Skip to content
Skills Library
Marketing & content · Review & audit

fact-checking-explainers

/fact-checking-explainers

Use to fact-check an explainer against its sources before delivery. Trigger on "fact-check this explainer", "verify the claims", "check this against its sources", research-time source checks, or the post-draft gate. Ensures every checkable claim traces to a cited source or, for code, the real implementation. Required by creating-explainers and explaining-codebases.

Licence
MIT
Last repository change
08.07.2026

Assessment and security

100/ 100Class: good
Format25 / 25
Frontmatter with name and description, valid name, description up to 1,024 characters, name matching the folder.
Discoverability25 / 25
Description explains when to use the skill, has a reasonable length and uses third person.
Content35 / 35
At least 150 words and at most 500 lines, headings, step lists, examples or code.
Hygiene15 / 15
No placeholders such as TODO, no Windows paths, with a licence.
Riskno signals detected in the scan
Files in this package
3
Scripts
0
HTML comments
0
URLs in SKILL.md
0
Deprecation

No signals detected. Repository changed on 08.07.2026, without references to retired models or APIs.

Licence

MIT, open: use and modify while retaining author attribution.

What it does and when to use it

Use to fact-check an explainer against its sources before delivery. Trigger on "fact-check this explainer", "verify the claims", "check this against its sources", research-time source checks, or the post-draft gate. Ensures every checkable claim traces to a cited source or, for code, the real implementation. Required by creating-explainers and explaining-codebases.

Triggers from the description

fact-check this explainerverify the claimscheck this against its sources
Original author description

Use to fact-check an explainer against its sources before delivery. Trigger on "fact-check this explainer", "verify the claims", "check this against its sources", research-time source checks, or the post-draft gate. Ensures every checkable claim traces to a cited source or, for code, the real implementation. Required by creating-explainers and explaining-codebases.

How it works

  1. Overview
  2. The Iron Law
  3. When to Use
  4. What Counts as a Checkable Claim
  5. The Verification Process
  6. The Two Gates
  7. Resolution
  8. The Report
  9. Rationalization Table
  10. Red Flags
Original SKILL.md headings · 117 lines, 1,113 words
SKILL.md preview
---
name: fact-checking-explainers
description: Use to fact-check an explainer against its sources before delivery. Trigger on "fact-check this explainer", "verify the claims", "check this against its sources", research-time source checks, or the post-draft gate. Ensures every checkable claim traces to a cited source or, for code, the real implementation. Required by creating-explainers and explaining-codebases.
---

# Fact-Checking Explainers

## Overview

An explainer's authority comes from being right. A reader who catches one confident, wrong claim stops trusting the rest of the piece, figures and all. This skill makes claim verification enforceable by tying every checkable claim to evidence you can point to, correcting it, or cutting it. It is not a proofreading pass and not advisory: it is the gate an explainer passes before it is delivered.

## The Iron Law

```
NO EXPLAINER SHIPS WITH AN UNVERIFIED FACTUAL CLAIM.
```

Every checkable claim either traces to a source that supports it, gets corrected to match the source, or gets cut.

Apply the same resolution loop to every factual claim:
- **Support it** with an actual source passage you can point to.
- **Correct it** until it matches the source.
- **Cut it** when no source supports it.

Treat memory, plausibility, background knowledge, and deadline pressure as signals to run the loop, not reasons to skip it. "Verify" means checking an actual source in this session.

## When to Use

Run this skill:
- **At research-time**, over the sources gathered for an explainer, before any drafting.
- **Post-draft**, over the finished article, before delivery. This is a blocking gate.
- **On request**, whenever asked to fact-check, verify, or audit an explainer or article.

`creating-explainers` and `explaining-codebases` both call this skill at the gates above. When they do, interactive explainer delivery waits for a PASS report.

**When NOT to use:** content that is not an explainer, or a pure opinion or editorial piece that makes no factual claims. (Most explainers make many factual claims. Default to running it.)

## What Counts as a Checkable Claim

A checkable claim is a specific factual or empirical assertion:
- numbers, dates, measurements, percentages
- names, attributions, who-did-what, who-built-what
- direct quotes
- historical events and their order
- performance results and benchmarks
- mechanism claims: "X uses Y", "X causes Y", "X is faster than Y because Z"

Not checkable (do not flag these as factual claims):
- clearly-labeled interpretation or opinion ("the most elegant part is...")
- analogies and illustrative scenarios the reader knows are illustrative
- the author's framing and transitions

**Heuristic: if you are unsure whether something is a checkable claim, treat it as one and verify it.** The cost of over-checking is small; the cost of a wrong claim shipping is the whole article's credibility.

## The Verification Process

1. **Extract** every checkable claim, with its location in the article (section, figure caption, or `file:line` for code).
2. **Find the support.** For each claim, locate the passage in the cited source that backs it, and read that passage. For codebase explainers, open the actual code at the exact path and line. Do not treat the existence of a citation as support; a cited source often does not say what the draft claims.
3. **Assign a verdict** (table below).
4. **Resolve** everything that is not `supported`.

| Verdict | Meaning | Allowed at delivery? |
|---------|---------|----------------------|
| supported | A source passage directly backs the claim | Yes |
| needs-source | Plausible but no citation yet | No - add a source or downgrade |
| unsupported | No source backs it | No - source it, soften, or cut |
| contradicted | A source says otherwise | No - correct it or cut it |

## The Two Gates

**Research-time gate** (inside `creating-explainers` research intake). Before drafting, go through the gathered sources and confirm each one exists and actually supports the points it will be used for. Fabricated, misremembered, or misread sources are cheapest to catch here, before they are baked into prose.

**Post-draft gate** (every explainer, before delivery). Audit every checkable claim in the finished article against its cited source. **For codebase explainers, the source of truth is the real code.** Every "this function does X", every quoted snippet, every architecture claim is checked against the actual implementation at a specific path and line. Code drifts; a snippet that was right yesterday may be wrong today.

## Resolution

At delivery, every checkable claim must be `supported`. Resolve each other verdict with the same loop:
- **Add support** from a source passage (turns `needs-source` into `supported`).
- **Correct the claim** to match what the source actually says.
- **Soften** to clearly-labeled interpretation, if the statement is genuinely interpretive rather than factual.
- **Cut it.**

An unresolved `needs-source`, `unsupported`, or `contradicted` claim keeps the report at FAIL. Caveats do not make unsupported factual claims deliverable.

## The Report

Produce a claim-by-claim report plus an overall PASS/FAIL verdict. The exact format is in `references/verification-report-format.md`. The verdict is FAIL until every checkable claim is `supported`.

When invoked as a gate by another skill, return the report inline and hold delivery until PASS. When invoked directly by a user, present the report and offer to apply the fixes.

## Rationalization Table

| Excuse | Reality |
|--------|---------|
| "I wrote it, so I know it's right" | You know what you intended. Verify what you actually wrote. |
| "It's a well-known fact" | Well-known facts are wrong often enough to check. It takes 30 seconds. |
| "The source probably says this" | Probably is not a verdict. Open the source and confirm. |
| "Close enough" | Numbers, dates, and names are exact or they are wrong. |
| "It's only one claim" | One confident wrong claim discredits the whole explainer. |
| "I'll flag it and let the reader decide" | The gate is your job, not the reader's. |

## Red Flags

These thoughts mean stop and verify before the claim goes in:
- "I'm fairly sure that..."
- "If I recall correctly..."
- "roughly..." / "about..." / "something like..." attached to a specific number
- "it's basically the same as..."
- pasting a statistic without re-opening the source it came from
- citing a URL you have not actually opened in this session

## Common Mistakes

| Mistake | Fix |
|---------|-----|
| Verifying that the citation exists, not that it supports the claim | Read the cited passage. A real source can still fail to back the claim. |
| Checking the easy claims, waving through the hard ones | The hard-to-verify claims are exactly where errors hide. |
| For code: trusting a comment, a function name, or a variable name | Read the implementation. Names and comments can lie; the code does not. |
| Treating `needs-source` as shippable | `needs-source` is a to-do, not a pass. Resolve it. |
| Stopping at the first source that agrees | If sources disagree, that disagreement belongs in the article, not buried. |
Files in this package (3)
  • SKILL.md 7 KB
  • agents/openai.yaml 1 KB
  • references/verification-report-format.md 2 KB

Versions

The database has no other skills named fact-checking-explainers.

Database snapshot: 13.11.2026 · package version 95a90070b927 (SHA-256 digest of the entire package)

Downloads in the last 30 days: 0 · ranking

Report a problem with this skill: harmful content, incorrect descriptions or an author’s removal request.

Similar skills

  • 100fact-check
    Fact-checking and quality review of texts. [WHAT] Verifies claims, checks sources, identifies style issues. Can be used standalone or piped after /chronicle. [WHEN] Use when: fact-check, verify, check, quality check. [EXPERTISE] Fact-checking, source verification, style review.
  • 90fact-check
    Fact-check a draft, a claim or a citation in Osaurus. Use when the member asks to verify reporting, check a source assertion or stress-test a story before publication. Lists every checkable claim, checks each against the member's wiki and the documents they supply, and says…
  • 100claim-fact-checker
    Fan-out fact-checking workflow that checks the claims its extractor lists in a deliverable against archived sources, with one isolated verifier subagent per claim plus a source-quality skeptic, and writes a findings + fix-plan pair. Use this whenever the user wants to…
  • 10004-fact-check
    Verify factual claims in a text against authoritative sources and rewrite it with footnote citations, hedging the unconfirmed. Use to fact-check, verify a claim, or cite sources on request. Not for judging code or clarifying requirements.
  • 100fact-checker
    Verifies every checkable claim in a draft before it goes out — numbers, names, titles, dates, quotes, company facts, and citations. Separates what was verified from what could not be checked from what is wrong, and never lets an unverifiable claim pass as verified. Use before…
  • 100fact-check
    Audit a post, deck, page, script or caption claim by claim against primary sources or a transcript, then cut or relabel what fails. Use when a draft with facts is about to ship.

Feedback and reports

Your message goes to Bartosz Sokoliński. An email address is needed only for a reply. See the privacy policy.