Guide · Agent skill
How to use the Research Compendium Builder agent skill
Updated · written with AI from the project's own documentation and checked against our catalogue data
The Research Compendium Builder is an agent skill that makes an AI agent run a full research project on a topic and save the results as a reusable knowledge page. This guide explains what the skill does, how to install it from a ZIP file or the Claude apps, how to trigger it with natural requests, and what it needs and touches. By the end you will know how to start a compendium, how to deepen one later, and where the safety limits are.
Research Compendium Buildergarrytan/gbrain · MIT Download the skill Quick start
- 1Download and unzip. Download the skill ZIP from this page and unzip it so you have a folder containing SKILL.md.
- 2Place the folder. Move the folder into ~/.claude/skills/ for personal use or into a project's .claude/skills/ directory. Alternatively, upload it under Settings → Capabilities → Skills in the Claude apps.
- 3Confirm the requirements. Make sure the gbrain knowledge base tool and web access are available, and that the companion convention files are in place.
- 4Make a request. Ask for a compendium or definitive guide on a named topic, and say you want the sources archived and summarized.
- 5State privacy limits. Tell the agent up front if any material is personal or sensitive so it skips verbatim archiving for that material.
- 6Choose a depth. Start at the base level and ask to go deeper later, which adds only the missing layers.
- 7Review the index. Open the index page to see archived sources, summaries, passes run and checks passed.
What the research-compendium skill makes the agent do
The skill turns a research question into a permanent asset in a knowledge base. The agent follows four phases: search your existing knowledge base first, fetch and archive the primary sources, write one summary for each source, and then write a single self-contained guide on the topic.
Each source is paired with its own summary, and the index page records what has been done so the work can be extended later without repeating finished steps. The skill also checks your retention preference before keeping verbatim copies of sensitive material, and it treats fetched web text as data rather than instructions.
- Phase 1: search the existing knowledge base before the open web.
- Phase 2: archive each primary source verbatim, unless your privacy or retention preference says otherwise.
- Phase 3: write one summary per source, paired by number.
- Phase 4: synthesize one compendium page with links back to the sources and summaries.
Install the skill from the ZIP file
Download the ZIP for this skill from this page and unzip it. The skill is a folder containing a file named SKILL.md. Place that folder in one of two locations: ~/.claude/skills/ to make it available across your projects, or .claude/skills/ inside a specific project to make it available only there.
If you use the Claude apps instead, open Settings, go to Capabilities, and upload the skill under Skills.
- Personal install: put the unzipped folder in ~/.claude/skills/.
- Project install: put the unzipped folder in the project's .claude/skills/ directory.
- Claude apps: Settings → Capabilities → Skills, then upload.
Trigger the skill with natural requests
You do not need a special command. The skill responds to requests that match its trigger phrases. Ask for a compendium, a definitive guide, or a comprehensive guide, and say you want the sources read and saved.
Be explicit about the topic and the outcome you want. A request that names the topic, asks for sources to be saved, and asks for one written result will match the skill most reliably.
- "Research everything about spaced repetition, read the sources, and write me a definitive guide."
- "Read them all and summarize, then build a compendium on zero-knowledge proofs."
- "Archive the sources then summarize them for a comprehensive guide to container security."
- "Deepen the compendium on spaced repetition."
Choose a depth level
Depth works as a dial. Each level includes everything below it, so you can start small and later ask for more without redoing finished work. If you do not specify a level, a fresh topic starts at the base level, and if a compendium already exists, saying go deeper raises it by one level.
The base level runs the four-phase pipeline. Higher levels add full-text primary sources, books with verbatim quotes, a search for critiques of the main ideas, and eventually full book mirrors. The source file says the highest level is a multi-day budget, so confirm with the user before starting it.
- compendium: base synthesis, the four-phase pipeline.
- compendium+: adds full-text primary sources and a fact-check gate.
- compendium++: adds books and a counter-canon hunt for critiques.
- compendium+++: adds full book mirrors on the most central books, with user opt-in.
- compendium++++: exhaustive, multi-day, confirm first.
Know the requirements and safety limits
The skill needs the gbrain knowledge base tool with commands such as search, link, lsd, files upload-raw and check-backlinks. It also needs web access to fetch sources. The skill refers to companion convention files for brain-first lookup, quality, untrusted content and filing rules; those files are not included in the reviewed file, so they must be present in your setup for the skill to follow those rules fully.
The skill does not run scripts, but it does use the network. Archiving is gated by privacy: for personal, sensitive or third-party-private material such as correspondence, medical or financial records, or private group content, the skill keeps a citation and summary and skips the verbatim copy. Fetched text containing instructions aimed at an AI is kept as quoted content, flagged, and never acted on.
- Requires: gbrain knowledge base tool, web access, companion convention files.
- Runs scripts: no.
- Uses network: yes.
- Writes under research/ in your knowledge base.
Read the output and tips
A finished run produces a folder of pages under research/<topic-slug>/: sources, summaries, a compendium page, and an index page that records the depth reached. Source and summary pages share a number so each pairs with one another, and links run in both directions.
For best results, state your privacy preference at the start, especially if the topic may involve personal material. Review the index page to see what was archived, what was only summarized, and which checks passed.
- Say early if some material should not be stored verbatim.
- Check the index page for what was archived versus summarized.
- Ask for a higher depth only when the base result is not enough.
What people use it for
- Building a lasting reference on a subject from many public papers, standards and articles.
- Maintaining a shared knowledge base where each claim is traced to an archived source with citations and backlinks.
- Producing a definitive guide with an audit trail showing where each claim came from.
- Extending an existing compendium later with another depth level without redoing finished work.
Frequently asked questions
How do I install the research-compendium skill?
Download the ZIP from this page and unzip it. Place the folder in ~/.claude/skills/ for personal use or in a project's .claude/skills/ directory. In the Claude apps, you can instead upload it under Settings → Capabilities → Skills.
What do I need to run research-compendium?
You need the gbrain knowledge base tool with commands such as search, link, lsd, files upload-raw and check-backlinks, plus web access to fetch sources. The companion convention files the skill references are not included in the reviewed file.
Is the research-compendium skill free?
The source file does not state pricing or licence terms for the skill beyond the listing's MIT licence. The source notes that the base level has a cost of tens of dollars and up to about an hour, so check that against your own usage.
Does research-compendium run scripts or access the internet?
It does not run scripts, according to this listing. It does use the network, because it fetches web sources to archive and summarize.
When should I not use this skill?
Avoid it for simple one-off questions, since the multi-phase process is heavy and slow. Use data-research for structured tables or trackers, and do not use it to collect private correspondence, medical or financial records, or other sensitive material.
How is this different from a normal web search or summary?
The skill archives each primary source, writes a summary for each one, and then produces a single synthesized guide with citations and backlinks. The source describes it as prose knowledge synthesis backed by an archived source corpus, rather than a quick web delta.
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 are a researcher, student or team building a lasting, citable reference from many public sources and want an audit trail for each claim. Pick something lighter if you only need a quick answer, need structured data in a table, or are working with private or sensitive material you do not want stored.
Download the skill