Guide · Agent skill
How to use the Concept Synthesis agent skill (concept-synthesis)
Updated · written with AI from the project's own documentation and checked against our catalogue data
Concept Synthesis is an agent skill: a set of written instructions that tells an AI agent how to clean up a large collection of short concept pages. It merges duplicates, scores each idea by how often and how long it recurs, sorts them into tiers from Canon to Riff, and writes a map of the result. This guide explains what the skill does, how to install it, how to ask for it in plain language, and what to check before accepting its changes.
Concept Synthesisgarrytan/gbrain · MIT Download the skill Quick start
- 1Download and unzip. Download the skill ZIP and unzip it into ~/.claude/skills/ or a project's .claude/skills/ folder.
- 2Check your knowledge base. Confirm you have gbrain concept pages and that its query and put_page operations are available to the agent.
- 3Back up concepts/. Copy your concept pages somewhere safe, since the skill rewrites and merges pages.
- 4Start with a small test. Ask for synthesis on three to five clusters first, rather than the whole collection.
- 5Ask in plain language. Say something like 'synthesize my concepts' or 'build my intellectual map' to trigger the skill.
- 6Review tiers and the map. Open concepts/README.md and spot-check several tier assignments and syntheses.
- 7Review cull verdicts. Read the keep, merge, rewrite and delete verdicts before applying any of them in bulk.
What the skill makes the agent do
Many ingestion tools create a concept page for every idea they notice in your notes, articles or voice memos. Over time this produces thousands of stubs, many of them duplicates, with no synthesis and no grouping. The skill works through the collection in phases: first it merges duplicates, then it scores and tiers each remaining concept, then it writes synthesis pages for the top tiers, and finally it builds a master map.
Tiers run from T1 Canon (ideas that recur across long spans) through T2 Developing and T3 Speculative to T4 Riff. Only the T1 and T2 concepts get a full written synthesis; lower tiers stay as short stubs with a quote and a source link.
- Phase 1: merge duplicates and near-duplicates, and fold their timelines and aliases into one canonical page
- Phase 2: score each concept on frequency, timespan and breadth, then assign a tier
- Phase 3: write an evolution narrative and best quote for T1 and T2 concepts
- Phase 4: group concepts into clusters and write concepts/README.md as the full map
- Phase 5: a curation cull that gives each concept a keep, merge, rewrite or delete verdict
Install the skill (ZIP download or Claude app upload)
Download the skill ZIP from the page where you found this guide. The skill ships as a folder containing SKILL.md. To make it available to Claude Code for your user account, unzip it into ~/.claude/skills/. To limit it to one project, unzip it into that project's .claude/skills/ folder instead.
If you use the Claude apps rather than Claude Code, open Settings, go to Capabilities, then Skills, and upload the skill there.
- Personal use across projects: ~/.claude/skills/
- Single project only: the .claude/skills/ folder inside that project
- Claude apps: Settings → Capabilities → Skills → upload
Prepare your concept pages
The skill expects a gbrain knowledge base that contains concept pages, and it uses gbrain's query and put_page operations to read and write them. An LLM must also be available to the agent, because the semantic duplicate check and the synthesis passes need it. The skill's own invocation example lists all concept pages with a query on type:concept.
Back up your concept folder before a large run. The skill writes into concepts/ and rewrites pages as it goes.
- A gbrain knowledge base with concept pages
- gbrain's query and put_page operations available to the agent
- An LLM the agent can call for semantic dedup and synthesis
gbrain query "type:concept" --limit 10000 --jsonTrigger the skill with natural requests
You do not need special syntax. Ask the agent in plain language, and the skill's trigger phrases will match. The source lists phrases such as these, and similar wording should work as well.
For a first run, ask for a limited pass on a few clusters rather than the whole collection. The skill's own convention is to cull 3 to 5 clusters first, read the output, and only then run the full pass.
- "Synthesize my concepts"
- "Build my intellectual map"
- "Trace idea evolution"
- "Canon vs riff" or "which concepts to keep"
- "Cull my concepts"
Read the output: tiers, synthesis and the map
Each T1 or T2 concept page gets tier metadata, a synthesis of two to four paragraphs, a best-articulation quote with its source, an evolution table and related-concept links. The concepts/README.md file lists every tier with counts and the earliest and latest source dates.
Scores come from frequency (distinct sources), timespan (first to last mention in days), breadth (distinct months) and, where available, engagement on the source material. Check a few tier assignments by hand before trusting the whole map.
- Tier 1 Canon: recurs across long spans
- Tier 2 Developing: sharpening, may become Canon
- Tier 3 Speculative: testing in public
- Tier 4 Riff: single or thin mentions, stub only
Review the Phase 5 curation cull
Phase 5 gives each concept one hard verdict: ELITE, KEEP, MERGE/REWRITE or DELETE. It scores six axes, with three substance axes weighted double and three packaging axes weighted single, applies a minimum substance gate, and adds grounding labels (VERIFIED, OPINION, NEEDS_SOURCE, UNSAFE). Merges fold a concept into a cluster canonical with backlinks, and the skill describes these as fully reversible.
The cull issues verdicts that can remove or merge pages. Read the verdicts before you let the agent apply them in bulk. The skill also includes a reputational-risk gate, but decisions about what to publish still need a person.
- Core test: would this sharpen a thought in two years, or would I scroll past it?
- Scroll-past verdicts are DELETE
- Review cluster by cluster before any bulk change
What people use it for
- Turning a few thousand auto-generated idea pages into a short list of recurring themes
- Seeing which of your ideas have recurred over many months, and which appeared once and faded
- Writing an evolution summary of one idea across the sources where you mentioned it
- Trimming hollow stub pages while keeping the concepts that carry real insight
Frequently asked questions
What do I need to use the concept-synthesis skill?
You need a gbrain knowledge base containing concept pages, with its query and put_page operations available to the agent. You also need an LLM the agent can call for semantic deduplication and synthesis. The skill is markdown instructions and does not run scripts.
Where do I install the concept-synthesis skill?
Unzip the downloaded skill into ~/.claude/skills/ for personal use, or into a project's .claude/skills/ folder to limit it to that project. In the Claude apps, you can upload it under Settings, then Capabilities, then Skills.
Is the concept-synthesis skill free?
The skill is listed under the MIT licence. This guide does not state any price, so check the licence and the page you downloaded it from for terms.
Does the skill delete my files automatically?
No, but its Phase 5 cull can issue delete or merge verdicts that remove or combine pages. The skill describes merges as reversible, yet you should review verdicts before applying them in bulk.
Does the skill use the network or run scripts?
According to its listing, it does not run scripts and does not use the network. It works through gbrain's operations and an LLM pass.
Is this worth using with only a few notes?
Probably not. The scoring adds little value when you only have a handful of concepts, so for a small collection you may be better off organising pages by hand.
New agent skills, repos and AI tools — every week
The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.
Our verdict
Use this skill if you have hundreds or thousands of auto-generated concept pages in gbrain and want them merged, tiered and mapped. Skip it for small note collections, or if you want changes applied automatically without reviewing the verdicts first.
Download the skill