cite-check
Skill0 starsChanged 5 months agoMITView on GitHub
Every factual claim either carries a source or wears an UNVERIFIED tag. Nothing real-world gets stated as settled fact on memory alone.
6 other projects use a version of this.
Adopted by
Most recently taken up by
- Walsworth-Group/physkittheir own version, first seen Nov 8, 2026
- enstw/thesisbug2their own version, first seen Nov 8, 2026
- zedxter/utility-belttheir own version, first seen Nov 1, 2026
- williamhorschak-dev/wisconsin-legal-ai-skillstheir own version, first seen Oct 25, 2026
- Sprintlaw/legal-skillstheir own version, first seen Oct 17, 2026
- reutavidan/zetesis-plugintheir own version, first seen Aug 6, 2026
Other versions6 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 6
- Walsworth-Group/physkit★ 7same name
- zedxter/utility-belt★ 4same name
- williamhorschak-dev/wisconsin-legal-ai-skills★ 2same name
- enstw/thesisbug2★ 0same name
- reutavidan/zetesis-plugin★ 0same name
- Sprintlaw/legal-skills★ 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 repository16 other files it gives its agents.
- assumption-auditSkill
- compactSkill
- confidence-tagSkill
- context-budgetSkill
- devils-advocateSkill
- distillSkill
- first-principlesSkill
- handoffSkill
- pre-mortemSkill
- red-teamSkill
- rubber-duckSkill
- scope-cutterSkill
- scratchpadSkill
- self-consistencySkill
- steelmanSkill
- teach-backSkill
The file
We split this skill into the rules it gives an agent
1 rule. 0 are common, 1 is distinctive, and 0 are very specific to this project.
0
Common
Most of the 6 edited versions we read keep these.
1
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 7
Every factual claim either carries a source or wears an UNVERIFIED tag.
“Common” is measured: kept by at least 6 in 10 of the 6 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
name: cite-check description: Forces every factual claim to carry a real source, and explicitly tags anything unsupported as UNVERIFIED rather than asserting it as fact — turning confident hallucination into a visible, honest gap. Use this skill when the user is doing research or factual writing, asks "is this true?" or "back this up," wants claims about real-world data, people, events, statistics, dates, or quotes, requests fact-checking of a draft, or whenever stating something wrong with confidence would be worse than admitting uncertainty.
Cite or Flag
Every factual claim either carries a source or wears an UNVERIFIED tag. Nothing real-world gets stated as settled fact on memory alone.
When to use
- Research, reports, or any writing that asserts real-world facts: statistics, dates, prices, quotes, names, attributions, study findings. - The user asks "is this true," "back this up," "what's your source," or hands you a draft to fact-check. - Anti-hallucination contexts where a confidently wrong figure could mislead a decision or get published. - Comparing or summarizing external sources, where it must stay clear which claim came from where.
When NOT to use
- Pure reasoning, math, or code logic with no external-fact dependency — verify those by re-derivation (self-consistency), not citation. - Clearly subjective or hypothetical content ("imagine a world where…", opinions, creative writing). - The user's own private context they just gave you (their numbers, their plans) — that's their input, not a claim to source. - Brainstorming where flagging every speculative idea as UNVERIFIED would just be noise — though still label anything that sounds like a hard fact.
The method
1. Extract the claims. Read your own draft answer and isolate each statement that asserts something checkable about the real world (a number, a date, a "X said Y," a "studies show"). 2. Classify each claim by support level: - SOURCED — backed by a specific, namable source you can point to (document, URL, dataset, the user's own input). - MEMORY — you believe it from training but have no live source; treat as a candidate for verification, not as confirmed. - UNVERIFIED / UNKNOWN — you're unsure, it's plausibly wrong, or it's beyond your knowledge cutoff. 3. Source what you can. For MEMORY claims that matter, look them up (search/fetch) and either upgrade to SOURCED with the reference, or down to UNVERIFIED if you can't confirm. 4. Tag the rest in the output. Anything still on memory gets an inline marker like "(unverified)" or a note; anything unknown is stated as "I couldn't confirm…". Never launder an uncertain claim into a confident sentence. 5. Be specific with sources. "According to [publisher/document, date]" beats "studies show." A citation that can't be located is not a citation. 6. Separate fact from your inference. If you're concluding something from the sources rather than quoting them, say so — don't dress an inference as a sourced fact. 7. Flag absence loudly. If a key claim has no support at all, say "I have no reliable source for this" rather than filling the gap with a plausible guess.
What good looks like
- A reader can trace every load-bearing fact to a source or see exactly which facts are unverified. - Numbers, dates, and quotes carry their origin. - "I don't know" and "I couldn't confirm X" appear when they're true, without apology or padding. - Speculation, inference, and sourced fact are visually distinct, never blended.
Anti-patterns
- Confident specificity with no source ("exactly 73% of users…") — the signature move of hallucination. - Vague authority laundering: "experts say," "research shows," "it's well known that." - Fake or unlocatable citations, mangled titles, invented DOIs or URLs. - Stating a memory-based fact and only admitting uncertainty if challenged. - Tagging everything UNVERIFIED as a CYA reflex, so the tag stops meaning anything.
Example
Draft: "Python was released in 1991 and is now the most popular language, used by 80% of developers." After cite-check: - "Python was first released in 1991" — SOURCED (python.org history; widely documented). - "the most popular language" — depends on the ranking; per the 2024 Stack Overflow Developer Survey it's among the top, but "most popular" varies by metric. (qualified, sourced) - "used by 80% of developers" — UNVERIFIED; I have no source for this figure and it looks inflated. Removed pending a real number.
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.

