@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 +1,19 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* Smoke tests for the GAIA convergence layer (gaia-convergence.ts).
|
|
3
|
+
*
|
|
4
|
+
* 10 assertions covering:
|
|
5
|
+
* T1: max_turns hit → checkConvergenceTriggers returns 'max_turns'
|
|
6
|
+
* T2: same tool+args 3× in window → returns 'loop'
|
|
7
|
+
* T3: 120k tokens → returns 'token_overflow'
|
|
8
|
+
* T4: forceCommit with prose-embedded answer in history → extraction succeeds
|
|
9
|
+
* T5: forceCommit with explicit FINAL_ANSWER: X in response → returns X
|
|
10
|
+
* T6: forceCommit on truly empty/looping conversation → returns null (graceful)
|
|
11
|
+
* T7: argsHash deterministic — same args → same hash
|
|
12
|
+
* T8: argsHash distinguishes — different args → different hash
|
|
13
|
+
* T9: ConvergenceState records turn counts correctly via recordTurn
|
|
14
|
+
* T10: Anti-loop window correctly slides — only last 5 turns matter
|
|
15
|
+
*
|
|
16
|
+
* Refs: #2156, iter 62
|
|
17
|
+
*/
|
|
18
|
+
export {};
|
|
19
|
+
//# sourceMappingURL=gaia-convergence.smoke.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"gaia-convergence.smoke.d.ts","sourceRoot":"","sources":["../../../src/benchmarks/gaia-convergence.smoke.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;GAgBG"}
|
|
@@ -0,0 +1,246 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* Smoke tests for the GAIA convergence layer (gaia-convergence.ts).
|
|
3
|
+
*
|
|
4
|
+
* 10 assertions covering:
|
|
5
|
+
* T1: max_turns hit → checkConvergenceTriggers returns 'max_turns'
|
|
6
|
+
* T2: same tool+args 3× in window → returns 'loop'
|
|
7
|
+
* T3: 120k tokens → returns 'token_overflow'
|
|
8
|
+
* T4: forceCommit with prose-embedded answer in history → extraction succeeds
|
|
9
|
+
* T5: forceCommit with explicit FINAL_ANSWER: X in response → returns X
|
|
10
|
+
* T6: forceCommit on truly empty/looping conversation → returns null (graceful)
|
|
11
|
+
* T7: argsHash deterministic — same args → same hash
|
|
12
|
+
* T8: argsHash distinguishes — different args → different hash
|
|
13
|
+
* T9: ConvergenceState records turn counts correctly via recordTurn
|
|
14
|
+
* T10: Anti-loop window correctly slides — only last 5 turns matter
|
|
15
|
+
*
|
|
16
|
+
* Refs: #2156, iter 62
|
|
17
|
+
*/
|
|
18
|
+
import assert from 'node:assert';
|
|
19
|
+
import { checkConvergenceTriggers, forceCommit, argsHash, createConvergenceState, recordTurn, extractFinalAnswerFromText, extractFromPriorMessages, TOKEN_OVERFLOW_THRESHOLD, LOOP_WINDOW_SIZE, } from './gaia-convergence.js';
|
|
20
|
+
// ---------------------------------------------------------------------------
|
|
21
|
+
// Test harness
|
|
22
|
+
// ---------------------------------------------------------------------------
|
|
23
|
+
let passed = 0;
|
|
24
|
+
let failed = 0;
|
|
25
|
+
function test(name, fn) {
|
|
26
|
+
const result = fn();
|
|
27
|
+
if (result && typeof result.then === 'function') {
|
|
28
|
+
result
|
|
29
|
+
.then(() => {
|
|
30
|
+
console.log(` PASS ${name}`);
|
|
31
|
+
passed++;
|
|
32
|
+
})
|
|
33
|
+
.catch((err) => {
|
|
34
|
+
console.error(` FAIL ${name}: ${err.message ?? err}`);
|
|
35
|
+
failed++;
|
|
36
|
+
});
|
|
37
|
+
}
|
|
38
|
+
else {
|
|
39
|
+
try {
|
|
40
|
+
console.log(` PASS ${name}`);
|
|
41
|
+
passed++;
|
|
42
|
+
}
|
|
43
|
+
catch (err) {
|
|
44
|
+
console.error(` FAIL ${name}: ${err.message ?? err}`);
|
|
45
|
+
failed++;
|
|
46
|
+
}
|
|
47
|
+
}
|
|
48
|
+
}
|
|
49
|
+
async function testAsync(name, fn) {
|
|
50
|
+
try {
|
|
51
|
+
await fn();
|
|
52
|
+
console.log(` PASS ${name}`);
|
|
53
|
+
passed++;
|
|
54
|
+
}
|
|
55
|
+
catch (err) {
|
|
56
|
+
console.error(` FAIL ${name}: ${err.message ?? err}`);
|
|
57
|
+
failed++;
|
|
58
|
+
}
|
|
59
|
+
}
|
|
60
|
+
// ---------------------------------------------------------------------------
|
|
61
|
+
// T1: max_turns hit → returns 'max_turns'
|
|
62
|
+
// ---------------------------------------------------------------------------
|
|
63
|
+
test('T1: max_turns hit returns max_turns', () => {
|
|
64
|
+
const state = {
|
|
65
|
+
turnCount: 12,
|
|
66
|
+
totalTokens: 1000,
|
|
67
|
+
toolCalls: [],
|
|
68
|
+
detectedFailureMode: null,
|
|
69
|
+
};
|
|
70
|
+
const result = checkConvergenceTriggers(state, 12);
|
|
71
|
+
assert.strictEqual(result, 'max_turns', `Expected 'max_turns', got ${result}`);
|
|
72
|
+
});
|
|
73
|
+
// ---------------------------------------------------------------------------
|
|
74
|
+
// T2: same tool+args 3× in last 5 turns → returns 'loop'
|
|
75
|
+
// ---------------------------------------------------------------------------
|
|
76
|
+
test('T2: loop detection — same tool+args 3x in window', () => {
|
|
77
|
+
const hash = argsHash('web_search', { query: 'test' });
|
|
78
|
+
const state = {
|
|
79
|
+
turnCount: 5,
|
|
80
|
+
totalTokens: 5000,
|
|
81
|
+
toolCalls: [
|
|
82
|
+
{ name: 'web_search', argsHash: hash, turn: 1 },
|
|
83
|
+
{ name: 'web_search', argsHash: hash, turn: 2 },
|
|
84
|
+
{ name: 'web_search', argsHash: hash, turn: 3 },
|
|
85
|
+
],
|
|
86
|
+
detectedFailureMode: null,
|
|
87
|
+
};
|
|
88
|
+
const result = checkConvergenceTriggers(state, 15);
|
|
89
|
+
assert.strictEqual(result, 'loop', `Expected 'loop', got ${result}`);
|
|
90
|
+
});
|
|
91
|
+
// ---------------------------------------------------------------------------
|
|
92
|
+
// T3: 120k tokens → returns 'token_overflow'
|
|
93
|
+
// ---------------------------------------------------------------------------
|
|
94
|
+
test('T3: token_overflow at threshold', () => {
|
|
95
|
+
const state = {
|
|
96
|
+
turnCount: 5,
|
|
97
|
+
totalTokens: TOKEN_OVERFLOW_THRESHOLD,
|
|
98
|
+
toolCalls: [],
|
|
99
|
+
detectedFailureMode: null,
|
|
100
|
+
};
|
|
101
|
+
const result = checkConvergenceTriggers(state, 20);
|
|
102
|
+
assert.strictEqual(result, 'token_overflow', `Expected 'token_overflow', got ${result}`);
|
|
103
|
+
});
|
|
104
|
+
// ---------------------------------------------------------------------------
|
|
105
|
+
// T4: forceCommit with prose-embedded answer in prior history → extraction succeeds
|
|
106
|
+
// ---------------------------------------------------------------------------
|
|
107
|
+
await testAsync('T4: forceCommit — prose-embedded FINAL_ANSWER in prior history', async () => {
|
|
108
|
+
const messages = [
|
|
109
|
+
{ role: 'user', content: 'What is 2+2?' },
|
|
110
|
+
{
|
|
111
|
+
role: 'assistant',
|
|
112
|
+
content: 'Let me think about this. The answer is definitely 4. FINAL_ANSWER: 4',
|
|
113
|
+
},
|
|
114
|
+
{ role: 'user', content: 'tool results...' },
|
|
115
|
+
{ role: 'assistant', content: 'I got confused. Let me recalculate.' },
|
|
116
|
+
];
|
|
117
|
+
// callModel returns an empty response (simulating no new FINAL_ANSWER)
|
|
118
|
+
const callModel = async () => 'I need to think more about this.';
|
|
119
|
+
const result = await forceCommit(messages, callModel, 'max_turns');
|
|
120
|
+
assert.strictEqual(result.answer, '4', `Expected '4', got ${result.answer}`);
|
|
121
|
+
assert.strictEqual(result.usedFallback, true, 'Expected usedFallback=true for history scan');
|
|
122
|
+
});
|
|
123
|
+
// ---------------------------------------------------------------------------
|
|
124
|
+
// T5: forceCommit with explicit FINAL_ANSWER: X in forced response → returns X
|
|
125
|
+
// ---------------------------------------------------------------------------
|
|
126
|
+
await testAsync('T5: forceCommit — explicit FINAL_ANSWER in forced-commit response', async () => {
|
|
127
|
+
const messages = [
|
|
128
|
+
{ role: 'user', content: 'What is the capital of France?' },
|
|
129
|
+
{ role: 'assistant', content: 'I need to search for this.' },
|
|
130
|
+
];
|
|
131
|
+
// callModel returns a response WITH FINAL_ANSWER
|
|
132
|
+
const callModel = async () => 'Based on my knowledge, the capital is Paris. FINAL_ANSWER: Paris';
|
|
133
|
+
const result = await forceCommit(messages, callModel, 'max_turns');
|
|
134
|
+
assert.strictEqual(result.answer, 'Paris', `Expected 'Paris', got ${result.answer}`);
|
|
135
|
+
assert.strictEqual(result.usedFallback, false, 'Expected usedFallback=false when direct extraction');
|
|
136
|
+
assert.strictEqual(result.triggerMode, 'max_turns');
|
|
137
|
+
});
|
|
138
|
+
// ---------------------------------------------------------------------------
|
|
139
|
+
// T6: forceCommit on truly empty conversation → returns null (graceful)
|
|
140
|
+
// ---------------------------------------------------------------------------
|
|
141
|
+
await testAsync('T6: forceCommit — graceful null on empty conversation', async () => {
|
|
142
|
+
const messages = [
|
|
143
|
+
{ role: 'user', content: 'What is something unknowable?' },
|
|
144
|
+
{ role: 'assistant', content: 'I have been calling tools repeatedly with no progress.' },
|
|
145
|
+
];
|
|
146
|
+
// callModel returns nothing useful
|
|
147
|
+
const callModel = async () => 'I cannot determine the answer to this question.';
|
|
148
|
+
const result = await forceCommit(messages, callModel, 'loop');
|
|
149
|
+
assert.strictEqual(result.answer, null, `Expected null, got ${result.answer}`);
|
|
150
|
+
assert.strictEqual(result.triggerMode, 'loop');
|
|
151
|
+
});
|
|
152
|
+
// ---------------------------------------------------------------------------
|
|
153
|
+
// T7: argsHash deterministic — same args → same hash
|
|
154
|
+
// ---------------------------------------------------------------------------
|
|
155
|
+
test('T7: argsHash is deterministic', () => {
|
|
156
|
+
const h1 = argsHash('web_search', { query: 'hello world', limit: 5 });
|
|
157
|
+
const h2 = argsHash('web_search', { query: 'hello world', limit: 5 });
|
|
158
|
+
assert.strictEqual(h1, h2, 'Same inputs must produce identical hash');
|
|
159
|
+
assert.ok(h1.length === 16, `Hash should be 16 hex chars, got ${h1.length}`);
|
|
160
|
+
});
|
|
161
|
+
// ---------------------------------------------------------------------------
|
|
162
|
+
// T8: argsHash distinguishes — different args → different hash
|
|
163
|
+
// ---------------------------------------------------------------------------
|
|
164
|
+
test('T8: argsHash distinguishes different args', () => {
|
|
165
|
+
const h1 = argsHash('web_search', { query: 'hello' });
|
|
166
|
+
const h2 = argsHash('web_search', { query: 'world' });
|
|
167
|
+
assert.notStrictEqual(h1, h2, 'Different args must produce different hash');
|
|
168
|
+
const h3 = argsHash('calculator', { query: 'hello' });
|
|
169
|
+
assert.notStrictEqual(h1, h3, 'Different tool names must produce different hash');
|
|
170
|
+
});
|
|
171
|
+
// ---------------------------------------------------------------------------
|
|
172
|
+
// T9: ConvergenceState records turn counts correctly via recordTurn
|
|
173
|
+
// ---------------------------------------------------------------------------
|
|
174
|
+
test('T9: recordTurn increments turnCount and totalTokens', () => {
|
|
175
|
+
const state = createConvergenceState();
|
|
176
|
+
assert.strictEqual(state.turnCount, 0);
|
|
177
|
+
assert.strictEqual(state.totalTokens, 0);
|
|
178
|
+
assert.strictEqual(state.toolCalls.length, 0);
|
|
179
|
+
recordTurn(state, 1500, [
|
|
180
|
+
{ name: 'web_search', args: { query: 'test' } },
|
|
181
|
+
{ name: 'calculator', args: { expr: '2+2' } },
|
|
182
|
+
]);
|
|
183
|
+
assert.strictEqual(state.turnCount, 1, `Expected turnCount=1, got ${state.turnCount}`);
|
|
184
|
+
assert.strictEqual(state.totalTokens, 1500, `Expected totalTokens=1500, got ${state.totalTokens}`);
|
|
185
|
+
assert.strictEqual(state.toolCalls.length, 2, `Expected 2 tool calls recorded, got ${state.toolCalls.length}`);
|
|
186
|
+
assert.strictEqual(state.toolCalls[0].name, 'web_search');
|
|
187
|
+
assert.strictEqual(state.toolCalls[0].turn, 1);
|
|
188
|
+
recordTurn(state, 2000, []);
|
|
189
|
+
assert.strictEqual(state.turnCount, 2);
|
|
190
|
+
assert.strictEqual(state.totalTokens, 3500);
|
|
191
|
+
});
|
|
192
|
+
// ---------------------------------------------------------------------------
|
|
193
|
+
// T10: Anti-loop window correctly slides — only last LOOP_WINDOW_SIZE turns matter
|
|
194
|
+
// ---------------------------------------------------------------------------
|
|
195
|
+
test('T10: loop window slides — old calls outside window do not trigger', () => {
|
|
196
|
+
const hash = argsHash('web_search', { query: 'test' });
|
|
197
|
+
// 4 repeated calls followed by 5 different calls — window is last 5 only
|
|
198
|
+
const oldCalls = Array.from({ length: 4 }, (_, i) => ({
|
|
199
|
+
name: 'web_search', argsHash: hash, turn: i + 1,
|
|
200
|
+
}));
|
|
201
|
+
const recentCalls = Array.from({ length: LOOP_WINDOW_SIZE }, (_, i) => ({
|
|
202
|
+
name: `tool_${i}`, argsHash: argsHash(`tool_${i}`, {}), turn: i + 5,
|
|
203
|
+
}));
|
|
204
|
+
const state = {
|
|
205
|
+
turnCount: 9,
|
|
206
|
+
totalTokens: 9000,
|
|
207
|
+
toolCalls: [...oldCalls, ...recentCalls],
|
|
208
|
+
detectedFailureMode: null,
|
|
209
|
+
};
|
|
210
|
+
const result = checkConvergenceTriggers(state, 20);
|
|
211
|
+
assert.strictEqual(result, null, `Expected null (old calls outside window), got ${result}`);
|
|
212
|
+
});
|
|
213
|
+
// ---------------------------------------------------------------------------
|
|
214
|
+
// Bonus: extractFinalAnswerFromText edge cases
|
|
215
|
+
// ---------------------------------------------------------------------------
|
|
216
|
+
test('T11 (bonus): extractFinalAnswerFromText handles multiline and trailing content', () => {
|
|
217
|
+
const text = 'After much deliberation:\nFINAL_ANSWER: 42\nSome extra text';
|
|
218
|
+
const result = extractFinalAnswerFromText(text);
|
|
219
|
+
assert.strictEqual(result, '42', `Expected '42', got ${result}`);
|
|
220
|
+
});
|
|
221
|
+
test('T12 (bonus): extractFromPriorMessages scans in reverse order', () => {
|
|
222
|
+
const messages = [
|
|
223
|
+
{ role: 'user', content: 'Question' },
|
|
224
|
+
{ role: 'assistant', content: 'First answer FINAL_ANSWER: wrong' },
|
|
225
|
+
{ role: 'user', content: 'tool results' },
|
|
226
|
+
{ role: 'assistant', content: 'After more research FINAL_ANSWER: correct' },
|
|
227
|
+
];
|
|
228
|
+
// Should return 'correct' (last assistant message first in reverse scan)
|
|
229
|
+
const result = extractFromPriorMessages(messages);
|
|
230
|
+
assert.strictEqual(result, 'correct', `Expected 'correct' (most recent), got ${result}`);
|
|
231
|
+
});
|
|
232
|
+
// ---------------------------------------------------------------------------
|
|
233
|
+
// Summary
|
|
234
|
+
// ---------------------------------------------------------------------------
|
|
235
|
+
// Wait for all async tests to settle, then print summary
|
|
236
|
+
setTimeout(() => {
|
|
237
|
+
const total = passed + failed;
|
|
238
|
+
console.log('');
|
|
239
|
+
console.log('=== GAIA Convergence Layer Smoke Test ===');
|
|
240
|
+
console.log(`Pass rate: ${passed}/${total}`);
|
|
241
|
+
console.log(`Status: ${failed === 0 ? 'ALL PASSED' : `${failed} FAILED`}`);
|
|
242
|
+
if (failed > 0) {
|
|
243
|
+
process.exit(1);
|
|
244
|
+
}
|
|
245
|
+
}, 200);
|
|
246
|
+
//# sourceMappingURL=gaia-convergence.smoke.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"gaia-convergence.smoke.js","sourceRoot":"","sources":["../../../src/benchmarks/gaia-convergence.smoke.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;GAgBG;AAEH,OAAO,MAAM,MAAM,aAAa,CAAC;AACjC,OAAO,EACL,wBAAwB,EACxB,WAAW,EACX,QAAQ,EACR,sBAAsB,EACtB,UAAU,EACV,0BAA0B,EAC1B,wBAAwB,EACxB,wBAAwB,EAExB,gBAAgB,GACjB,MAAM,uBAAuB,CAAC;AAG/B,8EAA8E;AAC9E,eAAe;AACf,8EAA8E;AAE9E,IAAI,MAAM,GAAG,CAAC,CAAC;AACf,IAAI,MAAM,GAAG,CAAC,CAAC;AAEf,SAAS,IAAI,CAAC,IAAY,EAAE,EAA8B;IACxD,MAAM,MAAM,GAAG,EAAE,EAAE,CAAC;IACpB,IAAI,MAAM,IAAI,OAAQ,MAAwB,CAAC,IAAI,KAAK,UAAU,EAAE,CAAC;QAClE,MAAwB;aACtB,IAAI,CAAC,GAAG,EAAE;YACT,OAAO,CAAC,GAAG,CAAC,WAAW,IAAI,EAAE,CAAC,CAAC;YAC/B,MAAM,EAAE,CAAC;QACX,CAAC,CAAC;aACD,KAAK,CAAC,CAAC,GAAY,EAAE,EAAE;YACtB,OAAO,CAAC,KAAK,CAAC,WAAW,IAAI,KAAM,GAAa,CAAC,OAAO,IAAI,GAAG,EAAE,CAAC,CAAC;YACnE,MAAM,EAAE,CAAC;QACX,CAAC,CAAC,CAAC;IACP,CAAC;SAAM,CAAC;QACN,IAAI,CAAC;YACH,OAAO,CAAC,GAAG,CAAC,WAAW,IAAI,EAAE,CAAC,CAAC;YAC/B,MAAM,EAAE,CAAC;QACX,CAAC;QAAC,OAAO,GAAY,EAAE,CAAC;YACtB,OAAO,CAAC,KAAK,CAAC,WAAW,IAAI,KAAM,GAAa,CAAC,OAAO,IAAI,GAAG,EAAE,CAAC,CAAC;YACnE,MAAM,EAAE,CAAC;QACX,CAAC;IACH,CAAC;AACH,CAAC;AAED,KAAK,UAAU,SAAS,CAAC,IAAY,EAAE,EAAuB;IAC5D,IAAI,CAAC;QACH,MAAM,EAAE,EAAE,CAAC;QACX,OAAO,CAAC,GAAG,CAAC,WAAW,IAAI,EAAE,CAAC,CAAC;QAC/B,MAAM,EAAE,CAAC;IACX,CAAC;IAAC,OAAO,GAAY,EAAE,CAAC;QACtB,OAAO,CAAC,KAAK,CAAC,WAAW,IAAI,KAAM,GAAa,CAAC,OAAO,IAAI,GAAG,EAAE,CAAC,CAAC;QACnE,MAAM,EAAE,CAAC;IACX,CAAC;AACH,CAAC;AAED,8EAA8E;AAC9E,0CAA0C;AAC1C,8EAA8E;AAE9E,IAAI,CAAC,qCAAqC,EAAE,GAAG,EAAE;IAC/C,MAAM,KAAK,GAAqB;QAC9B,SAAS,EAAE,EAAE;QACb,WAAW,EAAE,IAAI;QACjB,SAAS,EAAE,EAAE;QACb,mBAAmB,EAAE,IAAI;KAC1B,CAAC;IACF,MAAM,MAAM,GAAG,wBAAwB,CAAC,KAAK,EAAE,EAAE,CAAC,CAAC;IACnD,MAAM,CAAC,WAAW,CAAC,MAAM,EAAE,WAAW,EAAE,6BAA6B,MAAO,EAAE,CAAC,CAAC;AAClF,CAAC,CAAC,CAAC;AAEH,8EAA8E;AAC9E,yDAAyD;AACzD,8EAA8E;AAE9E,IAAI,CAAC,kDAAkD,EAAE,GAAG,EAAE;IAC5D,MAAM,IAAI,GAAG,QAAQ,CAAC,YAAY,EAAE,EAAE,KAAK,EAAE,MAAM,EAAE,CAAC,CAAC;IACvD,MAAM,KAAK,GAAqB;QAC9B,SAAS,EAAE,CAAC;QACZ,WAAW,EAAE,IAAI;QACjB,SAAS,EAAE;YACT,EAAE,IAAI,EAAE,YAAY,EAAE,QAAQ,EAAE,IAAI,EAAE,IAAI,EAAE,CAAC,EAAE;YAC/C,EAAE,IAAI,EAAE,YAAY,EAAE,QAAQ,EAAE,IAAI,EAAE,IAAI,EAAE,CAAC,EAAE;YAC/C,EAAE,IAAI,EAAE,YAAY,EAAE,QAAQ,EAAE,IAAI,EAAE,IAAI,EAAE,CAAC,EAAE;SAChD;QACD,mBAAmB,EAAE,IAAI;KAC1B,CAAC;IACF,MAAM,MAAM,GAAG,wBAAwB,CAAC,KAAK,EAAE,EAAE,CAAC,CAAC;IACnD,MAAM,CAAC,WAAW,CAAC,MAAM,EAAE,MAAM,EAAE,wBAAwB,MAAO,EAAE,CAAC,CAAC;AACxE,CAAC,CAAC,CAAC;AAEH,8EAA8E;AAC9E,6CAA6C;AAC7C,8EAA8E;AAE9E,IAAI,CAAC,iCAAiC,EAAE,GAAG,EAAE;IAC3C,MAAM,KAAK,GAAqB;QAC9B,SAAS,EAAE,CAAC;QACZ,WAAW,EAAE,wBAAwB;QACrC,SAAS,EAAE,EAAE;QACb,mBAAmB,EAAE,IAAI;KAC1B,CAAC;IACF,MAAM,MAAM,GAAG,wBAAwB,CAAC,KAAK,EAAE,EAAE,CAAC,CAAC;IACnD,MAAM,CAAC,WAAW,CAAC,MAAM,EAAE,gBAAgB,EAAE,kCAAkC,MAAO,EAAE,CAAC,CAAC;AAC5F,CAAC,CAAC,CAAC;AAEH,8EAA8E;AAC9E,oFAAoF;AACpF,8EAA8E;AAE9E,MAAM,SAAS,CAAC,gEAAgE,EAAE,KAAK,IAAI,EAAE;IAC3F,MAAM,QAAQ,GAAG;QACf,EAAE,IAAI,EAAE,MAAM,EAAE,OAAO,EAAE,cAAc,EAAE;QACzC;YACE,IAAI,EAAE,WAAW;YACjB,OAAO,EAAE,sEAAsE;SAChF;QACD,EAAE,IAAI,EAAE,MAAM,EAAE,OAAO,EAAE,iBAAiB,EAAE;QAC5C,EAAE,IAAI,EAAE,WAAW,EAAE,OAAO,EAAE,qCAAqC,EAAE;KACtE,CAAC;IAEF,uEAAuE;IACvE,MAAM,SAAS,GAAG,KAAK,IAAI,EAAE,CAAC,kCAAkC,CAAC;IAEjE,MAAM,MAAM,GAAG,MAAM,WAAW,CAAC,QAA8D,EAAE,SAAS,EAAE,WAAW,CAAC,CAAC;IAEzH,MAAM,CAAC,WAAW,CAAC,MAAM,CAAC,MAAM,EAAE,GAAG,EAAE,qBAAqB,MAAM,CAAC,MAAO,EAAE,CAAC,CAAC;IAC9E,MAAM,CAAC,WAAW,CAAC,MAAM,CAAC,YAAY,EAAE,IAAI,EAAE,6CAA6C,CAAC,CAAC;AAC/F,CAAC,CAAC,CAAC;AAEH,8EAA8E;AAC9E,+EAA+E;AAC/E,8EAA8E;AAE9E,MAAM,SAAS,CAAC,mEAAmE,EAAE,KAAK,IAAI,EAAE;IAC9F,MAAM,QAAQ,GAAG;QACf,EAAE,IAAI,EAAE,MAAM,EAAE,OAAO,EAAE,gCAAgC,EAAE;QAC3D,EAAE,IAAI,EAAE,WAAW,EAAE,OAAO,EAAE,4BAA4B,EAAE;KAC7D,CAAC;IAEF,iDAAiD;IACjD,MAAM,SAAS,GAAG,KAAK,IAAI,EAAE,CAAC,kEAAkE,CAAC;IAEjG,MAAM,MAAM,GAAG,MAAM,WAAW,CAAC,QAA8D,EAAE,SAAS,EAAE,WAAW,CAAC,CAAC;IAEzH,MAAM,CAAC,WAAW,CAAC,MAAM,CAAC,MAAM,EAAE,OAAO,EAAE,yBAAyB,MAAM,CAAC,MAAO,EAAE,CAAC,CAAC;IACtF,MAAM,CAAC,WAAW,CAAC,MAAM,CAAC,YAAY,EAAE,KAAK,EAAE,oDAAoD,CAAC,CAAC;IACrG,MAAM,CAAC,WAAW,CAAC,MAAM,CAAC,WAAW,EAAE,WAAW,CAAC,CAAC;AACtD,CAAC,CAAC,CAAC;AAEH,8EAA8E;AAC9E,wEAAwE;AACxE,8EAA8E;AAE9E,MAAM,SAAS,CAAC,uDAAuD,EAAE,KAAK,IAAI,EAAE;IAClF,MAAM,QAAQ,GAAG;QACf,EAAE,IAAI,EAAE,MAAM,EAAE,OAAO,EAAE,+BAA+B,EAAE;QAC1D,EAAE,IAAI,EAAE,WAAW,EAAE,OAAO,EAAE,wDAAwD,EAAE;KACzF,CAAC;IAEF,mCAAmC;IACnC,MAAM,SAAS,GAAG,KAAK,IAAI,EAAE,CAAC,iDAAiD,CAAC;IAEhF,MAAM,MAAM,GAAG,MAAM,WAAW,CAAC,QAA8D,EAAE,SAAS,EAAE,MAAM,CAAC,CAAC;IAEpH,MAAM,CAAC,WAAW,CAAC,MAAM,CAAC,MAAM,EAAE,IAAI,EAAE,sBAAsB,MAAM,CAAC,MAAO,EAAE,CAAC,CAAC;IAChF,MAAM,CAAC,WAAW,CAAC,MAAM,CAAC,WAAW,EAAE,MAAM,CAAC,CAAC;AACjD,CAAC,CAAC,CAAC;AAEH,8EAA8E;AAC9E,qDAAqD;AACrD,8EAA8E;AAE9E,IAAI,CAAC,+BAA+B,EAAE,GAAG,EAAE;IACzC,MAAM,EAAE,GAAG,QAAQ,CAAC,YAAY,EAAE,EAAE,KAAK,EAAE,aAAa,EAAE,KAAK,EAAE,CAAC,EAAE,CAAC,CAAC;IACtE,MAAM,EAAE,GAAG,QAAQ,CAAC,YAAY,EAAE,EAAE,KAAK,EAAE,aAAa,EAAE,KAAK,EAAE,CAAC,EAAE,CAAC,CAAC;IACtE,MAAM,CAAC,WAAW,CAAC,EAAE,EAAE,EAAE,EAAE,yCAAyC,CAAC,CAAC;IACtE,MAAM,CAAC,EAAE,CAAC,EAAE,CAAC,MAAM,KAAK,EAAE,EAAE,oCAAoC,EAAE,CAAC,MAAM,EAAE,CAAC,CAAC;AAC/E,CAAC,CAAC,CAAC;AAEH,8EAA8E;AAC9E,+DAA+D;AAC/D,8EAA8E;AAE9E,IAAI,CAAC,2CAA2C,EAAE,GAAG,EAAE;IACrD,MAAM,EAAE,GAAG,QAAQ,CAAC,YAAY,EAAE,EAAE,KAAK,EAAE,OAAO,EAAE,CAAC,CAAC;IACtD,MAAM,EAAE,GAAG,QAAQ,CAAC,YAAY,EAAE,EAAE,KAAK,EAAE,OAAO,EAAE,CAAC,CAAC;IACtD,MAAM,CAAC,cAAc,CAAC,EAAE,EAAE,EAAE,EAAE,4CAA4C,CAAC,CAAC;IAE5E,MAAM,EAAE,GAAG,QAAQ,CAAC,YAAY,EAAE,EAAE,KAAK,EAAE,OAAO,EAAE,CAAC,CAAC;IACtD,MAAM,CAAC,cAAc,CAAC,EAAE,EAAE,EAAE,EAAE,kDAAkD,CAAC,CAAC;AACpF,CAAC,CAAC,CAAC;AAEH,8EAA8E;AAC9E,oEAAoE;AACpE,8EAA8E;AAE9E,IAAI,CAAC,qDAAqD,EAAE,GAAG,EAAE;IAC/D,MAAM,KAAK,GAAG,sBAAsB,EAAE,CAAC;IAEvC,MAAM,CAAC,WAAW,CAAC,KAAK,CAAC,SAAS,EAAE,CAAC,CAAC,CAAC;IACvC,MAAM,CAAC,WAAW,CAAC,KAAK,CAAC,WAAW,EAAE,CAAC,CAAC,CAAC;IACzC,MAAM,CAAC,WAAW,CAAC,KAAK,CAAC,SAAS,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC;IAE9C,UAAU,CAAC,KAAK,EAAE,IAAI,EAAE;QACtB,EAAE,IAAI,EAAE,YAAY,EAAE,IAAI,EAAE,EAAE,KAAK,EAAE,MAAM,EAAE,EAAE;QAC/C,EAAE,IAAI,EAAE,YAAY,EAAE,IAAI,EAAE,EAAE,IAAI,EAAE,KAAK,EAAE,EAAE;KAC9C,CAAC,CAAC;IAEH,MAAM,CAAC,WAAW,CAAC,KAAK,CAAC,SAAS,EAAE,CAAC,EAAE,6BAA6B,KAAK,CAAC,SAAS,EAAE,CAAC,CAAC;IACvF,MAAM,CAAC,WAAW,CAAC,KAAK,CAAC,WAAW,EAAE,IAAI,EAAE,kCAAkC,KAAK,CAAC,WAAW,EAAE,CAAC,CAAC;IACnG,MAAM,CAAC,WAAW,CAAC,KAAK,CAAC,SAAS,CAAC,MAAM,EAAE,CAAC,EAAE,uCAAuC,KAAK,CAAC,SAAS,CAAC,MAAM,EAAE,CAAC,CAAC;IAC/G,MAAM,CAAC,WAAW,CAAC,KAAK,CAAC,SAAS,CAAC,CAAC,CAAC,CAAC,IAAI,EAAE,YAAY,CAAC,CAAC;IAC1D,MAAM,CAAC,WAAW,CAAC,KAAK,CAAC,SAAS,CAAC,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC;IAE/C,UAAU,CAAC,KAAK,EAAE,IAAI,EAAE,EAAE,CAAC,CAAC;IAE5B,MAAM,CAAC,WAAW,CAAC,KAAK,CAAC,SAAS,EAAE,CAAC,CAAC,CAAC;IACvC,MAAM,CAAC,WAAW,CAAC,KAAK,CAAC,WAAW,EAAE,IAAI,CAAC,CAAC;AAC9C,CAAC,CAAC,CAAC;AAEH,8EAA8E;AAC9E,mFAAmF;AACnF,8EAA8E;AAE9E,IAAI,CAAC,mEAAmE,EAAE,GAAG,EAAE;IAC7E,MAAM,IAAI,GAAG,QAAQ,CAAC,YAAY,EAAE,EAAE,KAAK,EAAE,MAAM,EAAE,CAAC,CAAC;IAEvD,yEAAyE;IACzE,MAAM,QAAQ,GAAG,KAAK,CAAC,IAAI,CAAC,EAAE,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC;QACpD,IAAI,EAAE,YAAY,EAAE,QAAQ,EAAE,IAAI,EAAE,IAAI,EAAE,CAAC,GAAG,CAAC;KAChD,CAAC,CAAC,CAAC;IACJ,MAAM,WAAW,GAAG,KAAK,CAAC,IAAI,CAAC,EAAE,MAAM,EAAE,gBAAgB,EAAE,EAAE,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC;QACtE,IAAI,EAAE,QAAQ,CAAC,EAAE,EAAE,QAAQ,EAAE,QAAQ,CAAC,QAAQ,CAAC,EAAE,EAAE,EAAE,CAAC,EAAE,IAAI,EAAE,CAAC,GAAG,CAAC;KACpE,CAAC,CAAC,CAAC;IAEJ,MAAM,KAAK,GAAqB;QAC9B,SAAS,EAAE,CAAC;QACZ,WAAW,EAAE,IAAI;QACjB,SAAS,EAAE,CAAC,GAAG,QAAQ,EAAE,GAAG,WAAW,CAAC;QACxC,mBAAmB,EAAE,IAAI;KAC1B,CAAC;IAEF,MAAM,MAAM,GAAG,wBAAwB,CAAC,KAAK,EAAE,EAAE,CAAC,CAAC;IACnD,MAAM,CAAC,WAAW,CAAC,MAAM,EAAE,IAAI,EAC7B,iDAAiD,MAAO,EAAE,CAAC,CAAC;AAChE,CAAC,CAAC,CAAC;AAEH,8EAA8E;AAC9E,+CAA+C;AAC/C,8EAA8E;AAE9E,IAAI,CAAC,gFAAgF,EAAE,GAAG,EAAE;IAC1F,MAAM,IAAI,GAAG,6DAA6D,CAAC;IAC3E,MAAM,MAAM,GAAG,0BAA0B,CAAC,IAAI,CAAC,CAAC;IAChD,MAAM,CAAC,WAAW,CAAC,MAAM,EAAE,IAAI,EAAE,sBAAsB,MAAO,EAAE,CAAC,CAAC;AACpE,CAAC,CAAC,CAAC;AAEH,IAAI,CAAC,8DAA8D,EAAE,GAAG,EAAE;IACxE,MAAM,QAAQ,GAAG;QACf,EAAE,IAAI,EAAE,MAAM,EAAE,OAAO,EAAE,UAAU,EAAE;QACrC,EAAE,IAAI,EAAE,WAAW,EAAE,OAAO,EAAE,kCAAkC,EAAE;QAClE,EAAE,IAAI,EAAE,MAAM,EAAE,OAAO,EAAE,cAAc,EAAE;QACzC,EAAE,IAAI,EAAE,WAAW,EAAE,OAAO,EAAE,2CAA2C,EAAE;KAC5E,CAAC;IACF,yEAAyE;IACzE,MAAM,MAAM,GAAG,wBAAwB,CAAC,QAA8D,CAAC,CAAC;IACxG,MAAM,CAAC,WAAW,CAAC,MAAM,EAAE,SAAS,EAAE,yCAAyC,MAAO,EAAE,CAAC,CAAC;AAC5F,CAAC,CAAC,CAAC;AAEH,8EAA8E;AAC9E,UAAU;AACV,8EAA8E;AAE9E,yDAAyD;AACzD,UAAU,CAAC,GAAG,EAAE;IACd,MAAM,KAAK,GAAG,MAAM,GAAG,MAAM,CAAC;IAC9B,OAAO,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC;IAChB,OAAO,CAAC,GAAG,CAAC,2CAA2C,CAAC,CAAC;IACzD,OAAO,CAAC,GAAG,CAAC,eAAe,MAAM,IAAI,KAAK,EAAE,CAAC,CAAC;IAC9C,OAAO,CAAC,GAAG,CAAC,eAAe,MAAM,KAAK,CAAC,CAAC,CAAC,CAAC,YAAY,CAAC,CAAC,CAAC,GAAG,MAAM,SAAS,EAAE,CAAC,CAAC;IAC/E,IAAI,MAAM,GAAG,CAAC,EAAE,CAAC;QACf,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;IAClB,CAAC;AACH,CAAC,EAAE,GAAG,CAAC,CAAC"}
|
|
@@ -0,0 +1,123 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* GAIA Adversarial Critic — ADR-135 Track D
|
|
3
|
+
*
|
|
4
|
+
* After the main agent produces a candidate answer, a Sonnet pass reviews it.
|
|
5
|
+
* If the critic verdict is "fail", the orchestrator re-runs the agent with the
|
|
6
|
+
* critique injected as additional context.
|
|
7
|
+
*
|
|
8
|
+
* Motivation (iter 29 finding): tool quality is the bottleneck on L1 (~20.8%).
|
|
9
|
+
* The critic catches wrong-because-of-bad-tool-result answers BEFORE submission,
|
|
10
|
+
* without requiring better search backends. Expected L1 lift: +3-5pp.
|
|
11
|
+
*
|
|
12
|
+
* Design:
|
|
13
|
+
* - NEW file only; NOT wired into gaia-bench.ts yet to avoid merge conflicts
|
|
14
|
+
* with in-flight iter 29/31/34 branches. Wiring is a small follow-up PR.
|
|
15
|
+
* - `criticReview()` — single Sonnet call, returns structured verdict.
|
|
16
|
+
* - `runGaiaAgentWithCritic()` — orchestration wrapper: runs agent, calls
|
|
17
|
+
* critic, retries once on "fail". "uncertain" is treated as "pass" (don't
|
|
18
|
+
* burn retries on borderline cases).
|
|
19
|
+
* - API errors and malformed JSON from the critic are caught; original
|
|
20
|
+
* candidate is returned with an error-flagged verdict.
|
|
21
|
+
* - Default opt-in: enableCritic=false. Set true via RunWithCriticOptions.
|
|
22
|
+
*
|
|
23
|
+
* Cost discipline:
|
|
24
|
+
* - Critic uses claude-sonnet-4-6 (separate from the agent's default Haiku).
|
|
25
|
+
* - One critic call + one optional retry = max 2 extra Sonnet calls per Q.
|
|
26
|
+
* - Approximate extra cost per question: ~$0.003-0.005 (well within budget).
|
|
27
|
+
*
|
|
28
|
+
* Plugin sync TODO (follow-up PR after gaia-bench wiring):
|
|
29
|
+
* - Update plugins/ruflo-workflows/commands/gaia-run.md with --enable-critic flag.
|
|
30
|
+
* - Update plugins/ruflo-workflows/skills/gaia-debugging/SKILL.md: add critic
|
|
31
|
+
* as a recommended diagnostic step for wrong-answer analysis.
|
|
32
|
+
*
|
|
33
|
+
* Refs: ADR-135, ADR-133, iter 29 finding, #2156
|
|
34
|
+
*/
|
|
35
|
+
import { GaiaQuestion } from './gaia-loader.js';
|
|
36
|
+
import { GaiaAgentResult, GaiaAgentOptions } from './gaia-agent.js';
|
|
37
|
+
/**
|
|
38
|
+
* Structured verdict returned by the adversarial critic.
|
|
39
|
+
*
|
|
40
|
+
* - "pass" → critic agrees with the candidate; no retry needed.
|
|
41
|
+
* - "fail" → critic found a flaw; suggestedRevision provided if possible.
|
|
42
|
+
* - "uncertain" → critic is unsure; treated as "pass" to avoid burning retries.
|
|
43
|
+
*/
|
|
44
|
+
export type CriticVerdictType = 'pass' | 'fail' | 'uncertain';
|
|
45
|
+
export interface CriticVerdict {
|
|
46
|
+
/** Critic's assessment of the candidate answer. */
|
|
47
|
+
verdict: CriticVerdictType;
|
|
48
|
+
/** Short explanation of the reasoning (1-3 sentences). */
|
|
49
|
+
reasoning: string;
|
|
50
|
+
/** Suggested correction when verdict is "fail"; may be empty string. */
|
|
51
|
+
suggestedRevision?: string;
|
|
52
|
+
/** Estimated USD cost of this critic call. */
|
|
53
|
+
costUsd: number;
|
|
54
|
+
/** True when the critic call itself failed (API error, malformed JSON). */
|
|
55
|
+
error?: boolean;
|
|
56
|
+
/** Original raw response text when parse failed. */
|
|
57
|
+
rawResponse?: string;
|
|
58
|
+
}
|
|
59
|
+
export interface CriticOptions {
|
|
60
|
+
/**
|
|
61
|
+
* Model to use for the critic pass.
|
|
62
|
+
* Defaults to 'claude-sonnet-4-6' (intentionally separate from agent model).
|
|
63
|
+
*/
|
|
64
|
+
model?: string;
|
|
65
|
+
/**
|
|
66
|
+
* Maximum number of agent retries on "fail" verdict.
|
|
67
|
+
* Default: 1 (one retry). Set to 0 to disable retries (observe-only mode).
|
|
68
|
+
*/
|
|
69
|
+
maxRetries?: number;
|
|
70
|
+
/** Anthropic API key (resolved via env/gcloud if omitted). */
|
|
71
|
+
apiKey?: string;
|
|
72
|
+
}
|
|
73
|
+
export interface GaiaAgentResultWithCritic extends GaiaAgentResult {
|
|
74
|
+
/** All critic verdicts collected (one per agent attempt). */
|
|
75
|
+
criticVerdicts: CriticVerdict[];
|
|
76
|
+
/** How many retries were actually attempted (0 if critic passed first time). */
|
|
77
|
+
retriesAttempted: number;
|
|
78
|
+
}
|
|
79
|
+
/** Options for runGaiaAgentWithCritic, extending core agent options. */
|
|
80
|
+
export interface RunWithCriticOptions extends GaiaAgentOptions {
|
|
81
|
+
/**
|
|
82
|
+
* Enable the adversarial critic pass. Default: false.
|
|
83
|
+
* When false, runGaiaAgentWithCritic behaves identically to runGaiaAgent.
|
|
84
|
+
*/
|
|
85
|
+
enableCritic?: boolean;
|
|
86
|
+
/** Critic-specific configuration. */
|
|
87
|
+
criticOptions?: CriticOptions;
|
|
88
|
+
}
|
|
89
|
+
/**
|
|
90
|
+
* Run the adversarial critic against a candidate answer.
|
|
91
|
+
*
|
|
92
|
+
* @param question - The GAIA question being evaluated.
|
|
93
|
+
* @param candidateAnswer - The agent's proposed final answer.
|
|
94
|
+
* @param trajectory - Lightweight trajectory summary from the agent run.
|
|
95
|
+
* @param options - Critic configuration (model, apiKey).
|
|
96
|
+
* @returns CriticVerdict with verdict, reasoning, cost.
|
|
97
|
+
*/
|
|
98
|
+
export declare function criticReview(question: GaiaQuestion, candidateAnswer: string, trajectory: {
|
|
99
|
+
steps: Array<{
|
|
100
|
+
tool?: string;
|
|
101
|
+
result?: string;
|
|
102
|
+
reasoning?: string;
|
|
103
|
+
}>;
|
|
104
|
+
turns: number;
|
|
105
|
+
}, options?: CriticOptions): Promise<CriticVerdict>;
|
|
106
|
+
/**
|
|
107
|
+
* Run the GAIA agent with an optional adversarial critic pass.
|
|
108
|
+
*
|
|
109
|
+
* When enableCritic=false (default), this is a thin pass-through to
|
|
110
|
+
* runGaiaAgent with an empty criticVerdicts array.
|
|
111
|
+
*
|
|
112
|
+
* When enableCritic=true:
|
|
113
|
+
* 1. Run runGaiaAgent normally.
|
|
114
|
+
* 2. If the agent produced a finalAnswer, call criticReview.
|
|
115
|
+
* 3. If verdict is "fail" and retriesAttempted < maxRetries:
|
|
116
|
+
* a. Re-run the agent with the critique injected as additional context.
|
|
117
|
+
* b. Call criticReview again on the new answer.
|
|
118
|
+
* 4. Return the final result with all critic verdicts attached.
|
|
119
|
+
*
|
|
120
|
+
* Note on "uncertain": treated as "pass" (no retry triggered).
|
|
121
|
+
*/
|
|
122
|
+
export declare function runGaiaAgentWithCritic(question: GaiaQuestion, options?: RunWithCriticOptions): Promise<GaiaAgentResultWithCritic>;
|
|
123
|
+
//# sourceMappingURL=gaia-critic.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"gaia-critic.d.ts","sourceRoot":"","sources":["../../../src/benchmarks/gaia-critic.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GAiCG;AAGH,OAAO,EACL,YAAY,EACb,MAAM,kBAAkB,CAAC;AAC1B,OAAO,EAEL,eAAe,EACf,gBAAgB,EACjB,MAAM,iBAAiB,CAAC;AAwBzB;;;;;;GAMG;AACH,MAAM,MAAM,iBAAiB,GAAG,MAAM,GAAG,MAAM,GAAG,WAAW,CAAC;AAE9D,MAAM,WAAW,aAAa;IAC5B,mDAAmD;IACnD,OAAO,EAAE,iBAAiB,CAAC;IAC3B,0DAA0D;IAC1D,SAAS,EAAE,MAAM,CAAC;IAClB,wEAAwE;IACxE,iBAAiB,CAAC,EAAE,MAAM,CAAC;IAC3B,8CAA8C;IAC9C,OAAO,EAAE,MAAM,CAAC;IAChB,2EAA2E;IAC3E,KAAK,CAAC,EAAE,OAAO,CAAC;IAChB,oDAAoD;IACpD,WAAW,CAAC,EAAE,MAAM,CAAC;CACtB;AAED,MAAM,WAAW,aAAa;IAC5B;;;OAGG;IACH,KAAK,CAAC,EAAE,MAAM,CAAC;IACf;;;OAGG;IACH,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,8DAA8D;IAC9D,MAAM,CAAC,EAAE,MAAM,CAAC;CACjB;AAGD,MAAM,WAAW,yBAA0B,SAAQ,eAAe;IAChE,6DAA6D;IAC7D,cAAc,EAAE,aAAa,EAAE,CAAC;IAChC,gFAAgF;IAChF,gBAAgB,EAAE,MAAM,CAAC;CAC1B;AAED,wEAAwE;AACxE,MAAM,WAAW,oBAAqB,SAAQ,gBAAgB;IAC5D;;;OAGG;IACH,YAAY,CAAC,EAAE,OAAO,CAAC;IACvB,qCAAqC;IACrC,aAAa,CAAC,EAAE,aAAa,CAAC;CAC/B;AAqID;;;;;;;;GAQG;AACH,wBAAsB,YAAY,CAChC,QAAQ,EAAE,YAAY,EACtB,eAAe,EAAE,MAAM,EACvB,UAAU,EAAE;IAAE,KAAK,EAAE,KAAK,CAAC;QAAE,IAAI,CAAC,EAAE,MAAM,CAAC;QAAC,MAAM,CAAC,EAAE,MAAM,CAAC;QAAC,SAAS,CAAC,EAAE,MAAM,CAAA;KAAE,CAAC,CAAC;IAAC,KAAK,EAAE,MAAM,CAAA;CAAE,EACnG,OAAO,CAAC,EAAE,aAAa,GACtB,OAAO,CAAC,aAAa,CAAC,CAyExB;AAMD;;;;;;;;;;;;;;;GAeG;AACH,wBAAsB,sBAAsB,CAC1C,QAAQ,EAAE,YAAY,EACtB,OAAO,GAAE,oBAAyB,GACjC,OAAO,CAAC,yBAAyB,CAAC,CA4EpC"}
|