0
- package/dist/src/benchmarks/gaia-gemini-pilot.js +168 -0
- package/dist/src/benchmarks/gaia-gemini-pilot.js.map +1 -0
- package/dist/src/benchmarks/gaia-hardness/features.d.ts +46 -0
- package/dist/src/benchmarks/gaia-hardness/features.d.ts.map +1 -0
- package/dist/src/benchmarks/gaia-hardness/features.js +170 -0
- package/dist/src/benchmarks/gaia-hardness/features.js.map +1 -0
- package/dist/src/benchmarks/gaia-hardness/predictor.d.ts +105 -0
- package/dist/src/benchmarks/gaia-hardness/predictor.d.ts.map +1 -0
- package/dist/src/benchmarks/gaia-hardness/predictor.js +260 -0
- package/dist/src/benchmarks/gaia-hardness/predictor.js.map +1 -0
- package/dist/src/benchmarks/gaia-hardness/predictor.smoke.d.ts +20 -0
- package/dist/src/benchmarks/gaia-hardness/predictor.smoke.d.ts.map +1 -0
- package/dist/src/benchmarks/gaia-hardness/predictor.smoke.js +235 -0
- package/dist/src/benchmarks/gaia-hardness/predictor.smoke.js.map +1 -0
- package/dist/src/benchmarks/gaia-hardness/train-data-loader.d.ts +51 -0
- package/dist/src/benchmarks/gaia-hardness/train-data-loader.d.ts.map +1 -0
- package/dist/src/benchmarks/gaia-hardness/train-data-loader.js +179 -0
- package/dist/src/benchmarks/gaia-hardness/train-data-loader.js.map +1 -0
- package/dist/src/benchmarks/gaia-judge.d.ts +88 -0
- package/dist/src/benchmarks/gaia-judge.d.ts.map +1 -0
- package/dist/src/benchmarks/gaia-judge.js +437 -0
- package/dist/src/benchmarks/gaia-judge.js.map +1 -0
- package/dist/src/benchmarks/gaia-loader.d.ts +87 -0
- package/dist/src/benchmarks/gaia-loader.d.ts.map +1 -0
- package/dist/src/benchmarks/gaia-loader.js +326 -0
- package/dist/src/benchmarks/gaia-loader.js.map +1 -0
- package/dist/src/benchmarks/gaia-mode-router.d.ts +67 -0
- package/dist/src/benchmarks/gaia-mode-router.d.ts.map +1 -0
- package/dist/src/benchmarks/gaia-mode-router.js +126 -0
- package/dist/src/benchmarks/gaia-mode-router.js.map +1 -0
- package/dist/src/benchmarks/gaia-mode-router.smoke.d.ts +15 -0
- package/dist/src/benchmarks/gaia-mode-router.smoke.d.ts.map +1 -0
- package/dist/src/benchmarks/gaia-mode-router.smoke.js +131 -0
- package/dist/src/benchmarks/gaia-mode-router.smoke.js.map +1 -0
- package/dist/src/benchmarks/gaia-tools/file_read.d.ts +35 -0
- package/dist/src/benchmarks/gaia-tools/file_read.d.ts.map +1 -0
- package/dist/src/benchmarks/gaia-tools/file_read.js +403 -0
- package/dist/src/benchmarks/gaia-tools/file_read.js.map +1 -0
- package/dist/src/benchmarks/gaia-tools/grounded_query.d.ts +126 -0
- package/dist/src/benchmarks/gaia-tools/grounded_query.d.ts.map +1 -0
- package/dist/src/benchmarks/gaia-tools/grounded_query.js +225 -0
- package/dist/src/benchmarks/gaia-tools/grounded_query.js.map +1 -0
- package/dist/src/benchmarks/gaia-tools/grounded_query.smoke.d.ts +24 -0
- package/dist/src/benchmarks/gaia-tools/grounded_query.smoke.d.ts.map +1 -0
- package/dist/src/benchmarks/gaia-tools/grounded_query.smoke.js +276 -0
- package/dist/src/benchmarks/gaia-tools/grounded_query.smoke.js.map +1 -0
- package/dist/src/benchmarks/gaia-tools/image_describe.d.ts +45 -0
- package/dist/src/benchmarks/gaia-tools/image_describe.d.ts.map +1 -0
- package/dist/src/benchmarks/gaia-tools/image_describe.js +222 -0
- package/dist/src/benchmarks/gaia-tools/image_describe.js.map +1 -0
- package/dist/src/benchmarks/gaia-tools/image_describe.smoke.d.ts +24 -0
- package/dist/src/benchmarks/gaia-tools/image_describe.smoke.d.ts.map +1 -0
- package/dist/src/benchmarks/gaia-tools/image_describe.smoke.js +169 -0
- package/dist/src/benchmarks/gaia-tools/image_describe.smoke.js.map +1 -0
- package/dist/src/benchmarks/gaia-tools/index.d.ts +32 -0
- package/dist/src/benchmarks/gaia-tools/index.d.ts.map +1 -0
- package/dist/src/benchmarks/gaia-tools/index.js +36 -0
- package/dist/src/benchmarks/gaia-tools/index.js.map +1 -0
- package/dist/src/benchmarks/gaia-tools/pdf_read.d.ts +27 -0
- package/dist/src/benchmarks/gaia-tools/pdf_read.d.ts.map +1 -0
- package/dist/src/benchmarks/gaia-tools/pdf_read.js +226 -0
- package/dist/src/benchmarks/gaia-tools/pdf_read.js.map +1 -0
- package/dist/src/benchmarks/gaia-tools/python_exec.d.ts +31 -0
- package/dist/src/benchmarks/gaia-tools/python_exec.d.ts.map +1 -0
- package/dist/src/benchmarks/gaia-tools/python_exec.js +94 -0
- package/dist/src/benchmarks/gaia-tools/python_exec.js.map +1 -0
- package/dist/src/benchmarks/gaia-tools/python_exec.smoke.d.ts +13 -0
- package/dist/src/benchmarks/gaia-tools/python_exec.smoke.d.ts.map +1 -0
- package/dist/src/benchmarks/gaia-tools/python_exec.smoke.js +98 -0
- package/dist/src/benchmarks/gaia-tools/python_exec.smoke.js.map +1 -0
- package/dist/src/benchmarks/gaia-tools/types.d.ts +62 -0
- package/dist/src/benchmarks/gaia-tools/types.d.ts.map +1 -0
- package/dist/src/benchmarks/gaia-tools/types.js +12 -0
- package/dist/src/benchmarks/gaia-tools/types.js.map +1 -0
- package/dist/src/benchmarks/gaia-tools/visit_webpage.d.ts +52 -0
- package/dist/src/benchmarks/gaia-tools/visit_webpage.d.ts.map +1 -0
- package/dist/src/benchmarks/gaia-tools/visit_webpage.js +304 -0
- package/dist/src/benchmarks/gaia-tools/visit_webpage.js.map +1 -0
- package/dist/src/benchmarks/gaia-tools/web_browse.d.ts +61 -0
- package/dist/src/benchmarks/gaia-tools/web_browse.d.ts.map +1 -0
- package/dist/src/benchmarks/gaia-tools/web_browse.js +217 -0
- package/dist/src/benchmarks/gaia-tools/web_browse.js.map +1 -0
- package/dist/src/benchmarks/gaia-tools/web_browse.smoke.d.ts +26 -0
- package/dist/src/benchmarks/gaia-tools/web_browse.smoke.d.ts.map +1 -0
- package/dist/src/benchmarks/gaia-tools/web_browse.smoke.js +152 -0
- package/dist/src/benchmarks/gaia-tools/web_browse.smoke.js.map +1 -0
- package/dist/src/benchmarks/gaia-tools/web_search.d.ts +30 -0
- package/dist/src/benchmarks/gaia-tools/web_search.d.ts.map +1 -0
- package/dist/src/benchmarks/gaia-tools/web_search.js +210 -0
- package/dist/src/benchmarks/gaia-tools/web_search.js.map +1 -0
- package/dist/src/benchmarks/gaia-tools/web_search.smoke.d.ts +24 -0
- package/dist/src/benchmarks/gaia-tools/web_search.smoke.d.ts.map +1 -0
- package/dist/src/benchmarks/gaia-tools/web_search.smoke.js +311 -0
- package/dist/src/benchmarks/gaia-tools/web_search.smoke.js.map +1 -0
- package/dist/src/benchmarks/gaia-visit-webpage.smoke.d.ts +25 -0
- package/dist/src/benchmarks/gaia-visit-webpage.smoke.d.ts.map +1 -0
- package/dist/src/benchmarks/gaia-visit-webpage.smoke.js +304 -0
- package/dist/src/benchmarks/gaia-visit-webpage.smoke.js.map +1 -0
- package/dist/src/benchmarks/gaia-voting.d.ts +88 -0
- package/dist/src/benchmarks/gaia-voting.d.ts.map +1 -0
- package/dist/src/benchmarks/gaia-voting.js +297 -0
- package/dist/src/benchmarks/gaia-voting.js.map +1 -0
- package/dist/src/benchmarks/gaia-voting.smoke.d.ts +20 -0
- package/dist/src/benchmarks/gaia-voting.smoke.d.ts.map +1 -0
- package/dist/src/benchmarks/gaia-voting.smoke.js +332 -0
- package/dist/src/benchmarks/gaia-voting.smoke.js.map +1 -0
- package/dist/src/commands/gaia-bench.d.ts +40 -0
- package/dist/src/commands/gaia-bench.d.ts.map +1 -0
- package/dist/src/commands/gaia-bench.js +597 -0
- package/dist/src/commands/gaia-bench.js.map +1 -0
- package/dist/src/commands/index.d.ts.map +1 -1
- package/dist/src/commands/index.js +2 -0
- package/dist/src/commands/index.js.map +1 -1
- package/dist/src/commands/init.d.ts.map +1 -1
- package/dist/src/commands/init.js +41 -4
- package/dist/src/commands/init.js.map +1 -1
- package/dist/src/commands/performance-capability.d.ts +30 -0
- package/dist/src/commands/performance-capability.d.ts.map +1 -0
- package/dist/src/commands/performance-capability.js +349 -0
- package/dist/src/commands/performance-capability.js.map +1 -0
- package/dist/src/commands/route.d.ts.map +1 -1
- package/dist/src/commands/route.js +9 -0
- package/dist/src/commands/route.js.map +1 -1
- package/dist/src/init/executor.d.ts.map +1 -1
- package/dist/src/init/executor.js +25 -6
- package/dist/src/init/executor.js.map +1 -1
- package/dist/src/init/mcp-generator.d.ts.map +1 -1
- package/dist/src/init/mcp-generator.js +9 -4
- package/dist/src/init/mcp-generator.js.map +1 -1
- package/dist/src/init/statusline-generator.d.ts.map +1 -1
- package/dist/src/init/statusline-generator.js +25 -2
- package/dist/src/init/statusline-generator.js.map +1 -1
- package/dist/src/mcp-tools/agentdb-tools.d.ts.map +1 -1
- package/dist/src/mcp-tools/agentdb-tools.js +20 -3
- package/dist/src/mcp-tools/agentdb-tools.js.map +1 -1
- package/dist/src/mcp-tools/system-tools.d.ts.map +1 -1
- package/dist/src/mcp-tools/system-tools.js +12 -1
- package/dist/src/mcp-tools/system-tools.js.map +1 -1
- package/dist/src/memory/memory-bridge.d.ts.map +1 -1
- package/dist/src/memory/memory-bridge.js +34 -0
- package/dist/src/memory/memory-bridge.js.map +1 -1
- package/dist/tsconfig.tsbuildinfo +1 -1
- package/package.json +1 -1
|
@@ -0,0 +1,126 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* GAIA Tool: grounded_query — ADR-135 / iter-33
|
|
3
|
+
*
|
|
4
|
+
* Calls Gemini 2.5 Flash with the google_search grounding tool and returns
|
|
5
|
+
* a synthesised factual answer plus source citations in a single API call.
|
|
6
|
+
*
|
|
7
|
+
* Architectural advantage over web_search (Google Custom Search):
|
|
8
|
+
* - web_search: raw snippets → agent reads + parses → 2-3 extra turns
|
|
9
|
+
* - grounded_query: Gemini synthesises an answer + cites sources in 1 call
|
|
10
|
+
*
|
|
11
|
+
* This is Ruflo's cross-provider stack in action:
|
|
12
|
+
* Gemini grounds (google_search tool) → Sonnet/Haiku reasons → answer
|
|
13
|
+
*
|
|
14
|
+
* Credential resolution (mirrors performance-capability.ts / web_search.ts):
|
|
15
|
+
* 1. process.env.GOOGLE_AI_API_KEY
|
|
16
|
+
* 2. gcloud secrets versions access latest --secret=GOOGLE_AI_API_KEY --project=ruv-dev
|
|
17
|
+
* 3. Throws with clear instructions if neither is available
|
|
18
|
+
*
|
|
19
|
+
* Free tier: 1500 grounded queries/day on Gemini 2.5 Flash.
|
|
20
|
+
* Pricing: ~$0.075/M input tokens, ~$0.30/M output tokens, grounding free <1500/day.
|
|
21
|
+
*
|
|
22
|
+
* Regression note (iter-47): this file was absent from feat/adr-135-integrate-tracks
|
|
23
|
+
* because the grounded-query-gemini branch was not cherry-picked during Track integration.
|
|
24
|
+
* The omission caused a 36-question empty-answer failure in iter-42 (13.2% vs 49.1% baseline).
|
|
25
|
+
*
|
|
26
|
+
* Refs: ADR-133, ADR-135, iter-30 research, iter-33, iter-47, #2156
|
|
27
|
+
* Live-tested 2026-05-27: Mercedes Sosa GAIA L1 question → HTTP 200, 4 sources.
|
|
28
|
+
*/
|
|
29
|
+
import { GaiaTool, ToolDefinition } from './types.js';
|
|
30
|
+
export interface GroundedQueryInput {
|
|
31
|
+
query: string;
|
|
32
|
+
max_tokens?: number;
|
|
33
|
+
}
|
|
34
|
+
export interface GroundedQuerySource {
|
|
35
|
+
title: string;
|
|
36
|
+
uri: string;
|
|
37
|
+
snippet?: string;
|
|
38
|
+
}
|
|
39
|
+
export interface GroundedQueryResult {
|
|
40
|
+
/** Synthesised factual answer from Gemini. */
|
|
41
|
+
answer: string;
|
|
42
|
+
/** Web pages Gemini cited when composing its answer. */
|
|
43
|
+
sources: GroundedQuerySource[];
|
|
44
|
+
/** Auto-formulated search queries issued by Gemini. */
|
|
45
|
+
search_queries_used: string[];
|
|
46
|
+
/** True when grounding metadata was returned (sources or search_queries present). */
|
|
47
|
+
grounded: boolean;
|
|
48
|
+
/** Always 'gemini-2.5-flash' for this tool. */
|
|
49
|
+
model: string;
|
|
50
|
+
/** Estimated cost in USD. */
|
|
51
|
+
cost_usd: number;
|
|
52
|
+
}
|
|
53
|
+
interface GeminiPart {
|
|
54
|
+
text?: string;
|
|
55
|
+
}
|
|
56
|
+
interface GeminiGroundingChunk {
|
|
57
|
+
web?: {
|
|
58
|
+
title?: string;
|
|
59
|
+
uri?: string;
|
|
60
|
+
};
|
|
61
|
+
}
|
|
62
|
+
interface GeminiGroundingSupport {
|
|
63
|
+
groundingChunkIndices?: number[];
|
|
64
|
+
segment?: {
|
|
65
|
+
text?: string;
|
|
66
|
+
};
|
|
67
|
+
}
|
|
68
|
+
interface GeminiGroundingMetadata {
|
|
69
|
+
groundingChunks?: GeminiGroundingChunk[];
|
|
70
|
+
webSearchQueries?: string[];
|
|
71
|
+
groundingSupports?: GeminiGroundingSupport[];
|
|
72
|
+
}
|
|
73
|
+
interface GeminiCandidate {
|
|
74
|
+
content?: {
|
|
75
|
+
parts?: GeminiPart[];
|
|
76
|
+
};
|
|
77
|
+
groundingMetadata?: GeminiGroundingMetadata;
|
|
78
|
+
}
|
|
79
|
+
interface GeminiUsageMetadata {
|
|
80
|
+
promptTokenCount?: number;
|
|
81
|
+
candidatesTokenCount?: number;
|
|
82
|
+
}
|
|
83
|
+
interface GeminiGenerateContentResponse {
|
|
84
|
+
candidates?: GeminiCandidate[];
|
|
85
|
+
usageMetadata?: GeminiUsageMetadata;
|
|
86
|
+
}
|
|
87
|
+
export interface GroundedQueryTestHooks {
|
|
88
|
+
/** Override the API key resolver. */
|
|
89
|
+
resolveKey?: () => Promise<string>;
|
|
90
|
+
/** Override the actual HTTP fetch so tests never make live calls. */
|
|
91
|
+
fetchResponse?: (query: string, maxTokens: number, apiKey: string) => Promise<GeminiGenerateContentResponse>;
|
|
92
|
+
}
|
|
93
|
+
export declare const groundedQueryTestHooks: GroundedQueryTestHooks;
|
|
94
|
+
/**
|
|
95
|
+
* Resolve a Google AI API key.
|
|
96
|
+
*
|
|
97
|
+
* Order: env var → gcloud secret → throw.
|
|
98
|
+
* Never returns null — throws with actionable guidance if no key is found.
|
|
99
|
+
*/
|
|
100
|
+
export declare function resolveGoogleAIApiKey(): Promise<string>;
|
|
101
|
+
/**
|
|
102
|
+
* Call Gemini 2.5 Flash with google_search grounding and parse the response.
|
|
103
|
+
*
|
|
104
|
+
* Exported so smoke tests can exercise the parser without live HTTP (via
|
|
105
|
+
* groundedQueryTestHooks.fetchResponse).
|
|
106
|
+
*/
|
|
107
|
+
export declare function callGeminiGrounded(query: string, maxTokens: number, apiKey: string): Promise<GroundedQueryResult>;
|
|
108
|
+
/**
|
|
109
|
+
* Parse a raw Gemini generateContent response into GroundedQueryResult.
|
|
110
|
+
* Exported for unit testing the parser independently.
|
|
111
|
+
*/
|
|
112
|
+
export declare function parseGeminiResponse(body: GeminiGenerateContentResponse): GroundedQueryResult;
|
|
113
|
+
export declare class GroundedQueryTool implements GaiaTool {
|
|
114
|
+
readonly name = "grounded_query";
|
|
115
|
+
readonly definition: ToolDefinition;
|
|
116
|
+
private readonly apiKey?;
|
|
117
|
+
constructor(opts?: {
|
|
118
|
+
apiKey?: string;
|
|
119
|
+
});
|
|
120
|
+
execute(input: Record<string, unknown>): Promise<string>;
|
|
121
|
+
}
|
|
122
|
+
export declare function createGroundedQueryTool(opts?: {
|
|
123
|
+
apiKey?: string;
|
|
124
|
+
}): GroundedQueryTool;
|
|
125
|
+
export {};
|
|
126
|
+
//# sourceMappingURL=grounded_query.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"grounded_query.d.ts","sourceRoot":"","sources":["../../../../src/benchmarks/gaia-tools/grounded_query.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;GA2BG;AAEH,OAAO,EAAE,QAAQ,EAAE,cAAc,EAAE,MAAM,YAAY,CAAC;AAoBtD,MAAM,WAAW,kBAAkB;IACjC,KAAK,EAAE,MAAM,CAAC;IACd,UAAU,CAAC,EAAE,MAAM,CAAC;CACrB;AAED,MAAM,WAAW,mBAAmB;IAClC,KAAK,EAAE,MAAM,CAAC;IACd,GAAG,EAAE,MAAM,CAAC;IACZ,OAAO,CAAC,EAAE,MAAM,CAAC;CAClB;AAED,MAAM,WAAW,mBAAmB;IAClC,8CAA8C;IAC9C,MAAM,EAAE,MAAM,CAAC;IACf,wDAAwD;IACxD,OAAO,EAAE,mBAAmB,EAAE,CAAC;IAC/B,uDAAuD;IACvD,mBAAmB,EAAE,MAAM,EAAE,CAAC;IAC9B,qFAAqF;IACrF,QAAQ,EAAE,OAAO,CAAC;IAClB,+CAA+C;IAC/C,KAAK,EAAE,MAAM,CAAC;IACd,6BAA6B;IAC7B,QAAQ,EAAE,MAAM,CAAC;CAClB;AAMD,UAAU,UAAU;IAClB,IAAI,CAAC,EAAE,MAAM,CAAC;CACf;AAED,UAAU,oBAAoB;IAC5B,GAAG,CAAC,EAAE;QAAE,KAAK,CAAC,EAAE,MAAM,CAAC;QAAC,GAAG,CAAC,EAAE,MAAM,CAAA;KAAE,CAAC;CACxC;AAED,UAAU,sBAAsB;IAC9B,qBAAqB,CAAC,EAAE,MAAM,EAAE,CAAC;IACjC,OAAO,CAAC,EAAE;QAAE,IAAI,CAAC,EAAE,MAAM,CAAA;KAAE,CAAC;CAC7B;AAED,UAAU,uBAAuB;IAC/B,eAAe,CAAC,EAAE,oBAAoB,EAAE,CAAC;IACzC,gBAAgB,CAAC,EAAE,MAAM,EAAE,CAAC;IAC5B,iBAAiB,CAAC,EAAE,sBAAsB,EAAE,CAAC;CAC9C;AAED,UAAU,eAAe;IACvB,OAAO,CAAC,EAAE;QAAE,KAAK,CAAC,EAAE,UAAU,EAAE,CAAA;KAAE,CAAC;IACnC,iBAAiB,CAAC,EAAE,uBAAuB,CAAC;CAC7C;AAED,UAAU,mBAAmB;IAC3B,gBAAgB,CAAC,EAAE,MAAM,CAAC;IAC1B,oBAAoB,CAAC,EAAE,MAAM,CAAC;CAC/B;AAED,UAAU,6BAA6B;IACrC,UAAU,CAAC,EAAE,eAAe,EAAE,CAAC;IAC/B,aAAa,CAAC,EAAE,mBAAmB,CAAC;CACrC;AAMD,MAAM,WAAW,sBAAsB;IACrC,qCAAqC;IACrC,UAAU,CAAC,EAAE,MAAM,OAAO,CAAC,MAAM,CAAC,CAAC;IACnC,qEAAqE;IACrE,aAAa,CAAC,EAAE,CACd,KAAK,EAAE,MAAM,EACb,SAAS,EAAE,MAAM,EACjB,MAAM,EAAE,MAAM,KACX,OAAO,CAAC,6BAA6B,CAAC,CAAC;CAC7C;AAED,eAAO,MAAM,sBAAsB,EAAE,sBAA2B,CAAC;AAMjE;;;;;GAKG;AACH,wBAAsB,qBAAqB,IAAI,OAAO,CAAC,MAAM,CAAC,CAuB7D;AAMD;;;;;GAKG;AACH,wBAAsB,kBAAkB,CACtC,KAAK,EAAE,MAAM,EACb,SAAS,EAAE,MAAM,EACjB,MAAM,EAAE,MAAM,GACb,OAAO,CAAC,mBAAmB,CAAC,CA8B9B;AAED;;;GAGG;AACH,wBAAgB,mBAAmB,CAAC,IAAI,EAAE,6BAA6B,GAAG,mBAAmB,CAyC5F;AA0CD,qBAAa,iBAAkB,YAAW,QAAQ;IAChD,QAAQ,CAAC,IAAI,oBAAoB;IAEjC,QAAQ,CAAC,UAAU,EAAE,cAAc,CAwBjC;IAEF,OAAO,CAAC,QAAQ,CAAC,MAAM,CAAC,CAAS;gBAErB,IAAI,CAAC,EAAE;QAAE,MAAM,CAAC,EAAE,MAAM,CAAA;KAAE;IAIhC,OAAO,CAAC,KAAK,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,GAAG,OAAO,CAAC,MAAM,CAAC;CA2B/D;AAMD,wBAAgB,uBAAuB,CAAC,IAAI,CAAC,EAAE;IAAE,MAAM,CAAC,EAAE,MAAM,CAAA;CAAE,GAAG,iBAAiB,CAErF"}
|
|
@@ -0,0 +1,225 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* GAIA Tool: grounded_query — ADR-135 / iter-33
|
|
3
|
+
*
|
|
4
|
+
* Calls Gemini 2.5 Flash with the google_search grounding tool and returns
|
|
5
|
+
* a synthesised factual answer plus source citations in a single API call.
|
|
6
|
+
*
|
|
7
|
+
* Architectural advantage over web_search (Google Custom Search):
|
|
8
|
+
* - web_search: raw snippets → agent reads + parses → 2-3 extra turns
|
|
9
|
+
* - grounded_query: Gemini synthesises an answer + cites sources in 1 call
|
|
10
|
+
*
|
|
11
|
+
* This is Ruflo's cross-provider stack in action:
|
|
12
|
+
* Gemini grounds (google_search tool) → Sonnet/Haiku reasons → answer
|
|
13
|
+
*
|
|
14
|
+
* Credential resolution (mirrors performance-capability.ts / web_search.ts):
|
|
15
|
+
* 1. process.env.GOOGLE_AI_API_KEY
|
|
16
|
+
* 2. gcloud secrets versions access latest --secret=GOOGLE_AI_API_KEY --project=ruv-dev
|
|
17
|
+
* 3. Throws with clear instructions if neither is available
|
|
18
|
+
*
|
|
19
|
+
* Free tier: 1500 grounded queries/day on Gemini 2.5 Flash.
|
|
20
|
+
* Pricing: ~$0.075/M input tokens, ~$0.30/M output tokens, grounding free <1500/day.
|
|
21
|
+
*
|
|
22
|
+
* Regression note (iter-47): this file was absent from feat/adr-135-integrate-tracks
|
|
23
|
+
* because the grounded-query-gemini branch was not cherry-picked during Track integration.
|
|
24
|
+
* The omission caused a 36-question empty-answer failure in iter-42 (13.2% vs 49.1% baseline).
|
|
25
|
+
*
|
|
26
|
+
* Refs: ADR-133, ADR-135, iter-30 research, iter-33, iter-47, #2156
|
|
27
|
+
* Live-tested 2026-05-27: Mercedes Sosa GAIA L1 question → HTTP 200, 4 sources.
|
|
28
|
+
*/
|
|
29
|
+
// ---------------------------------------------------------------------------
|
|
30
|
+
// Constants
|
|
31
|
+
// ---------------------------------------------------------------------------
|
|
32
|
+
const GEMINI_GENERATE_CONTENT_URL = 'https://generativelanguage.googleapis.com/v1beta/models/gemini-2.5-flash:generateContent';
|
|
33
|
+
const DEFAULT_MAX_TOKENS = 1024;
|
|
34
|
+
const REQUEST_TIMEOUT_MS = 30_000;
|
|
35
|
+
// Gemini 2.5 Flash pricing (USD per 1M tokens, grounding free under 1500/day)
|
|
36
|
+
const PRICE_INPUT_PER_M = 0.075;
|
|
37
|
+
const PRICE_OUTPUT_PER_M = 0.30;
|
|
38
|
+
export const groundedQueryTestHooks = {};
|
|
39
|
+
// ---------------------------------------------------------------------------
|
|
40
|
+
// Credential resolution
|
|
41
|
+
// ---------------------------------------------------------------------------
|
|
42
|
+
/**
|
|
43
|
+
* Resolve a Google AI API key.
|
|
44
|
+
*
|
|
45
|
+
* Order: env var → gcloud secret → throw.
|
|
46
|
+
* Never returns null — throws with actionable guidance if no key is found.
|
|
47
|
+
*/
|
|
48
|
+
export async function resolveGoogleAIApiKey() {
|
|
49
|
+
// 1. Environment variable (fastest path, used in test mocks and CI)
|
|
50
|
+
const envKey = process.env['GOOGLE_AI_API_KEY'];
|
|
51
|
+
if (envKey)
|
|
52
|
+
return envKey;
|
|
53
|
+
// 2. GCP Secrets Manager (matches resolveApiKey pattern in gaia-bench.ts)
|
|
54
|
+
try {
|
|
55
|
+
const { execSync } = await import('node:child_process');
|
|
56
|
+
const key = execSync('gcloud secrets versions access latest --secret=GOOGLE_AI_API_KEY --project=ruv-dev 2>/dev/null', { encoding: 'utf-8', timeout: 5_000 }).trim();
|
|
57
|
+
if (key)
|
|
58
|
+
return key;
|
|
59
|
+
}
|
|
60
|
+
catch {
|
|
61
|
+
// gcloud not installed, project unreachable, or secret not yet created
|
|
62
|
+
}
|
|
63
|
+
throw new Error('grounded_query: No Google AI API key found.\n' +
|
|
64
|
+
'Set GOOGLE_AI_API_KEY env var, or ensure `gcloud` is authenticated and\n' +
|
|
65
|
+
'the secret GOOGLE_AI_API_KEY exists in GCP project ruv-dev.\n' +
|
|
66
|
+
'Get a free key at: https://aistudio.google.com/apikey');
|
|
67
|
+
}
|
|
68
|
+
// ---------------------------------------------------------------------------
|
|
69
|
+
// Gemini Grounding API call
|
|
70
|
+
// ---------------------------------------------------------------------------
|
|
71
|
+
/**
|
|
72
|
+
* Call Gemini 2.5 Flash with google_search grounding and parse the response.
|
|
73
|
+
*
|
|
74
|
+
* Exported so smoke tests can exercise the parser without live HTTP (via
|
|
75
|
+
* groundedQueryTestHooks.fetchResponse).
|
|
76
|
+
*/
|
|
77
|
+
export async function callGeminiGrounded(query, maxTokens, apiKey) {
|
|
78
|
+
const url = `${GEMINI_GENERATE_CONTENT_URL}?key=${encodeURIComponent(apiKey)}`;
|
|
79
|
+
const requestBody = {
|
|
80
|
+
contents: [{ parts: [{ text: query }] }],
|
|
81
|
+
tools: [{ google_search: {} }],
|
|
82
|
+
generationConfig: { maxOutputTokens: maxTokens },
|
|
83
|
+
};
|
|
84
|
+
let rawBody;
|
|
85
|
+
if (groundedQueryTestHooks.fetchResponse) {
|
|
86
|
+
rawBody = await groundedQueryTestHooks.fetchResponse(query, maxTokens, apiKey);
|
|
87
|
+
}
|
|
88
|
+
else {
|
|
89
|
+
const resp = await fetch(url, {
|
|
90
|
+
method: 'POST',
|
|
91
|
+
headers: { 'Content-Type': 'application/json' },
|
|
92
|
+
signal: AbortSignal.timeout(REQUEST_TIMEOUT_MS),
|
|
93
|
+
body: JSON.stringify(requestBody),
|
|
94
|
+
});
|
|
95
|
+
if (!resp.ok) {
|
|
96
|
+
const errText = (await resp.text()).slice(0, 400);
|
|
97
|
+
throw new Error(`grounded_query: Gemini HTTP ${resp.status}: ${errText}`);
|
|
98
|
+
}
|
|
99
|
+
rawBody = (await resp.json());
|
|
100
|
+
}
|
|
101
|
+
return parseGeminiResponse(rawBody);
|
|
102
|
+
}
|
|
103
|
+
/**
|
|
104
|
+
* Parse a raw Gemini generateContent response into GroundedQueryResult.
|
|
105
|
+
* Exported for unit testing the parser independently.
|
|
106
|
+
*/
|
|
107
|
+
export function parseGeminiResponse(body) {
|
|
108
|
+
const candidate = body.candidates?.[0];
|
|
109
|
+
// Extract text answer
|
|
110
|
+
const answer = (candidate?.content?.parts ?? [])
|
|
111
|
+
.map((p) => p.text ?? '')
|
|
112
|
+
.join('');
|
|
113
|
+
// Extract grounding metadata
|
|
114
|
+
const meta = candidate?.groundingMetadata;
|
|
115
|
+
const chunks = meta?.groundingChunks ?? [];
|
|
116
|
+
const searchQueries = meta?.webSearchQueries ?? [];
|
|
117
|
+
// Build source list from grounding chunks (deduplicate by URI)
|
|
118
|
+
const seenUris = new Set();
|
|
119
|
+
const sources = [];
|
|
120
|
+
for (const chunk of chunks) {
|
|
121
|
+
const uri = chunk.web?.uri ?? '';
|
|
122
|
+
const title = chunk.web?.title ?? '';
|
|
123
|
+
if (uri && !seenUris.has(uri)) {
|
|
124
|
+
seenUris.add(uri);
|
|
125
|
+
sources.push({ title, uri });
|
|
126
|
+
}
|
|
127
|
+
}
|
|
128
|
+
// Estimate cost
|
|
129
|
+
const inputTokens = body.usageMetadata?.promptTokenCount ?? 0;
|
|
130
|
+
const outputTokens = body.usageMetadata?.candidatesTokenCount ?? 0;
|
|
131
|
+
const costUsd = (inputTokens / 1_000_000) * PRICE_INPUT_PER_M +
|
|
132
|
+
(outputTokens / 1_000_000) * PRICE_OUTPUT_PER_M;
|
|
133
|
+
return {
|
|
134
|
+
answer,
|
|
135
|
+
sources,
|
|
136
|
+
search_queries_used: searchQueries,
|
|
137
|
+
grounded: sources.length > 0 || searchQueries.length > 0,
|
|
138
|
+
model: 'gemini-2.5-flash',
|
|
139
|
+
cost_usd: costUsd,
|
|
140
|
+
};
|
|
141
|
+
}
|
|
142
|
+
// ---------------------------------------------------------------------------
|
|
143
|
+
// Format output for Claude (mirrors web_search formatResults pattern)
|
|
144
|
+
// ---------------------------------------------------------------------------
|
|
145
|
+
function formatResult(result) {
|
|
146
|
+
const lines = [];
|
|
147
|
+
lines.push(`[grounded_query model: ${result.model}]`);
|
|
148
|
+
if (result.answer) {
|
|
149
|
+
lines.push('');
|
|
150
|
+
lines.push(result.answer.trim());
|
|
151
|
+
}
|
|
152
|
+
if (result.sources.length > 0) {
|
|
153
|
+
lines.push('');
|
|
154
|
+
lines.push('Sources:');
|
|
155
|
+
for (let i = 0; i < result.sources.length; i++) {
|
|
156
|
+
const s = result.sources[i];
|
|
157
|
+
lines.push(`[${i + 1}] ${s.title}`);
|
|
158
|
+
lines.push(` ${s.uri}`);
|
|
159
|
+
}
|
|
160
|
+
}
|
|
161
|
+
if (result.search_queries_used.length > 0) {
|
|
162
|
+
lines.push('');
|
|
163
|
+
lines.push(`Search queries: ${result.search_queries_used.join('; ')}`);
|
|
164
|
+
}
|
|
165
|
+
if (!result.grounded) {
|
|
166
|
+
lines.push('');
|
|
167
|
+
lines.push('[Note: No grounding metadata returned — answer is unverified.]');
|
|
168
|
+
}
|
|
169
|
+
return lines.join('\n');
|
|
170
|
+
}
|
|
171
|
+
// ---------------------------------------------------------------------------
|
|
172
|
+
// GaiaTool implementation
|
|
173
|
+
// ---------------------------------------------------------------------------
|
|
174
|
+
export class GroundedQueryTool {
|
|
175
|
+
name = 'grounded_query';
|
|
176
|
+
definition = {
|
|
177
|
+
name: 'grounded_query',
|
|
178
|
+
description: 'Query Google Search via Gemini 2.5 Flash grounding. ' +
|
|
179
|
+
'Returns a synthesised factual answer with source citations in a single call — ' +
|
|
180
|
+
'more efficient than web_search for factoid questions because Gemini pre-synthesises ' +
|
|
181
|
+
'the answer rather than returning raw snippets. ' +
|
|
182
|
+
'Use this for: factual lookups, "what year did X happen", "how many Y", ' +
|
|
183
|
+
'"who is Z", or any question requiring current web information with a clean answer. ' +
|
|
184
|
+
'Use web_search instead when you need to read a source page in full.',
|
|
185
|
+
input_schema: {
|
|
186
|
+
type: 'object',
|
|
187
|
+
properties: {
|
|
188
|
+
query: {
|
|
189
|
+
type: 'string',
|
|
190
|
+
description: 'Question or factual lookup to ground via Google Search.',
|
|
191
|
+
},
|
|
192
|
+
max_tokens: {
|
|
193
|
+
type: 'number',
|
|
194
|
+
description: `Maximum response length in tokens (default: ${DEFAULT_MAX_TOKENS}).`,
|
|
195
|
+
},
|
|
196
|
+
},
|
|
197
|
+
required: ['query'],
|
|
198
|
+
},
|
|
199
|
+
};
|
|
200
|
+
apiKey;
|
|
201
|
+
constructor(opts) {
|
|
202
|
+
this.apiKey = opts?.apiKey;
|
|
203
|
+
}
|
|
204
|
+
async execute(input) {
|
|
205
|
+
const query = String(input['query'] ?? '').trim();
|
|
206
|
+
if (!query) {
|
|
207
|
+
throw new Error('grounded_query: `query` input is required and must be non-empty.');
|
|
208
|
+
}
|
|
209
|
+
const maxTokens = Math.min(Math.max(64, Number(input['max_tokens'] ?? DEFAULT_MAX_TOKENS)), 8192);
|
|
210
|
+
const resolveKeyFn = groundedQueryTestHooks.resolveKey ?? resolveGoogleAIApiKey;
|
|
211
|
+
const apiKey = this.apiKey ?? (await resolveKeyFn());
|
|
212
|
+
process.stderr.write(`[grounded_query] model=gemini-2.5-flash query=${JSON.stringify(query)}\n`);
|
|
213
|
+
const result = await callGeminiGrounded(query, maxTokens, apiKey);
|
|
214
|
+
process.stderr.write(`[grounded_query] grounded=${result.grounded} sources=${result.sources.length} ` +
|
|
215
|
+
`cost_usd=${result.cost_usd.toFixed(6)}\n`);
|
|
216
|
+
return formatResult(result);
|
|
217
|
+
}
|
|
218
|
+
}
|
|
219
|
+
// ---------------------------------------------------------------------------
|
|
220
|
+
// Convenience factory
|
|
221
|
+
// ---------------------------------------------------------------------------
|
|
222
|
+
export function createGroundedQueryTool(opts) {
|
|
223
|
+
return new GroundedQueryTool(opts);
|
|
224
|
+
}
|
|
225
|
+
//# sourceMappingURL=grounded_query.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"grounded_query.js","sourceRoot":"","sources":["../../../../src/benchmarks/gaia-tools/grounded_query.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;GA2BG;AAIH,8EAA8E;AAC9E,YAAY;AACZ,8EAA8E;AAE9E,MAAM,2BAA2B,GAC/B,0FAA0F,CAAC;AAE7F,MAAM,kBAAkB,GAAG,IAAI,CAAC;AAChC,MAAM,kBAAkB,GAAG,MAAM,CAAC;AAElC,8EAA8E;AAC9E,MAAM,iBAAiB,GAAG,KAAK,CAAC;AAChC,MAAM,kBAAkB,GAAG,IAAI,CAAC;AAqFhC,MAAM,CAAC,MAAM,sBAAsB,GAA2B,EAAE,CAAC;AAEjE,8EAA8E;AAC9E,wBAAwB;AACxB,8EAA8E;AAE9E;;;;;GAKG;AACH,MAAM,CAAC,KAAK,UAAU,qBAAqB;IACzC,oEAAoE;IACpE,MAAM,MAAM,GAAG,OAAO,CAAC,GAAG,CAAC,mBAAmB,CAAC,CAAC;IAChD,IAAI,MAAM;QAAE,OAAO,MAAM,CAAC;IAE1B,0EAA0E;IAC1E,IAAI,CAAC;QACH,MAAM,EAAE,QAAQ,EAAE,GAAG,MAAM,MAAM,CAAC,oBAAoB,CAAC,CAAC;QACxD,MAAM,GAAG,GAAG,QAAQ,CAClB,gGAAgG,EAChG,EAAE,QAAQ,EAAE,OAAO,EAAE,OAAO,EAAE,KAAK,EAAE,CACtC,CAAC,IAAI,EAAE,CAAC;QACT,IAAI,GAAG;YAAE,OAAO,GAAG,CAAC;IACtB,CAAC;IAAC,MAAM,CAAC;QACP,uEAAuE;IACzE,CAAC;IAED,MAAM,IAAI,KAAK,CACb,+CAA+C;QAC7C,0EAA0E;QAC1E,+DAA+D;QAC/D,uDAAuD,CAC1D,CAAC;AACJ,CAAC;AAED,8EAA8E;AAC9E,4BAA4B;AAC5B,8EAA8E;AAE9E;;;;;GAKG;AACH,MAAM,CAAC,KAAK,UAAU,kBAAkB,CACtC,KAAa,EACb,SAAiB,EACjB,MAAc;IAEd,MAAM,GAAG,GAAG,GAAG,2BAA2B,QAAQ,kBAAkB,CAAC,MAAM,CAAC,EAAE,CAAC;IAE/E,MAAM,WAAW,GAAG;QAClB,QAAQ,EAAE,CAAC,EAAE,KAAK,EAAE,CAAC,EAAE,IAAI,EAAE,KAAK,EAAE,CAAC,EAAE,CAAC;QACxC,KAAK,EAAE,CAAC,EAAE,aAAa,EAAE,EAAE,EAAE,CAAC;QAC9B,gBAAgB,EAAE,EAAE,eAAe,EAAE,SAAS,EAAE;KACjD,CAAC;IAEF,IAAI,OAAsC,CAAC;IAE3C,IAAI,sBAAsB,CAAC,aAAa,EAAE,CAAC;QACzC,OAAO,GAAG,MAAM,sBAAsB,CAAC,aAAa,CAAC,KAAK,EAAE,SAAS,EAAE,MAAM,CAAC,CAAC;IACjF,CAAC;SAAM,CAAC;QACN,MAAM,IAAI,GAAG,MAAM,KAAK,CAAC,GAAG,EAAE;YAC5B,MAAM,EAAE,MAAM;YACd,OAAO,EAAE,EAAE,cAAc,EAAE,kBAAkB,EAAE;YAC/C,MAAM,EAAE,WAAW,CAAC,OAAO,CAAC,kBAAkB,CAAC;YAC/C,IAAI,EAAE,IAAI,CAAC,SAAS,CAAC,WAAW,CAAC;SAClC,CAAC,CAAC;QAEH,IAAI,CAAC,IAAI,CAAC,EAAE,EAAE,CAAC;YACb,MAAM,OAAO,GAAG,CAAC,MAAM,IAAI,CAAC,IAAI,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC,CAAC;YAClD,MAAM,IAAI,KAAK,CAAC,+BAA+B,IAAI,CAAC,MAAM,KAAK,OAAO,EAAE,CAAC,CAAC;QAC5E,CAAC;QAED,OAAO,GAAG,CAAC,MAAM,IAAI,CAAC,IAAI,EAAE,CAAkC,CAAC;IACjE,CAAC;IAED,OAAO,mBAAmB,CAAC,OAAO,CAAC,CAAC;AACtC,CAAC;AAED;;;GAGG;AACH,MAAM,UAAU,mBAAmB,CAAC,IAAmC;IACrE,MAAM,SAAS,GAAG,IAAI,CAAC,UAAU,EAAE,CAAC,CAAC,CAAC,CAAC;IAEvC,sBAAsB;IACtB,MAAM,MAAM,GAAG,CAAC,SAAS,EAAE,OAAO,EAAE,KAAK,IAAI,EAAE,CAAC;SAC7C,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,IAAI,EAAE,CAAC;SACxB,IAAI,CAAC,EAAE,CAAC,CAAC;IAEZ,6BAA6B;IAC7B,MAAM,IAAI,GAAG,SAAS,EAAE,iBAAiB,CAAC;IAC1C,MAAM,MAAM,GAAG,IAAI,EAAE,eAAe,IAAI,EAAE,CAAC;IAC3C,MAAM,aAAa,GAAG,IAAI,EAAE,gBAAgB,IAAI,EAAE,CAAC;IAEnD,+DAA+D;IAC/D,MAAM,QAAQ,GAAG,IAAI,GAAG,EAAU,CAAC;IACnC,MAAM,OAAO,GAA0B,EAAE,CAAC;IAE1C,KAAK,MAAM,KAAK,IAAI,MAAM,EAAE,CAAC;QAC3B,MAAM,GAAG,GAAG,KAAK,CAAC,GAAG,EAAE,GAAG,IAAI,EAAE,CAAC;QACjC,MAAM,KAAK,GAAG,KAAK,CAAC,GAAG,EAAE,KAAK,IAAI,EAAE,CAAC;QACrC,IAAI,GAAG,IAAI,CAAC,QAAQ,CAAC,GAAG,CAAC,GAAG,CAAC,EAAE,CAAC;YAC9B,QAAQ,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC;YAClB,OAAO,CAAC,IAAI,CAAC,EAAE,KAAK,EAAE,GAAG,EAAE,CAAC,CAAC;QAC/B,CAAC;IACH,CAAC;IAED,gBAAgB;IAChB,MAAM,WAAW,GAAG,IAAI,CAAC,aAAa,EAAE,gBAAgB,IAAI,CAAC,CAAC;IAC9D,MAAM,YAAY,GAAG,IAAI,CAAC,aAAa,EAAE,oBAAoB,IAAI,CAAC,CAAC;IACnE,MAAM,OAAO,GACX,CAAC,WAAW,GAAG,SAAS,CAAC,GAAG,iBAAiB;QAC7C,CAAC,YAAY,GAAG,SAAS,CAAC,GAAG,kBAAkB,CAAC;IAElD,OAAO;QACL,MAAM;QACN,OAAO;QACP,mBAAmB,EAAE,aAAa;QAClC,QAAQ,EAAE,OAAO,CAAC,MAAM,GAAG,CAAC,IAAI,aAAa,CAAC,MAAM,GAAG,CAAC;QACxD,KAAK,EAAE,kBAAkB;QACzB,QAAQ,EAAE,OAAO;KAClB,CAAC;AACJ,CAAC;AAED,8EAA8E;AAC9E,sEAAsE;AACtE,8EAA8E;AAE9E,SAAS,YAAY,CAAC,MAA2B;IAC/C,MAAM,KAAK,GAAa,EAAE,CAAC;IAC3B,KAAK,CAAC,IAAI,CAAC,0BAA0B,MAAM,CAAC,KAAK,GAAG,CAAC,CAAC;IAEtD,IAAI,MAAM,CAAC,MAAM,EAAE,CAAC;QAClB,KAAK,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;QACf,KAAK,CAAC,IAAI,CAAC,MAAM,CAAC,MAAM,CAAC,IAAI,EAAE,CAAC,CAAC;IACnC,CAAC;IAED,IAAI,MAAM,CAAC,OAAO,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QAC9B,KAAK,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;QACf,KAAK,CAAC,IAAI,CAAC,UAAU,CAAC,CAAC;QACvB,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,MAAM,CAAC,OAAO,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;YAC/C,MAAM,CAAC,GAAG,MAAM,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC;YAC5B,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,GAAG,CAAC,KAAK,CAAC,CAAC,KAAK,EAAE,CAAC,CAAC;YACpC,KAAK,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC,GAAG,EAAE,CAAC,CAAC;QAC7B,CAAC;IACH,CAAC;IAED,IAAI,MAAM,CAAC,mBAAmB,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QAC1C,KAAK,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;QACf,KAAK,CAAC,IAAI,CAAC,mBAAmB,MAAM,CAAC,mBAAmB,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;IACzE,CAAC;IAED,IAAI,CAAC,MAAM,CAAC,QAAQ,EAAE,CAAC;QACrB,KAAK,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;QACf,KAAK,CAAC,IAAI,CAAC,gEAAgE,CAAC,CAAC;IAC/E,CAAC;IAED,OAAO,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;AAC1B,CAAC;AAED,8EAA8E;AAC9E,0BAA0B;AAC1B,8EAA8E;AAE9E,MAAM,OAAO,iBAAiB;IACnB,IAAI,GAAG,gBAAgB,CAAC;IAExB,UAAU,GAAmB;QACpC,IAAI,EAAE,gBAAgB;QACtB,WAAW,EACT,sDAAsD;YACtD,gFAAgF;YAChF,sFAAsF;YACtF,iDAAiD;YACjD,yEAAyE;YACzE,qFAAqF;YACrF,qEAAqE;QACvE,YAAY,EAAE;YACZ,IAAI,EAAE,QAAQ;YACd,UAAU,EAAE;gBACV,KAAK,EAAE;oBACL,IAAI,EAAE,QAAQ;oBACd,WAAW,EAAE,yDAAyD;iBACvE;gBACD,UAAU,EAAE;oBACV,IAAI,EAAE,QAAQ;oBACd,WAAW,EAAE,+CAA+C,kBAAkB,IAAI;iBACnF;aACF;YACD,QAAQ,EAAE,CAAC,OAAO,CAAC;SACpB;KACF,CAAC;IAEe,MAAM,CAAU;IAEjC,YAAY,IAA0B;QACpC,IAAI,CAAC,MAAM,GAAG,IAAI,EAAE,MAAM,CAAC;IAC7B,CAAC;IAED,KAAK,CAAC,OAAO,CAAC,KAA8B;QAC1C,MAAM,KAAK,GAAG,MAAM,CAAC,KAAK,CAAC,OAAO,CAAC,IAAI,EAAE,CAAC,CAAC,IAAI,EAAE,CAAC;QAClD,IAAI,CAAC,KAAK,EAAE,CAAC;YACX,MAAM,IAAI,KAAK,CAAC,kEAAkE,CAAC,CAAC;QACtF,CAAC;QAED,MAAM,SAAS,GAAG,IAAI,CAAC,GAAG,CACxB,IAAI,CAAC,GAAG,CAAC,EAAE,EAAE,MAAM,CAAC,KAAK,CAAC,YAAY,CAAC,IAAI,kBAAkB,CAAC,CAAC,EAC/D,IAAI,CACL,CAAC;QAEF,MAAM,YAAY,GAAG,sBAAsB,CAAC,UAAU,IAAI,qBAAqB,CAAC;QAChF,MAAM,MAAM,GAAG,IAAI,CAAC,MAAM,IAAI,CAAC,MAAM,YAAY,EAAE,CAAC,CAAC;QAErD,OAAO,CAAC,MAAM,CAAC,KAAK,CAClB,iDAAiD,IAAI,CAAC,SAAS,CAAC,KAAK,CAAC,IAAI,CAC3E,CAAC;QAEF,MAAM,MAAM,GAAG,MAAM,kBAAkB,CAAC,KAAK,EAAE,SAAS,EAAE,MAAM,CAAC,CAAC;QAElE,OAAO,CAAC,MAAM,CAAC,KAAK,CAClB,6BAA6B,MAAM,CAAC,QAAQ,YAAY,MAAM,CAAC,OAAO,CAAC,MAAM,GAAG;YAC9E,YAAY,MAAM,CAAC,QAAQ,CAAC,OAAO,CAAC,CAAC,CAAC,IAAI,CAC7C,CAAC;QAEF,OAAO,YAAY,CAAC,MAAM,CAAC,CAAC;IAC9B,CAAC;CACF;AAED,8EAA8E;AAC9E,sBAAsB;AACtB,8EAA8E;AAE9E,MAAM,UAAU,uBAAuB,CAAC,IAA0B;IAChE,OAAO,IAAI,iBAAiB,CAAC,IAAI,CAAC,CAAC;AACrC,CAAC"}
|
|
@@ -0,0 +1,24 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* Smoke tests for grounded_query.ts — ADR-135 / iter-33
|
|
3
|
+
*
|
|
4
|
+
* All tests are mock-based using groundedQueryTestHooks for ESM-safe injection.
|
|
5
|
+
* NO live API calls are made.
|
|
6
|
+
*
|
|
7
|
+
* Coverage:
|
|
8
|
+
* 1. resolveGoogleAIApiKey — returns key from env var
|
|
9
|
+
* 2. resolveGoogleAIApiKey — throws when env var absent and gcloud unavailable
|
|
10
|
+
* 3. parseGeminiResponse — well-formed response → correct GroundedQueryResult shape
|
|
11
|
+
* 4. parseGeminiResponse — empty grounding metadata → grounded=false, empty sources
|
|
12
|
+
* 5. parseGeminiResponse — deduplicates sources by URI
|
|
13
|
+
* 6. parseGeminiResponse — cost calculation uses token counts
|
|
14
|
+
* 7. GroundedQueryTool.execute — happy path: answer + sources in output
|
|
15
|
+
* 8. GroundedQueryTool.execute — HTTP 4xx surfaces useful error message
|
|
16
|
+
* 9. GroundedQueryTool.execute — ungrounded response includes note in output
|
|
17
|
+
* 10. GroundedQueryTool.execute — rejects empty query
|
|
18
|
+
* 11. GroundedQueryTool.execute — max_tokens clamped to 8192
|
|
19
|
+
* 12. GroundedQueryTool.execute — uses injected apiKey from constructor
|
|
20
|
+
*
|
|
21
|
+
* Run with: npx tsx grounded_query.smoke.ts (from gaia-tools directory)
|
|
22
|
+
*/
|
|
23
|
+
export {};
|
|
24
|
+
//# sourceMappingURL=grounded_query.smoke.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"grounded_query.smoke.d.ts","sourceRoot":"","sources":["../../../../src/benchmarks/gaia-tools/grounded_query.smoke.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;GAqBG"}
|