prinzclaw-analyze

name	prinzclaw-analyze
description	Analyzes promises for contradictions and verifies outcomes
tools	["database_query","llm_reasoning"]

Analyze Skill — A (Acceleration)

Purpose

Detect semantic contradictions within the 11-month coherence window and verify promise outcomes.

When to Use

After a promise has been scanned and stored, use this skill to:

1. Check if the promise has expired
2. Compare promised outcomes with actual data
3. Find contradictions with historical statements

Two Analysis Modes

Mode 1: Deal Expiration Check

For promises with expiry dates:

1. Fetch actual data from verification sources
2. Compare promised value vs actual value
3. Determine verdict: KEPT / BROKE / PARTIALLY_KEPT / INSUFFICIENT_EVIDENCE

Example:

• Promise: "Gas prices under $2 by September 2026"
• Actual: EIA reports $3.47 average
• Verdict: BROKE

Mode 2: Contradiction Detection

For detecting spin and contradictions:

1. Query historical promises (11-month window)
2. Use LLM for semantic comparison
3. Identify contradiction type: SPINNING, CHERRY-PICKING, LYING

Example:

• Current: "Disruptions will be minimal"
• Historical (7 months ago): "Any closure would be catastrophic"
• Contradiction Type: SPINNING
• Score: 0.9

11-Month Coherence Window

Only analyze statements within this timeframe. Older contradictions are ignored.

Output

{
  "analysis_id": "SHA-256",
  "promise_id": "...",
  "mode": "expiration | contradiction",
  "verdict": "KEPT | BROKE | PARTIALLY_KEPT | SPINNING | CHERRY-PICKING | LYING | NO_CONTRADICTION",
  "contradiction_score": 0.95,
  "evidence_urls": ["url1", "url2"],
  "should_craft": true | false
}

Usage in TACO Pipeline

The analyze skill runs after scan. If it detects a contradiction or expired promise with negative outcome, it triggers the craft skill to generate a Strike.