concept-synthesis

Deduplicate, tier, and synthesize raw concept stubs into a tiered intellectual map.

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

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Ingestion pipelines create a concept page for every idea mentioned, producing thousands of duplicate, flat stubs with no synthesis, tiering, or clustering. This Skill transforms that raw material into a curated intellectual map with deduplication, tier assignments, synthesis narratives, and a reversible curation cull. ## Core Features & Use Cases - Multi-phase dedup and tiering: Jaccard, substring, and semantic dedup merge stubs into canonical concepts, then score and tier them from T1 Canon to T4 Riff based on frequency, timespan, and breadth. - LLM synthesis and clustering: T1/T2 concepts get evolution narratives, best-articulation quotes, and cross-links, while all concepts are grouped into named clusters with a master concepts/README.md map. - Reversible curation cull (Phase 5): A 6-axis weighted rubric issues ELITE/KEEP/MERGE/DELETE verdicts with grounding labels, cluster budgets, and merge-with-backlinks so nothing is hard-deleted. - Use Case: After months of ingesting meeting notes and voice memos, run the cull pass to fold 40 near-identical stubs on one theme into a single canonical concept whose merge ledger drives tier promotion. ## Quick Start Ask the agent to run concept synthesis on your brain to dedupe the concept stubs, tier them from canon to riff, and build your intellectual map.

Frequently Asked Questions about concept-synthesis

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

FAQPage Schema
How do I deduplicate thousands of concept notes in my knowledge base?

Run the concept synthesis pass, which applies Jaccard word-overlap and substring dedup deterministically, then an LLM pass to confirm semantic duplicates. Duplicates are merged into canonical pages with aliases and timelines preserved in frontmatter.

How are concepts tiered from Canon to Riff?

Each canonical concept is scored on frequency of distinct sources, timespan between first and last mention, breadth across months, and engagement. Scores map to four tiers: T1 Canon, T2 Developing, T3 Speculative, and T4 Riff, with only T1 and T2 receiving full LLM synthesis.

Does the curation cull permanently delete my concept pages?

No, every destructive step is reversible. Absorbed pages are archived verbatim under a _merged/ tree, soft-deleted rather than purged, and page history supports restore and revert operations at every step.

When should I run the concept cull versus full synthesis?

Run synthesis weekly or after major ingestion batches for newly promoted T1/T2 concepts. Run the Phase 5 cull less often, monthly or after large ingestion waves, and always test on 3-5 clusters before a full pass.

Why are similar concepts not being merged automatically?

The distinctness guard is a hard veto: an LLM judge must affirmatively confirm two concepts state the same mechanism before any merge is written. Shared vocabulary alone is insufficient, and the judge's verdict is logged per cluster.