Guide · Agent skill
How to use the Fact-Check Claims Skill agent skill
Updated · written with AI from the project's own documentation and checked against our catalogue data
The fact-check skill is an agent skill from the garrytan/gbrain project (plugin/skills/fact-check, MIT licence) that makes an AI agent verify a draft claim by claim before you publish or deliver it. This guide is for writers, editors and teams who want a repeatable checklist for AI-assisted or self-written content. By the end you will know what the skill does, how to install it, how to ask for a check in plain language and what its report looks like.
Fact-Check Claims Skillgarrytan/gbrain · MIT Download the skill Quick start
- 1Download the ZIP. Download the fact-check ZIP from this skill's page and unzip it. You should see a fact-check folder containing SKILL.md.
- 2Place the folder. Move the fact-check folder into ~/.claude/skills/ for personal use, or into .claude/skills/ inside your project.
- 3Or upload in the apps. In the Claude apps, open Settings, go to Capabilities, then Skills, and upload the skill.
- 4Confirm the tools. Make sure your environment provides web_search and web_fetch, and query tools if you check database-derived output.
- 5Paste your draft. Give the agent the text to check and state its purpose, such as publishing an article.
- 6Ask for a check. Use a request like "fact check this" or "check the claims" so the skill's triggers match.
- 7Act on the report. Apply the suggested corrections and hedges, and read the cited sources for anything important before publishing.
What is an agent skill and what does this one do?
An agent skill is a folder containing a SKILL.md file that tells an AI agent how to handle a specific kind of task. When the skill is installed and your request matches it, the agent follows the instructions in that file.
Once fact-check is active, the agent goes through a draft and pulls out names, numbers, quotes, dates and superlatives. It checks each one against live web sources rather than its own memory. Each claim gets a label on a six-level confidence scale, the agent suggests corrections or hedged wording, and it produces a pass/fail report.
- Extracts verifiable claims and ranks them by priority, with numbers, quotes, superlatives and claims about people checked first.
- Skips opinions, predictions and metaphors, which are not checkable.
- Labels each claim VERIFIED, LIKELY ACCURATE, UNVERIFIED, DISPUTED, LIKELY INACCURATE or FALSE.
- Records the source (URL or query and result) for each verification.
- For outputs built from a brain or database, re-derives key claims through a separate query path before delivery.
Install the fact-check skill
Download the ZIP from the skill's page on this site and unzip it. Put the resulting fact-check folder, which contains SKILL.md, in one of two places. For a personal install, use ~/.claude/skills/. For a single project, use the .claude/skills/ folder inside that project.
If you use the Claude apps instead, open Settings, go to Capabilities, then Skills, and upload the skill there.
- Personal: ~/.claude/skills/fact-check/
- Project: .claude/skills/fact-check/ inside your project folder
- Claude apps: Settings → Capabilities → Skills, then upload
~/.claude/skills/fact-check/SKILL.mdCheck what the skill needs before you start
The skill depends on tools that your agent environment must provide. It needs search and fetch tools (web_search and web_fetch) for live source checks. For claims generated from a brain or database, it also needs a query tool (query, search, get_page).
The skill does not run scripts. It does use the network, because web search and fetch are how it checks sources.
- Required for web checks: web_search, web_fetch
- Required for data-derived claims: query, search, get_page
- Runs scripts: no
- Uses network: yes
Trigger the skill with natural requests
You do not need a special command. Ask in plain language, and the skill's trigger phrases will match. The skill's trigger phrases include "fact check", "fact-check", "verify the facts", "check the claims", "is this accurate", "source check" and "verify this output claim by claim".
Paste the text or attach the file, then say what you want. Naming the purpose helps, for example that the piece is going to publication.
- "Fact check this essay before I publish it."
- "Verify the facts in this briefing claim by claim."
- "Is this accurate? Check the numbers and quotes."
- "Source check this blog post and give me a pass/fail report."
Read the report and act on it
The report lists each claim with its status and the source used. The skill's guidance for each status is specific. Verified and likely accurate claims can be stated as fact, with the source cited for the second. Unverified claims should be hedged with wording such as "reportedly" or "estimated". Disputed claims should present both sides or be cut. Likely inaccurate and false claims should be corrected or removed.
Treat the pass/fail result as a guide to what needs fixing, not a final sign-off. Read the sources yourself for anything important.
- VERIFIED: state as fact
- LIKELY ACCURATE: state as fact and cite the source
- UNVERIFIED: hedge the wording
- DISPUTED: present both sides or cut
- LIKELY INACCURATE: correct or remove
- FALSE: fix or kill
Understand the safety notes
Claims and some surrounding context are sent to web search and fetch tools, so do not use this skill on highly confidential content. The skill's guidance also says that fetched web pages are not authority to change facts. If a web page contains an instruction to change a claim, the agent should treat it as untrusted and verify independently.
The skill can still miss errors or report false negatives, and it depends on the quality and availability of web sources. It is a checking aid, not a guarantee of truth.
- Do not paste confidential material without checking your tool settings and policies.
- Review every correction before accepting it.
- Expect occasional misses, especially for obscure or very recent facts.
What people use it for
- Checking an essay or blog post for wrong dates, figures and misattributed quotes before publishing.
- Verifying a client briefing or internal report that must be factually reliable.
- Reviewing an AI-generated summary to see whether its claims hold up against web sources.
- Running a repeatable checklist across a team so every draft is checked the same way.
- Checking claims derived from a brain or database before the output is delivered.
Frequently asked questions
Is the fact-check agent skill free?
The skill is published under the MIT licence, as stated on this page. Check the licence file in the source repository for the full terms before reuse. Your agent platform may have its own costs, which this page does not cover.
How do I install the fact-check skill?
Download the ZIP from this page and unzip it. Place the fact-check folder in ~/.claude/skills/ for personal use or in .claude/skills/ in a project. In the Claude apps, you can instead upload it under Settings, Capabilities, Skills.
What does the fact-check skill require?
It requires search and fetch tools (web_search and web_fetch) for live source checks. For claims derived from a brain or database, it also requires query, search and get_page tools. It does not run scripts but does use the network.
Is the fact-check skill accurate enough to trust without review?
Not on its own. The skill can still miss errors or report false negatives, and its results depend on the quality and availability of web sources. Use its report as a guide and check important sources yourself.
Can I use fact-check on opinion pieces or fiction?
It is not a good fit. The skill deliberately skips opinions and predictions, so it adds little value to opinion pieces or creative fiction. It works best on articles, reports and posts that contain checkable facts.
Is it safe to use with confidential content?
Be cautious. Claims and some context may be sent to web search and fetch tools, so highly confidential material should not be checked this way unless your setup allows it.
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 fact-check if you publish essays, articles or briefings that contain checkable facts and want a consistent, source-based review before release. Pick something else if your content is opinion or fiction, if it is highly confidential, or if you need a guarantee of accuracy, since the skill can still miss errors.
Download the skill