/extract-claims

Extract specific claims from a source document for the evidence chain.

Usage

/extract-claims <url> [--source-id <id>]

• •<url>: Source URL to fetch and analyze
• •--source-id <id>: Optional source ID (defaults to derived from URL)

Examples

/extract-claims https://www.anthropic.com/engineering/claude-code-best-practices
/extract-claims https://arize.com/blog/claude-md-optimization-study --source-id arize-study

Process

Step 1: Source Admission Check

Before extracting, validate source meets admission criteria.

Admission threshold: weight >= 0.4

See evidence-chain.md for weight table and auto-reject criteria.

If rejected: Stop. Do not proceed to claim extraction.

Step 2: Extract Claims

1. •Fetch source content from URL
2. •Identify enforceable claims:
   • •Specific recommendations ("Keep under 300 lines")
   • •Thresholds ("No more than 5 emphases")
   • •Patterns ("Use @imports for...")
   • •Anti-patterns ("Don't include style rules")

Step 3: Output YAML

claims:
  - id: claim-id
    text: "The specific claim"
    quote: true
    section: "Section Name"
    rules: [S1, C2]

Reference

• •Evidence chain - Admission criteria and weights

Quick Reference

Good claims:

• •"Keep CLAUDE.md under 300 lines"
• •"Use @imports for documentation"
• •"5 IMPORTANT markers maximum"

Bad claims:

• •"Has advice about file size"
• •"Talks about imports"
• •"Mentions emphasis"

Rule mapping: