@core / search

1.0.0 ▾

1.0.0

verifiedMIT

GitHub

Site search on Postgres full-text: per-language stemming, prefix matching, safe snippets and typo fallback

Code8 filesContext~511 tokensscan passed

~511 tokens→ src/lib/search/AGENTS.md→ .cursor/rules/genpm-core-search.mdc

This is exactly what your AI reads when it works in src/lib/search. Nothing else is added to its context.

@core/search — rules for AI agents

Purpose

Site search on Postgres full-text search: one search_documents table fed by any SearchSource (@core/contracts), weighted title/body vectors with the stemmer of each document's language, prefix matching for search-as-you-type, snippets returned as plain parts (never HTML) and, if the pg_trgm extension exists, a typo-tolerant fallback on titles. No external engine, no semantic search.

Map

index.ts — public API: search, registerSearchSource, reindex, indexDocument, removeDocument, reindexJob.
search.ts — indexing and querying. schema.ts — table and GIN index.
adapters/hono.ts — searchRoutes(). adapters/next.ts — searchRoute (GET).

Integration

Generate and apply migrations (see src/lib/db/AGENTS.md). Optional typo tolerance: run create extension if not exists pg_trgm; once (available on Neon, Supabase, RDS).
Register sources in the project file src/genpm/search.ts, imported at startup: registerSearchSource(contentSearchSource('pages', { path: ({ slug }) => /${slug} })).
Build the index once and then daily: await reindexJob.enqueue({}) and schedule('search.reindex', '0 3 * * *') (@core/jobs).
Query: const { hits } = await search(q, { locale }), or mount searchRoutes() / searchRoute at /api/search.
Render snippets by mapping parts: hit.snippet.map(p => p.match ? <mark>{p.text}</mark> : p.text).
Verify: index a document and find it by the first letters of a word in its title.

Conventions

Sources yield plain text bodies (strip Markdown/HTML first, e.g. with toPlainText of @core/rich-text).
Pass locale when you know it: the query uses that language's stemmer and the GIN index.
Only index published, public content; private data never goes into search_documents.

Don't

Don't build SQL from the query string; search() already sanitizes it to words.
Don't render snippets with dangerouslySetInnerHTML.
Don't call reindex() inside a request: use the job.

# @core/search — rules for AI agents

## Purpose
Site search on Postgres full-text search: one `search_documents` table fed by any `SearchSource` (@core/contracts),
weighted title/body vectors with the stemmer of each document's language, prefix matching for search-as-you-type,
snippets returned as plain parts (never HTML) and, if the `pg_trgm` extension exists, a typo-tolerant fallback on
titles. No external engine, no semantic search.

## Map
- `index.ts` — public API: `search`, `registerSearchSource`, `reindex`, `indexDocument`, `removeDocument`, `reindexJob`.
- `search.ts` — indexing and querying. `schema.ts` — table and GIN index.
- `adapters/hono.ts` — `searchRoutes()`. `adapters/next.ts` — `searchRoute` (GET).

## Integration
1. Generate and apply migrations (see `src/lib/db/AGENTS.md`). Optional typo tolerance: run
   `create extension if not exists pg_trgm;` once (available on Neon, Supabase, RDS).
2. Register sources in the project file `src/genpm/search.ts`, imported at startup:
   `registerSearchSource(contentSearchSource('pages', { path: ({ slug }) => `/${slug}` }))`.
3. Build the index once and then daily: `await reindexJob.enqueue({})` and `schedule('search.reindex', '0 3 * * *')` (@core/jobs).
4. Query: `const { hits } = await search(q, { locale })`, or mount `searchRoutes()` / `searchRoute` at `/api/search`.
5. Render snippets by mapping parts: `hit.snippet.map(p => p.match ? <mark>{p.text}</mark> : p.text)`.
6. Verify: index a document and find it by the first letters of a word in its title.

## Conventions
- Sources yield plain text bodies (strip Markdown/HTML first, e.g. with `toPlainText` of @core/rich-text).
- Pass `locale` when you know it: the query uses that language's stemmer and the GIN index.
- Only index published, public content; private data never goes into `search_documents`.

## Don't
- Don't build SQL from the query string; `search()` already sanitizes it to words.
- Don't render snippets with `dangerouslySetInnerHTML`.
- Don't call `reindex()` inside a request: use the job.

The exact tree that will be injected, after .genpmignore. Pinned to

src/lib/search/LICENSEread-only · ae66128

MIT License

Copyright (c) 2026 GenPM

Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all
copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
SOFTWARE.