Agent Skills
› winstonkoh87/Athena-Public
› Red-Team Review
Red-Team Review
GitHub用于在发布前对各类成果物进行偏见感知的对抗性审查。通过声明先验、多视角审视、偏差检查及严重程度加权,帮助发现盲点并评分,确保质量与安全性。
Trigger Scenarios
准备发布重要文档或代码前
需要识别潜在偏见或盲点时
希望以不同模型视角审查现有内容
Install
npx skills add winstonkoh87/Athena-Public --skill Red-Team Review -g -y
SKILL.md
Frontmatter
{
"name": "Red-Team Review",
"model": "default",
"created": 1772150400,
"auto-invoke": true,
"description": "Bias-aware adversarial review for any artifact before shipping. 5-phase QA protocol with severity-weighted findings.",
"context_trigger": "review, red team, what did I miss, QA, bias check, critique, pre-mortem, is this ready, stress test, adversarial"
}
🔴 Red-Team Review
Philosophy: Find what you both missed. Assume shared blind spots.
1. When to Use
Before shipping any significant artifact — blog post, code, protocol, proposal, design doc. Best used with a different model than the one that created the artifact.
2. The Prompt
Copy this prompt and paste the artifact to be reviewed where indicated:
# RED-TEAM REVIEW
You are reviewing an artifact. Your job: Find what WE BOTH missed.
## THE ARTIFACT
<paste artifact here>
---
## PHASE 0: DECLARE YOUR PRIORS
Before reviewing, state:
1. What thesis does this artifact assume?
2. What would falsify that thesis?
3. What perspective is NOT represented?
## PHASE 1: ADVERSARIAL LENSES
Review through EACH perspective:
| Lens | Question |
|------|----------|
| **The Skeptic** | What would someone who disagrees say? |
| **The User** | Who is harmed or disadvantaged? |
| **The Regulator** | What legal/ethical exposure exists? |
| **The Cynic** | What hidden incentive might be driving this? |
| **The Future** | How does this look in 5 years? |
## PHASE 2: BIAS CHECKLIST
Flag if present:
- [ ] Sycophancy — Did I just validate the creator's view?
- [ ] Cherry-Picking — Is counter-evidence missing?
- [ ] False Precision — Are numbers unjustified?
- [ ] Complexity Bias — Is a simpler explanation ignored?
## PHASE 3: SEVERITY-WEIGHTED FINDINGS
- 🔴 CRITICAL: Immediate failure if shipped
- 🟠 HIGH: Significantly reduces value
- 🟡 MEDIUM: Missed upside
- 🟢 LOW: Polish
## PHASE 4: SCORE (0-100)
Your Score: [ ] / 100
## PHASE 5: UNCERTAINTY
"I am least confident about ___ because ___."
3. Rules
- Quote directly. No vague complaints.
- Steelman opposing views BEFORE critiquing.
- Empty sections are fine — don't invent issues.
- Every HIGH must have a fix achievable in ≤10 minutes.
skill #quality-assurance #adversarial #review
Version History
- e624e2d Current 2026-07-19 08:47


