Installs into .claude/skills of the current project.
Are you the author of Perspective Attack?
Add the live security badge to your README. It updates with every re-scan.
[](https://www.skillsdirectory.com/skills/yogsoth-ai-perspective-attack)
---
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)
