corpus-gbrain-query

name: corpus-gbrain-query description: 用于显式配置目标 corpus 后，从该 LoreKit corpus 进行图谱、多跳关系或 GBrain 增强候选检索，并保持 LoreKit 知识库为事实源。

corpus-gbrain-query

Optional read-only GBrain query gateway for a configured Lorekit corpus.

Config

Read ~/.config/lorekit/global-corpus.json.

Required:

• default_corpus
• lorekit_bin or <default_corpus>/bin/lorekit

Optional:

• gbrain_bin

Prefer <lorekit_bin> gbrain query over direct gbrain query, because Lorekit maps candidates back to canonical corpus pages.

Boundary

Lorekit 知识库/ is the only source of truth.

GBrain is a derived candidate recall layer. It must not write 知识库/, 原料/, or 输出/.

Do not run native GBrain mutating commands from this skill.

When To Use

Use when:

• ordinary corpus-query recall is weak
• the user asks for multi-hop relationships
• the question is about graph connections, adjacent concepts, or indirect links
• the user explicitly asks for GBrain

Do not use for:

• first-pass exact keyword search
• writes or fileback
• ingest

Action

1. Read config.
2. cd <default_corpus>.
3. Run <lorekit_bin> gbrain status --json when freshness or setup is unclear.
4. Run <lorekit_bin> gbrain query "<q>" --json.
5. Map every candidate back to canonical 知识库/ pages using Lorekit output.
6. Read the canonical pages before answering.

If the GBrain index is stale, report the stale warning. Do not auto-sync unless the user asks.

Output

Return:

• candidate pages
• relationship or multi-hop explanation
• freshness warning if present
• next step: run corpus-query synthesis or corpus-fileback only after user confirmation