query

name	query
description	GBrain-inspired local-brain query for Codex. Use when answering questions from the repo-local brain corpus with explicit file grounding.

Query

Use this skill when the user wants an answer grounded in the local brain/ corpus.

This port is adapted from garrytan/gbrain at commit 0b47afbf402a4e27a648bb9d131ce584461461ea.

Important Adaptation

Upstream GBrain uses hybrid database-backed search. In this repo, the local brain substrate is file-based, so the deterministic search path is:

1. python3 scripts/brain_search.py "<query>"
2. read the top matching files
3. answer from those files with explicit path grounding

Workflow

1. Search the local brain first.
2. Read the top relevant brain pages.
3. Synthesize only from what those pages actually say.
4. If nothing relevant exists, say the local brain does not yet contain the answer.

Current Upstream Coverage

Use source precedence explicitly: primary/raw sources, structured brain pages, then synthesized concept pages. Traverse backlinks and related entities when direct keyword hits are incomplete, state search-quality gaps, and stay within the user's requested detail budget.

Guardrails

• Do not answer from general memory when the user asked for brain-grounded context.
• Make missing coverage explicit.
• Cite the file paths you relied on.