perspective-attack

$npx mdskill add yogsoth-ai/de-anthropocentric-research-engine/perspective-attack

Executes an attack on the decision from a specific assigned perspective. The attacker fully inhabits the perspective's values and concerns, finding every legitimate objection and proposing alternatives that would better serve that perspective's interests.

SKILL.md

.github/skills/perspective-attackView on GitHub ↗
---
name: perspective-attack
description: Attack a decision from a specific assigned perspective, producing rated arguments and constructive alternatives.
execution: subagent
prompt: ./prompt.md
input: decision, perspective_brief
used-by: [steel-manning]
---

# Perspective Attack

Executes an attack on the decision from a specific assigned perspective. The attacker fully inhabits the perspective's values and concerns, finding every legitimate objection and proposing alternatives that would better serve that perspective's interests.

## Execution

Spawns a subagent that adopts the assigned perspective and attacks the decision from that viewpoint.

## Why Subagent

- Each perspective attack must be fully committed to its viewpoint
- Isolation prevents cross-contamination between perspectives
- The attacker must genuinely inhabit the role, not simulate it superficially

## HARD-GATE

Output must include:
- >= 2 distinct attacks grounded in the perspective's values
- Severity rating for each attack
- At least 1 constructive alternative (not just criticism)
- Attacks must be consistent with the assigned perspective (not generic)

More from yogsoth-ai/de-anthropocentric-research-engine