04-fact-check — Claude Code skill: what it does, how to install
04-fact-check verifies factual statements against authoritative sources and rewrites prose with citations, hedging claims that lack proof. Not meant for judging code.
MIT ★ 571 updated 2026-10-29
/plugin marketplace add ai-driven-dev/framework
/plugin install aidd-refine@aidd-frameworkWhat 04-fact-check does
The skill processes target prose through three linked steps. It strips opinions to isolate verifiable facts, runs a cheapest-first check cascade, and rewrites the passage with numbered citations. Web searches stay strictly a last resort, triggered only after project memory and repository files fail to substantiate a claim.
Unconfirmed statements remain in the output with an explicit (unverified - no source found) tag rather than disappearing. If sources dispute each other, the skill surfaces both sides alongside their origins instead of picking a winner.
The final deliverable stays clean: only the revised draft, a ## Sources section, and a ## Unverified claims block. Internal classification labels and tier traces are stripped from the final prose.
What it helps with
- audit factual statements across draft documentation or summaries
- attach footnote citations to every validated claim in a draft
- highlight disagreements between conflicting reference sources
- flag unverified claims explicitly without dropping them from context
Example prompts for the agent
- Fact-check this project summary and cite authoritative references:
summary.md - Verify the claims in this draft against the repo and flag unconfirmed items
- Run fact verification on this text and rewrite it with footnote citations
How to install 04-fact-check
1Plugin marketplace
/plugin marketplace add ai-driven-dev/framework/plugin install aidd-refine@aidd-frameworkThis installs the whole aidd-refine plugin with all its skills
2Skills CLI (Vercel)
npx skills add ai-driven-dev/framework --skill 04-fact-check -a claude-code3Manually: put the files in place
~/.claude/skills/04-fact-check/SKILL.mdCopy the whole skill folder; for one project use .claude/skills/04-fact-check/
1Skills CLI (Vercel)
npx skills add ai-driven-dev/framework --skill 04-fact-check -a codex2Manually: put the files in place
.agents/skills/04-fact-check/SKILL.mdFor all projects use ~/.agents/skills/
1Skills CLI (Vercel)
npx skills add ai-driven-dev/framework --skill 04-fact-check -a cursor2Manually: put the files in place
.cursor/skills/04-fact-check/SKILL.mdCursor also reads .agents/skills/ and .claude/skills/
1GitHub CLI
gh skill install ai-driven-dev/framework plugins/aidd-refine/skills/04-fact-check2Manually: put the files in place
.github/skills/04-fact-check/SKILL.md1Skills CLI (Vercel)
npx skills add ai-driven-dev/framework --skill 04-fact-check -a gemini-cli2Platform CLI
gemini skills install https://github.com/ai-driven-dev/framework --path plugins/aidd-refine/skills/04-fact-checkManually: put the files in place
.devin/skills/04-fact-check/SKILL.mdThe legacy .windsurf/skills/ path still works
Skills CLI (Vercel)
npx skills add ai-driven-dev/framework --skill 04-fact-check -a opencodeClaude.ai
Zip the folder and upload it in Settings → Capabilities → Skills
Step by step, with commands for each platform: How to install skills in Claude Code: five ways
Questions about 04-fact-check
How to install 04-fact-check in Claude Code?
Run npx skills add ai-driven-dev/framework --skill 04-fact-check -a claude-code in your terminal.
What does 04-fact-check do?
It identifies factual assertions in text, checks them across project records and external data, then delivers a rewritten response with source footnotes.
What happens when two sources disagree?
The skill reports both conflicting accounts together with their origins instead of quietly selecting one.
Source text
SKILL.md · MIT
--- name: 04-fact-check description: 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. argument-hint: text --- # Fact-check Verifies the factual claims inside a target text and rewrites it grounded in evidence. The skill extracts each verifiable claim, runs a cheapest-first verification cascade, and emits a rewritten answer where every confirmed claim carries a footnote citation and every unconfirmed claim is explicitly hedged. When sources disagree, both are reported rather than one being silently chosen. ## Actions | # | Action | Role | Input | | --- | ----------------- | ----------------------------------------------------------------------------- | --------------------------- | | 01 | `identify-claims` | Extract verifiable factual claims from the text, classify each, drop opinion | target text | | 02 | `verify` | Run the verification cascade per claim, produce a verdict and sources | claim list from 01 | | 03 | `report` | Rewrite the text with footnote citations, hedge unverified, surface conflicts | verdict list from 02 | Run `01 → 02 → 03`; pass each `## Test` before the next. Before running an action, read its file in `actions/`, not only the table or assets. ## Transversal rules - Never alter the meaning of a claim while verifying it: verify what was stated, not a charitable reinterpretation. - The verification cascade is cheapest-first and short-circuits: stop at the first tier that resolves a claim. See `references/verification-cascade.md`. - A web lookup is the last resort, never the first. Reach it only when project memory and codebase inspection both fail to resolve a claim. - Claim categories come from the locked set in `references/claim-categories.md`. Opinion, preference, and trivially-known statements are not claims and are skipped. - When two sources disagree, report both with their origin, never silently pick one. - An unverified claim is never deleted and never asserted as fact: it is kept and marked `(unverified - no source found)`. - Caching a verified fact is opt-in: propose it with a recommendation, never cache silently. The skill persists nothing; the mechanics live in action 03. - The final report is reader-facing prose: the corrected text, `## Sources`, and `## Unverified claims`, nothing else. Internal mechanics never appear in the output: no cascade or tier trace (`Cascade:`, `tier 1/2/3`, `miss`, `resolved`), no category labels, no raw verdict words. State conclusions, not the process. Action 03 holds the exhaustive forbidden list. - The report is rendered in plain prose and is never restyled by an active output mode. The skill's output format is fixed by action 03 alone. ## References - `references/claim-categories.md`: locked taxonomy of verifiable claim categories with definition and example. - `references/verification-cascade.md`: the three-tier cascade, short-circuit rule, web-cost guardrail. ## Assets - `assets/report-template.md`: rewritten-answer skeleton with the `## Sources` footnote block.
© AI-Driven Dev · MIT · ai-driven-dev/framework