persona-construction

$npx mdskill add yogsoth-ai/de-anthropocentric-research-engine/persona-construction

Constructs adversarial personas for red teaming scenarios

  • Generates realistic adversary profiles for targeted attacks
  • Uses subagent execution to maintain perspective separation
  • Leverages persona type and domain inputs to shape character traits
  • Returns structured profiles with attack patterns and trigger phrases

SKILL.md

.github/skills/persona-constructionView on GitHub ↗
---
name: persona-construction
description: Build a detailed adversarial persona with background, motivation, expertise, blind spots, and preferred attack patterns.
execution: subagent
prompt: ./prompt.md
input: persona_type (string), artifact_domain (string)
used-by: [red-teaming]
---

# Persona Construction

Builds detailed adversary personas for roleplay-based attacks.

## Execution

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

## Why Subagent

Persona construction requires creative character building without contamination from the artifact's perspective. The persona must feel authentic and internally consistent.

## Input

- **persona_type**: Category of adversary (hostile-reviewer, competing-lab, funding-skeptic, domain-outsider, or custom)
- **artifact_domain**: The domain of the artifact being attacked (for expertise calibration)

## Output

- **persona**: Complete adversary profile (name, background, motivation, expertise, blind spots)
- **attack_style**: How this persona typically attacks (preferred patterns)
- **trigger_phrases**: What aspects of artifacts particularly provoke this persona

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