Garry's Opinionated OpenClaw/Hermes Agent Brain
TypeScript
/**
* #5749: a bare frontmatter wikilink with a unique global-basename match must
* resolve to that page, not to whatever page the live-mode keyword search
* ranks first. The managed stale-link sweep runs the live resolver over every
* stale page, so before the fix a transcript that repeated the term silently
* took over correct `related:` edges.
*
* Precedence mirrors the entity resolver (#5769): an exact name match (here, a
* page whose slug basename IS the link text) wins before fuzzy or search
* evidence about some other page.
*
* PGLite in-memory, no embeddings ($0).
*/
import { describe, test, expect, beforeAll, afterAll } from 'bun:test';
import { PGLiteEngine } from '../src/core/pglite-engine.ts';
import { importFromContent } from '../src/core/import-file.ts';
import { extractPageLinks, makeResolver } from '../src/core/link-extraction.ts';
import { withEnv } from './helpers/with-env.ts';
// A brain that boosts transcripts (GBRAIN_SOURCE_BOOST) lets a transcript that
// repeats a term clear the live keyword step's 0.8 score bar, as in the report.
const BOOSTED = { GBRAIN_SOURCE_BOOST: 'conversations/:2.5' };
let engine: PGLiteEngine;
const TRANSCRIPT = 'conversations/sessions/2026-09-01-planning-chat';
const PRODUCT = 'products/zorblax';
beforeAll(async () => {
engine = new PGLiteEngine();
await engine.connect({});
await engine.initSchema();
await importFromContent(engine, PRODUCT,
'---\ntype: project\ntitle: Acme Gizmo Line\n---\n\nThe gizmo product line.\n', { noEmbed: true });
await importFromContent(engine, 'analysis/quibbleton',
'---\ntype: concept\ntitle: Market Sizing Notes\n---\n\nSizing work.\n', { noEmbed: true });
const chatter = Array.from({ length: 60 }, () => 'zorblax zorblax quibbleton quibbleton zorblax quibbleton.').join('\n');
await importFromContent(engine, TRANSCRIPT,
`---\ntype: conversation\ntitle: Zorblax and quibbleton planning chat\n---\n\n${chatter}\n`, { noEmbed: true });
}, 60_000);
afterAll(async () => {
if (engine) await engine.disconnect();
}, 60_000);
async function relatedTargets(globalBasename: boolean, mode: 'live' | 'batch'): Promise<string[]> {
const resolver = makeResolver(engine, { mode, sourceId: 'default' });
const { candidates } = await withEnv(BOOSTED, () => extractPageLinks(
'wiki/overview', 'Body.', { related: ['[[zorblax]]', '[[quibbleton]]'] }, 'concept', resolver, { globalBasename },
));
return candidates.filter(c => c.linkType === 'related_to').map(c => c.targetSlug).sort();
}
describe('#5749 frontmatter wikilink resolution order', () => {
test('the live keyword search can reach the transcript (precondition)', async () => {
const hits = await withEnv(BOOSTED, () => engine.searchKeyword('zorblax', { limit: 3, sourceId: 'default' }));
expect(hits[0]?.slug).toBe(TRANSCRIPT);
expect(hits[0]!.score).toBeGreaterThanOrEqual(0.8);
});
test('live mode with global_basename: the unique basename wins over keyword search', async () => {
expect(await relatedTargets(true, 'live')).toEqual(['analysis/quibbleton', PRODUCT]);
});
test('live and batch mode agree on the same frontmatter', async () => {
expect(await relatedTargets(true, 'live')).toEqual(await relatedTargets(true, 'batch'));
});
test('an ambiguous basename does not win; the later steps still decide', async () => {
await importFromContent(engine, 'archive/zorblax',
'---\ntype: project\ntitle: Old Gizmo Line\n---\n\nArchived.\n', { noEmbed: true });
try {
const resolver = makeResolver(engine, { mode: 'batch', sourceId: 'default' });
const { candidates } = await extractPageLinks(
'wiki/overview', 'Body.', { related: ['[[zorblax]]'] }, 'concept', resolver, { globalBasename: true },
);
expect(candidates.filter(c => c.linkType === 'related_to')).toEqual([]);
} finally {
await engine.executeRaw(`DELETE FROM pages WHERE slug = 'archive/zorblax'`);
}
});
test('without global_basename the resolver order is unchanged', async () => {
const resolver = makeResolver(engine, { mode: 'batch', sourceId: 'default' });
expect(await resolver.resolve('zorblax')).toBeNull();
});
test('a page never links to itself through its own basename', async () => {
const resolver = makeResolver(engine, { mode: 'live', sourceId: 'default' });
const { candidates } = await withEnv(BOOSTED, () => extractPageLinks(
PRODUCT, 'Body.', { related: ['[[zorblax]]'] }, 'project', resolver, { globalBasename: true },
));
expect(candidates.some(c => c.linkType === 'related_to' && c.targetSlug === PRODUCT)).toBe(false);
});
});