# Skills Judgment

> Judge, manage, and evolve the Hermes skills ecosystem using natural selection principles — with progressive loading for deep dives

- Skill: `zzyong24/skills-judgment` (Agent Skill, multi-file: 6 files)
- Install (CLI): `npx skillmds@latest add zzyong24/skills-judgment`
- Raw SKILL.md: https://api.skillmd.com/api/skills/zzyong24/skills-judgment/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- License: MIT
- Author: zzyong24 (https://skillmd.com/u/zzyong24)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/zzyong24/skills-judgment

---


# 🌲 Skills Judgment / 技能生态系统审判

<!-- ProTip: If you see this skill loaded, scan the system first before making any judgment.
     Never execute verdicts without explicit user approval. -->

## Language / 语言

> **Language auto-detection active.** This document adapts based on your environment.
>
> 当前语言检测：`{{ env.LANG | default(env.LC_ALL | default("en_US.UTF-8")) }}`
>
> - `en*` → English mode
> - `zh*` → 中文模式
>
> If detection fails, defaults to **English**. To force Chinese, set `LANG=zh_CN.UTF-8` in your environment.

---

## Quick Decision Tree / 快速决策树

```
用户说"开始审判"或类似语句？
        │
        ▼
Yes → Load this skill
        │
        ▼
Run: hermes tools run scripts/skills_judgment scan
        │
        ▼
Report generated → Show to user
        │
        ▼
User approves? ─────────────────┐
        │                       │
       Yes                      No
        │                       │
        ▼                       ▼
Execute all verdicts      Maintain status quo
   (documented)          (no changes)
```

---

## Core Metrics Dashboard / 核心指标看板

| Metric | 当前值 | 状态 | 行动阈值 |
|---------|--------|------|----------|
| 总 Skills 数 | `{{ stats.total_skills }}` | `{{ stats.status }}` | > 150 需分流 |
| 顶级掠食者 | `{{ stats.apex }}` | 🟢 健康 | — |
| 普通公民 | `{{ stats.citizen }}` | 🟢 健康 | — |
| 濒危物种 | `{{ stats.endangered }}` | 🟡 注意 | > 20 需干预 |
| 待处决 | `{{ stats.condemned }}` | 🔴 警戒 | > 5 立即审判 |
| Graveyard 总数 | `{{ stats.graveyard }}` | ⚪ 中性 | — |
| 可复活 | `{{ stats.revivable }}` | 🟢 好事 | — |

> **Note**: Metrics populate after first `scan`. If you see `{{` here, run `skills_judgment scan` first.

---

## Fitness Tiers / Fitness 分级

### Six Dimensions / 六维评估

| 维度 | 指标 | 计算逻辑 | 低分信号 |
|------|------|---------|----------|
| **活跃度 / Active** | `updated` + mtime | 有更新且 < 14 天 = 2 | 14 天无更新 |
| **关系度 / Relationship** | `related_skills` 图 | 活跃引用 +1, 死锁引用 +0.1 | 孤立无引用 |
| **完整度 / Completeness** | 5 个 metadata 字段 | 每字段 +1, 上限 5 | 缺 name/description/tags |
| **健壮度 / Robustness** | 4 个内容章节 | 每章节 +2, 上限 8 | 缺 Procedure/Pitfalls |
| **适应度 / Adaptability** | platforms + prereqs + env_vars | 每项 +1, 上限 3 | 无 OS/环境适配 |
| **竞争位 / Competition** | 冲突组内对比 | 组内完整性最高者 +2 | 组内最差 |

### Tier Thresholds / 分级阈值

| 分值 | 级别 | 行为 |
|------|------|------|
| > 12 | 顶级掠食者 / Apex Predator | 保护，不进入审判 |
| 7-12 | 普通公民 / Citizen | 正常存在，不干预 |
| 3-6 | 濒危物种 / Endangered | 进入 pending 观察 |
| < 3 | 待处决 / Condemned | 进入审判流程 |

---

## When to Use / 何时使用

### English / 英语触发
- "Start judgment" / "Judge skills"
- "Skills health check" / "Clean up skills"
- "Weekly judgment" / "Skills ecosystem review"

### 中文触发
- "开始审判" / "审判 skills"
- "检查 skills 生态系统"
- "skills 健康检查"
- "清理过时的 skill"
- "每周例行审判"

---

## Deep Dive — Progressive Loading / 渐进式加载

> **For detailed documentation, load specific references on demand.**
> **详细文档，按需加载：**

| Reference | 内容 | 何时加载 |
|-----------|--------|----------|
| `references/architecture.md` | 系统架构 + 数据流图 | 理解组件如何连接 |
| `references/fitness-guide.md` | 六维评估标准 + 阈值 | 分析某个 skill 为何得此分 |
| `references/verdict-guide.md` | 判决类型 + 执行机制 + 可逆性 | 准备执行判决 |
| `references/conflict-groups.md` | 冲突组分析 + 存活策略 | 处理竞争性 skills |
| `references/examples.md` | 来自生态系统的真实案例 | 从实际审判中学习 |

### How to Load / 如何加载

```python
# 加载架构参考
skill_view("skills-judgment", file_path="references/architecture.md")

# 加载 fitness 评估指南
skill_view("skills-judgment", file_path="references/fitness-guide.md")

# 加载判决执行指南
skill_view("skills-judgment", file_path="references/verdict-guide.md")

# 加载冲突组分析
skill_view("skills-judgment", file_path="references/conflict-groups.md")

# 加载案例研究
skill_view("skills-judgment", file_path="references/examples.md")
```

---

## Judgment Workflow / 审判工作流

```mermaid
flowchart LR
    A[User says "开始审判"] --> B[Run scan]
    B --> C[Generate Report]
    C --> D[Show to User]
    D --> E{User approves?}
    E -->|Yes| F[Execute Verdicts]
    E -->|No| G[Status Quo]
    F --> H{Verdict Type}
    H -->|Probation| I[Patch frontmatter]
    H -->|Exile| J[Move to .judgment/exiled/]
    H -->|Graveyard| K[Move to .graveyard/]
    H -->|Delete| L[Permanent deletion]
    I --> M[Graveyard Review]
    J --> M
    K --> M
    M --> N{7-day Review}
    N -->|Original exists| O[Revival possible]
    N -->|Replaced| P[Confirmed death]
    N -->|Orphan| Q[Unresolved]
```

---

## Verdict Reference / 判决参考

| Verdict | 适用分值 | 执行方式 | 可逆 | 风险 |
|---------|----------|---------|------|------|
| **Probation 缓刑** | 3-6 | 写入 frontmatter | ✅ | 低 |
| **Exile 流放** | 0-3 | 移动到 `.judgment/exiled/` | ✅ | 中 |
| **Graveyard 处决** | 0-3 | 移动到 `.graveyard/` | ✅ | 低 |
| **Delete 彻底删除** | 0-3 | `shutil.rmtree` | ❌ | 高 |

---

## CLI Commands / CLI 命令

```bash
# 完整扫描 + 报告生成
hermes tools run scripts/skills_judgment scan

# 扫描 + 输出到 ThirdSpace（供 cron 使用）
hermes tools run scripts/skills_judgment scan --output ~/ThirdSpace/.../report.md

# 列出所有审判分区状态
hermes tools run scripts/skills_judgment list

# 运行 Graveyard 复审
hermes tools run scripts/skills_judgment graveyard
```

---

## Pitfalls / 陷阱

### ✅ 应该做 / DO
- 总是先运行 `scan` 再做审判决定
- 总要等用户明确审批后再执行判决
- 总要在判决报告中记录推理过程
- 总要在 exile 前检查冲突组

### ❌ 不应该做 / DON'T
- 绝不能不审批就执行判决
- 绝不能删除有 `related_skills` 指向它的 skill
- 绝不能忽略 probation 中的 skill — 它们有 3 天改善窗口
- 绝不能跳过 Graveyard 复审 — 7 天是故意的反思周期

---

## Verification / 验证

任何审判会话后，验证系统健康：

```bash
# 验证各分区正确填充
hermes tools run scripts/skills_judgment list

# 验证扫描结果与实际相符
hermes tools run scripts/skills_judgment scan
# 对比"生态系统概览"数字与预期是否一致
```

---

## Next Steps / 加载后的下一步

1. **首次使用？** 读取 `references/architecture.md` 理解系统
2. **分析某个 skill？** 加载 `references/fitness-guide.md` 查看评分标准
3. **即将执行判决？** 加载 `references/verdict-guide.md` 了解可逆性
4. **冲突组决策？** 加载 `references/conflict-groups.md` 查看策略
5. **案例学习？** 加载 `references/examples.md` 查看真实案例

---

*Built without modifying Hermes source code. All intervention via file movement + frontmatter patching.*