EFC Standalone — Fact-Check Skill for AI Agents

One file. Zero dependencies. Any AI agent.

Drop SKILL.md into your agent's skill directory and it gains a rigorous fact-checking protocol — catching hallucinated numbers, fabricated data points, and exaggerated claims in AI-generated research reports.

What it does

EFC (Everything Fact-Checked) gives your agent a structured 6-step workflow to audit reports:

1. Inventory every specific, checkable claim
2. Triage by risk (P0 critical → P3 cosmetic)
3. Verify P0/P1 claims against primary sources
4. Cross-check charts and tables for consistency
5. Audit source links and attribution chains
6. Report with verdicts, evidence, and a reliability rating

It's trained to catch the five systematic failure modes that LLMs produce at scale:

Failure Mode	Example
Unit/scale errors	$5.3B → $530M (dropped conversion)
Fabricated interpolation	6-point chart where only 2 were sourced
Source conflation	GMV reported as revenue
Stale data as current	2023 figures presented as 2025
Attribution laundering	Blog cited as regulatory filing

Requirements

• Best results: Agent with web browsing capability, so it can open cited URLs and verify claims against source content.
• Minimum: Agent that can read the document. Without browsing, the skill degrades to an internal consistency review (cross-checking figures within the document, verifying table totals, flagging unsupported claims). All external claims are marked "unverifiable."
• No code, no install, no API keys. This is a protocol, not a script.

Install

The skill name is fact-check. Install it under a folder with the same name:

Pi Agent

mkdir -p ~/.pi/agent/skills/fact-check
cp SKILL.md ~/.pi/agent/skills/fact-check/SKILL.md

Claude Code

mkdir -p .claude/skills/fact-check
cp SKILL.md .claude/skills/fact-check/SKILL.md

OpenClaw / Hermes / Other agents

Place SKILL.md in <agent-skill-dir>/fact-check/SKILL.md. The file is self-contained — no scripts, no schemas, no dependencies.

For agents that don't support skill files: copy the entire content of SKILL.md into your system prompt or custom instructions.

Example

See examples/sample-report.md — a deliberately flawed fictional report with one of each failure mode — and examples/expected-fact-check.md for the target output.

Why a standalone version?

The full EFC-Plugin repo includes Python scripts, JSON schemas, a CLI (efc), a GitHub Action, and a Claude Code plugin. That's great for CI pipelines and automated workflows.

But most AI agents just need the protocol — the structured thinking about what to check and how. That's what this is: one Markdown file that teaches any agent to be a rigorous fact-checker.

	EFC-Plugin	EFC-Standalone
Files	25+	1 (SKILL.md)
Dependencies	Python 3.11+	None
CLI	efc command	No
CI integration	GitHub Action	No
Schema enforcement	JSON Schema + validator	No
Core protocol	✅	✅
Agent compatibility	Claude Code	Any

License

MIT — use it, fork it, ship it.

Related

• Full plugin + CLI: EFC-Plugin — CLI, schemas, CI, GitHub Action
• Feedback / Issues: Open an issue on either repo