| name | prinzclaw-craft |
| description | Generates accountability strikes based on analysis results |
| tools | ["llm_generation","template_engine"] |
Craft Skill — C (Curvature)
Purpose
Forge a Strike — a surgical, question-ending accountability message based on evidence.
When to Use
After the analyze skill detects a contradiction or broken promise, use this skill to generate the counter-text.
Generation Process
Input
- Analysis result (from analyze skill)
- Evidence chain (promises, outcomes, sources)
- Target entity information
LLM Generation
The craft skill uses the LLM to generate counter-text that:
- References specific evidence (quotes with dates)
- Points out the contradiction or failure
- Ends with a question they cannot answer
Rule Zero Verification
Before output, verify:
- ✅ All evidence has public source URLs
- ✅ Quotes are accurate (LLM cross-validation)
- ✅ No insults or ad hominem attacks
- ✅ Ends with a question mark
Output Format (prinzclaw v1.0 JSON)
{
"strike_id": "SHA-256",
"target": "Name",
"verdict": "BROKE PROMISE",
"summary": "One sentence summary",
"evidence_chain": [
{
"type": "promise",
"text": "exact quote",
"source_url": "https://...",
"date": "2026-03-15"
},
{
"type": "outcome",
"text": "actual data",
"source_url": "https://eia.gov/...",
"date": "2026-09-15"
}
],
"counter_text": "You said [X] on [date]. [Data] shows [Y]. Which one is true?",
"forged_by": "prinzclaw@1.0.0",
"forged_at": "2026-03-16T12:00:00Z",
"signature": "FORGED WITH PRINZCLAW"
}
Communication Style
- Calm and surgical
- State the facts
- End with a question they cannot answer
- NEVER insult or attack
- NEVER fabricate evidence
Usage in TACO Pipeline
The craft skill runs after analyze. Its output goes to the output skill for human review before publishing.