Fact-Check Gates
Fact-Check Gates
Hard verification gates for factual content — video scripts, articles, newsletters, social posts. Every gate exists because a real published piece had to be deleted after publication, called out by viewers or domain experts in the comments.
The pattern behind those deletions was never "the source was fake." It was subtler: the sources were real, but the claim outran them.
What it does
Install this skill and your agent becomes a strict pre-publication reviewer. Give it a draft and it runs every factual claim through five hard gates:
| Gate | Catches |
|---|---|
| 1. Scope of claim | "People have used X since ancient times" — whose ancestors, exactly? Historical continuity claims almost always describe elites and surviving artifacts, not ordinary people. |
| 2. Primary data | Causal claims about tides, sunrise, moon phases, weather records — checked against the official table, not a news report repeating the claim. |
| 3. Source scope | Two articles repeating the same study is ONE source. A claim broader than its source is unsupported, no matter how many outlets repeat it. |
| 4. Time-point re-verification | Facts rot between research and publication — and titles written last from memory reintroduce the debunked version. |
| 5. Don't widen the answer | "This weather system exits tonight" is not "the rain is over." One-sided reassurance is the same offense as fearmongering. |
It also enforces review hygiene: the review itself never asserts unverified replacement facts. Corrections are phrased as verification instructions ("check the tide table for that date"), not new claims to take on faith.
Example
Input:
Please fact-check this draft script before I publish it:
"Most of the world eats from ceramic. But here, people have used metal rice bowls since the Three Kingdoms era — an unbroken 1,500-year tradition. And last week's coastal flooding? It got so bad because the downpour coincided with the pre-dawn high tide. 70% of people still prefer metal bowls today, so this tradition isn't going anywhere. No need to worry."
Sources: two news articles.
Output (abridged):
## Fact-Check Gate Report
1. "used metal rice bowls since the Three Kingdoms era — unbroken"
Gate 1: REJECT — era marker + continuity verb with no scope
qualifier. The era's metal vessels are elite/burial finds; "people"
reads as "everyone."
Fix: "Surviving artifacts show metal bowls in use since the Three
Kingdoms era — mostly among royalty and in burial finds."
2. "flooding got so bad because of the pre-dawn high tide"
Gate 2: REJECT — causal tide claim backed only by media re-reporting.
Check the national hydrographic agency's tide table for that date
and location before publishing.
3. "70% of people still prefer metal bowls"
Gate 3: REJECT — no survey org, no n, no population. The denominator
is missing.
4. "No need to worry"
Gate 5: REJECT — one-sided future reassurance no source bounds.
### Verdict: BLOCKED (4 claims rejected) — smallest fixes included above.
Who it's for
- Knowledge-content creators (YouTube / Shorts / Reels scripts)
- Newsletter and blog writers
- Social media managers
- Anyone whose credibility dies with one deleted post
What you get
A single SKILL.md — clear, auditable instructions your agent follows. No scripts, no network calls, no credentials, no data leaves your machine. Works with Claude Code and SKILL.md-compatible agents.
FAQ
Does it do the web research for me?
It directs verification: which claims need which primary source, and what to check. If your agent has web access, it will use it — but the skill's core value is refusing to let unverified claims through, including in its own suggested rewrites.
Is this a plagiarism/AI-detection tool?
No. It reviews factual accuracy and claim scope, not authorship.
Note: This skill is a set of instructions for an AI agent; output quality depends on the underlying model. It reduces factual-risk systematically but is not a substitute for your own editorial judgment on high-stakes claims.
