Дмитрий
50 lines
2.0 KiB
JavaScript
50 lines
2.0 KiB
JavaScript
import { describe, it, expect } from 'vitest';
|
|
import { tmpdir } from 'node:os';
|
|
import { mkdtempSync, writeFileSync, rmSync } from 'node:fs';
|
|
import { join } from 'node:path';
|
|
import { extractV4Signals } from './observer-v4-signals.mjs';
|
|
|
|
function write(dir, name, lines) {
|
|
writeFileSync(join(dir, name), lines.map((l) => JSON.stringify(l)).join('\n') + '\n');
|
|
}
|
|
|
|
describe('extractV4Signals', () => {
|
|
it('counts rationalization flags inside the turn window only', () => {
|
|
const dir = mkdtempSync(join(tmpdir(), 'v4-'));
|
|
write(dir, 'rationalization-flags-s.jsonl', [
|
|
{ ts: '2026-05-31T10:00:00.000Z', kind: 'x' }, // before window
|
|
{ ts: '2026-05-31T10:05:00.000Z', kind: 'y' }, // in
|
|
{ ts: '2026-05-31T10:06:00.000Z', kind: 'z' }, // in
|
|
]);
|
|
write(dir, 'llm-judge-verdicts-s.jsonl', [
|
|
{ ts: '2026-05-31T10:05:30.000Z', tool: 'Edit', verdict: 'YES' },
|
|
{ ts: '2026-05-31T10:05:40.000Z', tool: 'Bash', verdict: 'block' },
|
|
]);
|
|
write(dir, 'safe-baseline-actions-s.jsonl', [
|
|
{ ts: '2026-05-31T10:05:10.000Z', tool: 'Edit', action: 'soft_flag' },
|
|
{ ts: '2026-05-31T10:05:50.000Z', tool: 'Write', action: 'hard_block' },
|
|
]);
|
|
writeFileSync(join(dir, 'llm-judge-budget-s.json'), JSON.stringify({ calls: 7 }));
|
|
const sig = extractV4Signals('s', {
|
|
startMs: Date.parse('2026-05-31T10:04:00.000Z'),
|
|
endMs: Date.parse('2026-05-31T10:07:00.000Z'),
|
|
baseDir: dir,
|
|
});
|
|
expect(sig.rationalization_flag_count).toBe(2);
|
|
expect(sig.judge_verdict).toBe('block'); // last in-window verdict
|
|
expect(sig.safe_baseline_action).toBe('hard_block'); // worst in-window action
|
|
expect(sig.judge_calls).toBe(7);
|
|
rmSync(dir, { recursive: true, force: true });
|
|
});
|
|
|
|
it('returns zero/null defaults when files absent', () => {
|
|
const sig = extractV4Signals('nope', { startMs: 0, endMs: 1, baseDir: tmpdir() });
|
|
expect(sig).toEqual({
|
|
rationalization_flag_count: 0,
|
|
judge_verdict: null,
|
|
safe_baseline_action: null,
|
|
judge_calls: 0,
|
|
});
|
|
});
|
|
});
|