fact-checking
Skill37 starsChanged 37 days agoMITView on GitHub
Use when verifying claims. Covers tracing a claim to its origin, evaluating sources, detecting laundered statistics, and reporting a verdict with the evidence rather than an assertion. Trigger on "is this true", "where does this statistic come from", "fact check this article". Do not use for open-ended multi-source research — see the deep-research skill.
7 other projects use a version of this.
Adopted by
Most recently taken up by
- lengrufeng2008/dsh-skill-librarytheir own version, first seen Nov 15, 2026
- github/squadtheir own version, first seen Nov 8, 2026
- axiomantic/spellbooktheir own version, first seen Nov 1, 2026
- KMaruthi2002/universal-llm-skillstheir own version, first seen Oct 25, 2026
- microsoft/github-vteam-industry-demostheir own version, first seen Oct 25, 2026
- ccgargantua/gars-claude-skillstheir own version, first seen Sep 16, 2026
Other versions7 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 7
- github/gh-aw★ 5.6ksame name
- github/squad★ 3.4ksame name
- axiomantic/spellbook★ 13same name
- microsoft/github-vteam-industry-demos★ 2same name
- ccgargantua/gars-claude-skills★ 1same name
- KMaruthi2002/universal-llm-skills★ 1same name
- lengrufeng2008/dsh-skill-library★ 0same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository137 other files it gives its agents.
- agent-instructionsSkill
- agent-memorySkill
- hooksSkill
- plugin-developmentSkill
- skill-authoringSkill
- skill-reviewSkill
- slash-commandsSkill
- subagentsSkill
- agent-designSkill
- context-engineeringSkill
- fine-tuningSkill
- llm-cost-optimizationSkill
- llm-evaluationSkill
- llm-integrationSkill
- mcp-serverSkill
- ml-pipelineSkill
- model-selectionSkill
- prompt-engineeringSkill
- ragSkill
- structured-outputSkill
- api-designSkill
- background-jobsSkill
- cachingSkill
- djangoSkill
- event-driven-architectureSkill
- fastapiSkill
- graphqlSkill
- microservicesSkill
- nestjsSkill
- realtime-websocketsSkill
- spring-bootSkill
- competitive-analysisSkill
- lead-researchSkill
- product-analysisSkill
- resumeSkill
- user-researchSkill
- database-performanceSkill
- data-modelingSkill
- data-qualitySkill
- pandasSkill
- postgresSkill
- sparkSkill
- brand-guidelinesSkill
- generative-artSkill
- themingSkill
- ui-reviewSkill
- visual-designSkill
- architecture-decisionsSkill
- bug-fix-protocolSkill
- cli-developmentSkill
- code-reviewSkill
- debuggingSkill
- design-patternsSkill
- domain-modelingSkill
- git-workflowSkill
- legacy-modernizationSkill
- refactoringSkill
- repository-explorationSkill
- simple-designSkill
- system-designSkill
- The first 60 shown.
The file
We split this skill into the rules it gives an agent
3 rules. 0 are common, 3 are distinctive, and 0 are very specific to this project.
0
Common
Most of the 7 edited versions we read keep these.
3
Distinctive
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 14
Verify a claim by tracing it to its origin.
- Distinctive: line 56
Check the date. A statistic that was true in 2011 may have no relationship to the present, and it will still be quoted.
- Distinctive: line 57
Check who funded the study, and what they wanted it to show.
“Common” is measured: kept by at least 6 in 10 of the 7 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Fact-Checking
Purpose
Verify a claim by tracing it to its origin. Most false claims in circulation are not fabrications; they are real findings that have been stripped of their qualifications, or numbers that have acquired a citation through repetition.
When to Use
- Verifying a statistic or a claim before publishing it.
- A number that is widely repeated and never sourced.
- Checking an assertion in a document, an article, or a report.
- Evaluating a claim that seems too clean.
Capabilities
- Tracing a claim to its primary source.
- Detecting circular citation.
- Evaluating study quality and its actual conclusion.
- Identifying the qualifications that were dropped.
- Reporting a verdict with evidence.
Inputs
- The claim, quoted exactly.
- Where it appeared and what it was used to support.
Outputs
- A verdict: supported, unsupported, misleading, or unverifiable.
- The chain of citation, traced.
- What the original source actually says.
Workflow
- Quote the claim exactly — Vague paraphrases are unfalsifiable. "Most projects fail" cannot be checked; "70% of software projects fail" can.
- Follow the citation chain — Each source cites another. Follow it until you reach a primary source, or until it loops, or until it disappears. All three outcomes are informative.
- Read what the original actually says — Very frequently, the original is a limited finding about a specific population that has been generalized into a universal claim by the time it reaches you.
- Check for independence — Twelve sources citing one study is one study.
- Check the qualifications that were dropped — "In a survey of 43 startups in one accelerator cohort" becomes "70% of startups" within three citations.
- Report the chain, not just the verdict — The chain is the evidence.
Best Practices
- A claim with no traceable origin is not established, however widely it is repeated. Repetition is not evidence.
- Numbers that are suspiciously round (90%, 70%, 50%) are frequently rhetorical rather than measured. Check them.
- The most common falsification is not fabrication but the dropping of qualifications: a finding about a specific population, under specific conditions, becomes a universal law.
- Check the date. A statistic that was true in 2011 may have no relationship to the present, and it will still be quoted.
- Check who funded the study, and what they wanted it to show. This does not invalidate it, but it determines how much scrutiny it warrants.
- "Unverifiable" is a legitimate verdict, and a useful one. It is not the same as "false".
Examples
Tracing a claim to its origin:
## Claim
"70% of digital transformation projects fail."
## Chain
Appears in: a 2025 consultancy report.
Cited to -> a 2023 industry magazine article.
Cited to -> a 2019 blog post by a management consultancy.
Cited to -> a 2016 book.
Cited to -> a 2013 article in a business publication.
Cited to -> "research suggests" — no citation. The chain ends.
There is no study. The number entered circulation without a source and acquired
authority through repetition. Each citation cited the previous citation, and by
the fifth hop it was being reported as an established statistic.
## What the closest thing to evidence actually says
A separate 2011 survey — sometimes cited in the same context — found that
of 1,471 IT projects, the average cost overrun was 27%, and one in six projects
had a cost overrun exceeding 200%. That is a real finding, and it does not say
70% of anything failed. It is about cost overrun, not failure, and "failure" is
never defined in any of the sources in the chain above.
## Verdict: UNSUPPORTED
The number has no traceable origin. It should not be used. If a figure is
needed, cite the 2011 study for what it actually measured — cost overrun — and
state the definition.
Note also: "digital transformation" is not defined in any source in the chain,
which makes the claim unfalsifiable even in principle.Detecting a dropped qualification:
## Claim as it appeared
"Remote workers are 13% more productive."
## What the original study found
A 2015 randomized trial at a single Chinese travel agency, on call-center staff
performing a highly measurable, repetitive task, over nine months.
The 13% figure breaks down as: ~9% from working more minutes per shift (fewer
breaks, fewer sick days) and ~4% from more calls per minute.
The same study found that half the participants asked to return to the office,
and that promotion rates for remote workers fell.
## Verdict: MISLEADING
The finding is real. The claim, as stated, is not what the study found. It
generalizes a result about call-center piecework at one company to knowledge
work generally, and it omits both the mechanism (more hours, not more output
per hour) and the negative findings from the same study.Notes
- The citation chain ending in "research suggests" with no source is the single most common pattern in laundered statistics, and it is why tracing to the primary source is not optional.
- A finding whose mechanism is "they worked more hours" being reported as "they are more productive" is not a lie, but it is not the truth either. Read the mechanism.
- Reporting the chain, rather than just the verdict, is what makes the fact-check checkable. Otherwise you are asking to be trusted, which is the problem you started with.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

