Garry's Opinionated OpenClaw/Hermes Agent Brain
TypeScript
/**
* N2-2 — bare `gbrain find-contradictions` reads the latest run.
*
* skills/correction-pipeline/SKILL.md documents the bare command as reading
* the latest probe run. makeContext always populates ctx.sourceId (falling
* back to 'default'), so the op's source-filter refusal fired for every local
* CLI call. Driven through the real CLI seam (makeContext) so the implicit
* source marker and the op handler are tested together; user-selected
* sources (--source, GBRAIN_SOURCE) get the scoped view (D9): each side is
* re-read in that source, and this fixture's chunk ids name no row there.
*/
import { afterAll, beforeAll, beforeEach, describe, expect, test } from 'bun:test';
import { makeContext } from '../src/cli/main.ts';
import { PGLiteEngine } from '../src/core/pglite-engine.ts';
import { operationsByName } from '../src/core/operations.ts';
import { resetPgliteState } from './helpers/reset-pglite.ts';
import { withEnv } from './helpers/with-env.ts';
let engine: PGLiteEngine;
beforeAll(async () => {
engine = new PGLiteEngine();
await engine.connect({ database_url: '' });
await engine.initSchema();
}, 60_000);
afterAll(async () => {
await engine.disconnect();
});
beforeEach(async () => {
await resetPgliteState(engine);
await engine.executeRaw(`INSERT INTO sources (id, name) VALUES ('alpha', 'alpha')`);
await engine.writeContradictionsRun({
run_id: 'cli-run-1', judge_model: 'test', prompt_version: 'v1',
queries_evaluated: 1, queries_with_contradiction: 1, total_contradictions_flagged: 1,
wilson_ci_lower: 0, wilson_ci_upper: 1, judge_errors_total: 0,
cost_usd_total: 0, duration_ms: 1, source_tier_breakdown: {},
report_json: {
per_query: [{
contradictions: [{
kind: 'cross_slug_chunks', severity: 'medium', axis: 'headcount', confidence: 0.9,
a: { slug: 'notes/a', chunk_id: 1, take_id: null },
b: { slug: 'notes/b', chunk_id: 2, take_id: null },
resolution_kind: 'manual_review', resolution_command: '# manual review',
}],
}],
},
});
});
const findContradictions = (params: Record<string, unknown>) =>
withEnv({ GBRAIN_SOURCE: undefined }, async () => {
const ctx = await makeContext(engine, params);
return operationsByName['find_contradictions'].handler(ctx, {}) as Promise<{
run_id?: string; ran_at?: unknown; contradictions: unknown[]; note?: string; total_visible?: number;
}>;
});
describe('find-contradictions through the local CLI context', () => {
test('bare command (implicit default source) reads the latest run', async () => {
const res = await findContradictions({});
expect(res.note).toBeUndefined();
expect(res.run_id).toBe('cli-run-1');
expect(res.contradictions).toHaveLength(1);
});
test('--source <id> is a filter: the scoped view, without the run total', async () => {
const res = await findContradictions({ source: 'alpha' });
expect(res).toEqual({ run_id: 'cli-run-1', ran_at: expect.anything(), contradictions: [], total_visible: 0 });
});
test('--source __all__ still spans the brain', async () => {
const res = await findContradictions({ source: '__all__' });
expect(res.run_id).toBe('cli-run-1');
});
test('GBRAIN_SOURCE is user-selected scope: the scoped view', async () => {
const res = await withEnv({ GBRAIN_SOURCE: 'alpha' }, async () => {
const ctx = await makeContext(engine, {});
return operationsByName['find_contradictions'].handler(ctx, {});
});
expect(res).toEqual({ run_id: 'cli-run-1', ran_at: expect.anything(), contradictions: [], total_visible: 0 });
});
});