AgentSkillTree
compositeUnverifiedmedium autonomyeasy setup

Fact Verification

Check every load-bearing claim against a named source and flag anything unverifiable.

Scope
Verification of claims in a given document or draft. Not research generation.
Last reviewed
Checked Jul 31, 2026

Executes these capabilities

Inside this skill

The internal capability tree shows how this skill gets its work done.

SFact Verification
  • claim extraction
  • source lookup
  • cross-checking
  • source quality rating

The contract

Input

  • A draft containing factual claims
  • Optional candidate sources

Output

  • Claim-by-claim verification table
  • Flagged unverifiable claims

Deliverables

  • Verification table with check dates

Use cases

  • Pre-publish content check
  • Reviewing AI-generated drafts
  • Auditing existing pages

Dependencies and replacements

Dependencies

  • Web access

Required tools

  • A browser

Required permissions

  • Web access

Compatible with

  • seo-content-workflow
  • validation-checklist

Alternatives

  • Human fact-checking
  • Specialized verification tools

Replaceable components

  • The source list
  • The verification table format

Reliability

Human checkpoints

  • Human decides what happens to flagged claims

Limitations

  • Cannot verify opinions presented as facts
  • Source quality judgments are partly subjective
  • Fast-changing topics need recent sources

Failure modes

  • Marking claims verified from weak sources
  • Verifying the statement instead of the source
  • Missing implied claims in the text

Recovery steps

  • Require at least two independent sources for key claims
  • Check the source actually says what the claim says
  • Read the draft for implied factual claims, not just explicit ones

Validation methods

  • A sample of verified claims is audited by a second pass

Sources

Entry checked on Jul 31, 2026. Verification status reflects what AgentSkillTree has independently run, not the skill's own claims.

When to use

Use on any draft that will be published or used for decisions. The output is a table: claim, source, check date, and verdict, so nothing depends on the agent’s memory.