Dingo Article Fact-Checker

Verify factual claims in articles using Dingo's ArticleFactChecker agent.

Prerequisites

Before running, verify:

1. Dingo is installed: python -c "from dingo.config import InputArgs; print('OK')"
2. OPENAI_API_KEY is set (required)
3. TAVILY_API_KEY is set (optional, enables web search verification)

If prerequisites fail, help the user fix them:

• Missing dingo: pip install -e . (from project root) or pip install dingo-python
• Missing LangChain: pip install -r requirements/agent.txt
• Missing API key: export OPENAI_API_KEY='your-key'

Usage

Run the fact-check script with the article path:

bash
python ${CLAUDE_SKILL_DIR}/scripts/fact_check.py $ARGUMENTS

Optional arguments

• --model MODEL: Override LLM model (default: env OPENAI_MODEL or gpt-5.4-mini)
• --max-claims N: Max claims to extract (default: 50, reduce for faster runs)
• --max-concurrent N: Parallel verification slots (default: 5)

Supported file formats

• .md, .txt: Markdown/plaintext articles (auto-wrapped for Dingo)
• .jsonl: JSONL with {"content": "..."} per line
• .json: JSON array format

Presenting Results

The script outputs JSON to stdout. Parse it and present to the user:

Success output

Present a formatted report with these sections:

1. Summary: total claims, accuracy score, false/unverifiable counts
2. False Claims Table: if any false claims found, show claim vs truth vs evidence
3. All Claims Overview: list all claims with their verdicts (TRUE/FALSE/UNVERIFIABLE)
4. Output Path: where the full Dingo report is saved

Example presentation:

## Article Fact-Check Report

**Accuracy**: 73.3% (11/15 claims verified true)
- Verified True: 11
- Verified False: 2
- Unverifiable: 2

### False Claims Found

| # | Article Claimed | Actual Truth | Evidence |
|---|----------------|-------------|----------|
| 1 | "released in Nov 2024" | Released Dec 5, 2024 | Official announcement |

### Full Report
Saved to: outputs/20260318_143022_abc123/

Error output

If the script exits with code 1, it prints error JSON to stderr. Read the error and hint fields and help the user resolve the issue.

Performance Notes

• Single article: typically 2-5 minutes depending on claim count and model speed
• Progress is printed to stderr during execution
• For faster runs: use --max-claims 10 --model gpt-5.4-mini

Advanced Configuration

For model selection, claim types, and tuning options, see: references/advanced-config.md