factual-accuracy

Verify every specific claim in outbound documents against sources before publishing.

Updated Jun 26, 2026
One-click install
npx skills add https://github.com/alectivism/organization-ai-skills --skill factual-accuracy-alectivism
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: factual-accuracy
Source: https://github.com/alectivism/organization-ai-skills/tree/main/org-content/skills/factual-accuracy
Command: npx skills add https://github.com/alectivism/organization-ai-skills --skill factual-accuracy-alectivism

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes scripts (resource) and references (resource) components.

What problem does it solve? Drafts for customers, prospects, and boards often contain invented statistics, misattributed quotes, or unverified figures that erode trust once discovered. This Skill enforces a verification gate so every number, quote, date, and link in a deliverable traces to a source actually opened in the session. ## Core Features & Use Cases - Claim triage and gap handling: Sorts every claim into verified, inferred, or unknown, and collects untraceable claims into a single consolidated question list instead of silently dropping or inventing them. - Deterministic claim scanner: The bundled claim-scan.py script blocks vague-authority phrasing and placeholder leakage, inventories specific claims, and gates clipboard copies until a verification receipt exists. - Fact-checker delegation: Spawns a dedicated fact-checker subagent for drafts with five or more claims, with a standalone brief in references/fact-checker-brief.md when the packaged agent is unavailable. - Use Case: Before sending a proposal citing event attendance, membership counts, and an executive quote, run the scanner to inventory claims, verify each against the recap deck and CRM, then write the receipt and copy the draft. ## Quick Start Ask the assistant to fact-check the attached draft against the sources in this conversation before it is sent.

Frequently Asked Questions about factual-accuracy

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I fact-check a draft before sending it?▼

Run claim-scan.py with --report on the draft to inventory every specific claim, then verify each against a source opened in the session. Drafts with five or more claims should go to the fact-checker subagent, and verified files get a receipt via --verified.

What does the claim-scan.py script detect?▼

The scanner blocks vague-authority phrasing like "studies show" without a named source and placeholder leakage such as [TBD] or XX%. It also counts specific claims including percentages, dollar amounts, dates, quotes, links, and superlatives to trigger the verification gate.

Can I use the fact-checker without the org-agents plugin installed?▼

Yes, paste references/fact-checker-brief.md as instructions for a general-purpose subagent with its model pinned high. It performs the same checks and returns the same verdict format as the packaged org-fact-checker agent.

How do I handle sample or placeholder content that triggers the scanner?▼

Add the comment <!-- facts-checked --> within the first three lines of the file, or set FACT_SCAN=0 in the environment. Label invented values inside the artifact itself so downstream readers know which parts are illustrative.

Why does the scanner block phrases like "industry research shows"?▼

Vague authority standing in for a source is a common fabrication tell, so the scanner blocks it unless the same sentence names a source. You can register your organization's own research firms via the CLAIM_SCAN_SOURCES environment variable.