cross-examination

$npx mdskill add yogsoth-ai/de-anthropocentric-research-engine/cross-examination

Identifies weaknesses in defender responses through targeted probing

  • Detects inconsistencies, logical gaps, and unsupported claims in defenses
  • Uses structured input from debates and artifacts for context
  • Analyzes defenses independently to avoid framing bias
  • Returns targeted follow-up questions and a verdict on defense validity
SKILL.md
.github/skills/cross-examinationView on GitHub ↗
---
name: cross-examination
description: Probes defender responses for inconsistencies, logical gaps, and unsupported claims. Acts as follow-up interrogation after initial defense.
execution: subagent
prompt: ./prompt.md
input: defenses (string), attacks (string), artifact (string)
used-by: [multiagent-debate]
---

# Cross-Examination

Probes defender responses for inconsistencies and logical gaps.

## Execution

Subagent — spawned via subagent-spawning/spawn-agent.

## Why Subagent

Cross-examination requires fresh analytical perspective on the defense — isolated context prevents anchoring to either attack or defense framing.

## Input

- **defenses**: Structured defenses from debate-defender
- **attacks**: Original attacks that prompted the defenses
- **artifact**: The artifact being debated (for reference)

## Output

- **probes**: List of follow-up questions targeting defense weaknesses
- **inconsistencies**: Contradictions found within or between defenses
- **unsupported_claims**: Defense claims lacking evidence
- **verdict_suggestion**: Whether defense held up under examination

## Budget

One unit = one cross-examination pass per round.
More from yogsoth-ai/de-anthropocentric-research-engine