brain-ingest-gate

Validates named-entity resolution and semantic dedup before writing pages into a knowledge base.

Updated Aug 17, 2026
One-click install
npx skills add https://github.com/AMC-JTC/gbrain-1 --skill brain-ingest-gate-amc-jtc
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: brain-ingest-gate
Source: https://github.com/AMC-JTC/gbrain-1/tree/main/plugin/skills/brain-ingest-gate
Command: npx skills add https://github.com/AMC-JTC/gbrain-1 --skill brain-ingest-gate-amc-jtc

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Duplicate pages and unresolved named entities silently pollute a knowledge base when content is copied or migrated in without checks. This Skill enforces a pre-write quality gate so every new page is resolved against existing entities and deduplicated before it lands. ## Core Features & Use Cases - Named-Entity Resolution: Resolves people, companies, projects, and concepts registry-first via gbrain entity, expanding through aliases before concluding anything is new. - Semantic Dedup Decision Tree: Searches for the core claim, reads the top hit, and assigns a clear-dup / plausible-dup / clear band that decides write, link, or skip. - Delegation and Verification: Routes approved writes to enrichment skills (ingest, enrich, concept-synthesis) and verifies results with backlink checks. - Use Case: When asked to "move these project notes into the brain," the gate checks each file for existing entity pages and duplicate insights, links duplicates instead of cloning them, and only writes genuinely new content. ## Quick Start Ask the agent to migrate your notes folder into the brain while checking for duplicates before writing anything.

Frequently Asked Questions about brain-ingest-gate

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I check for duplicates before writing to a knowledge base?

Extract the core claim of the new content in one or two sentences, run gbrain search with a limit of 5, then open and read the top hit with gbrain get. Band the result as clear-dup, plausible-dup, or clear, and link instead of writing when the insight already exists.

How to resolve named entities before creating a new page?

Run gbrain entity with the name first to get the zero-LLM entity card with aliases and near-miss suggestions. On a miss, fall back to gbrain query, expand through all known aliases, and read any plausible candidate page before concluding the entity is new.

Why does vector similarity miss duplicate named-entity pages?

A page stored under its chosen name does not embed close to the generic phrase someone uses to describe it, so the canonical page can score below the prose floor. The gate treats vector scores as a floor for prose only and always reads candidate pages for named things.

Does gbrain import or sync deduplicate content automatically?

No. gbrain import and sync only skip matching frontmatter IDs, and gbrain capture dedupes only identical bytes within 24 hours. Semantic dedup and named-entity resolution are handled by this gate, not the native pipeline.

When should dedup be skipped during ingestion?

Skip semantic dedup for operational or state files, meeting transcripts, timeline entries on existing pages, and media files. These are unique by definition or deduplicated by filename and hash rather than semantic similarity.

What is the difference between brain-ingest-gate and concept-synthesis?

brain-ingest-gate is prevention at write time: it decides whether a new page should exist before anything is written. concept-synthesis is retroactive cleanup that deduplicates concept stubs already inside the brain.