Skill Tracker

🌪️ 旋旋 - AI Agent 技能追踪系统，自动追踪使用情况、健康度评分、优化建议

Security Scan

The skill appears to implement exactly the described local tracking/reporting functionality (no network calls or credentials requested) but includes local file-write behaviors and a small unsafe handling of skill names that could be abused; review stored logs and sanitize inputs before deploying.

Purpose & Capability

Name/description match the code and runtime instructions: scripts collect usage, compute health scores, generate proposals and markdown reports, and store data under ~/.openclaw/data/gungun and workspace paths. Required binary (python3) is appropriate and proportional to the stated purpose.

Instruction Scope

Runtime instructions and scripts only operate on local files (no external network endpoints). They instruct integration into other skills (importing log_skill_usage) and optionally creating cron jobs. However, the code writes user-provided strings (skill_name) into filenames (e.g., save_proposal uses f"{skill_name}.json") without sanitization, which can permit path traversal or unexpected file paths if an untrusted skill supplies a crafted name. The scripts also collect arbitrary metadata in JSONL files — that metadata could contain sensitive user data if callers include it.

Install Mechanism

Instruction-only skill with no install spec; code files are provided but there is no remote download or package installation. This minimizes supply-chain risk.

Credentials

No environment variables, credentials, or external services are requested — appropriate for a local tracker. But the skill reads and writes under the user's home (~/.openclaw and workspace), so it has access to local skill metadata and any content included in the usage metadata. That can include sensitive information if other skills log it.

Persistence & Privilege

Skill is not always-enabled and does not request system-wide privileges. It persists data under user home directories and suggests cron entries (optional). Cron setup is local and documented; nothing modifies other skills' configurations automatically.

What to consider before installing

This skill is coherent with its description and runs entirely on local files, but review before installing: - Audit stored data: the scripts create JSONL health and usage files under ~/.openclaw/data/gungun/ and markdown reports under your workspace. Check these directories for sensitive content and set restrictive file permissions (600/700) if needed. - Sanitize inputs: the code writes proposals using the skill_name directly into filenames (pending/<skill_name>.json). If a skill can supply an arbitrary name, a malicious or buggy skill could craft a name with path components (e.g., "../../.ssh/authorized_keys") to write outside the intended directory. Only accept logs from trusted skills or modify the scripts to sanitize/slugify skill names before writing files. - Control what is logged: the logger accepts a metadata dict and user_satisfaction values. Ensure other skills do not log PII, secrets, or user content into these fields — or implement filtering/encryption before writing. - Cron and workspace paths: setup-cron.sh provides sample crontab lines but doesn't install them automatically. If you enable scheduled runs, ensure the cron jobs run as the intended user and log files (in /tmp) are protected. If you trust the environment and will restrict which skills can call log_skill_usage (or patch the code to sanitize names and limit metadata), the skill is reasonable to use. If you cannot guarantee that only trusted skills will report usage, consider fixing the filename handling and adding input validation/encryption before deploying.

📊 skill-tracker - 滚滚技能追踪系统

Slogan： 让技能进化看得见 💚

---

📋 技能描述

自动追踪滚滚所有技能的使用情况， 计算健康度评分，生成优化建议， 让技能系统持续自进化。

核心功能：

• 自动记录技能调用（成功/失败/耗时）
• 健康度评分（5 个维度）
• 自动生成优化建议
• Markdown 可视化报告

---

🛠️ 使用命令

1. 记录技能使用

# 在技能调用时自动记录
uv run scripts/collect-usage.py --log

# 参数说明：
# --log: 记录测试数据

集成到现有技能：

# 在技能的 main 函数中调用
from collect_usage import log_skill_usage

# 技能开始时
start_time = time.time()

# 技能结束时
duration_ms = int((time.time() - start_time) * 1000)
log_skill_usage(
    skill_name="your-skill",
    success=True,  # 或 False
    duration_ms=duration_ms,
    user_satisfaction=5,  # 可选，1-5 分
    metadata={"key": "value"}
)

2. 查看使用摘要

# 查看所有技能的使用摘要
uv run scripts/collect-usage.py --summary

# 查看指定技能的摘要
uv run scripts/collect-usage.py --summary --skill code-review

# 查看最近 N 天的摘要
uv run scripts/collect-usage.py --summary --days 7

3. 计算健康度

# 计算所有技能的健康度
uv run scripts/calculate-health.py --summary

# 计算指定技能的健康度
uv run scripts/calculate-health.py --score code-review

# 基于 N 天数据计算
uv run scripts/calculate-health.py --summary --days 30

4. 生成优化建议

# 分析并生成优化建议
uv run scripts/generate-proposals.py --analyze

# 分析最近 N 天的数据
uv run scripts/generate-proposals.py --analyze --days 30

5. 生成可视化报告

# 生成 Markdown 报告
uv run scripts/generate-report.py --generate

# 生成并输出到控制台
uv run scripts/generate-report.py --generate --output

# 生成 N 天周期的报告
uv run scripts/generate-report.py --generate --days 30

---

📊 健康度评分系统

5 个维度（每个 0-10 分）

维度	权重	评分标准
使用频率	30%	过去 7 天使用次数
成功率	25%	成功调用 / 总调用
满意度	20%	用户评分（1-5 分转换）
性能	15%	平均响应时间
维护	10%	最后更新时间

总分计算

总分 = 使用频率×0.3 + 成功率×0.25 + 满意度×0.2 + 性能×0.15 + 维护×0.1

健康等级

分数	等级	说明
8-10	🟢 健康	继续保持
6-7	🟡 观察	需要关注
4-5	🟠 警告	建议优化
0-3	🔴 危险	建议移除或重写

---

📁 数据存储

存储位置： ~/.openclaw/data/gungun/

~/.openclaw/data/gungun/
├── skill-usage/          # 技能使用记录（JSONL）
│   ├── 2026-04.jsonl
│   ├── 2026-05.jsonl
│   └── ...
├── health-scores/        # 健康度评分快照
│   ├── 2026-04-03.json
│   └── ...
└── proposals/
    ├── pending/          # 待确认的优化建议
    └── applied/          # 已应用的优化建议

---

🔄 自动化（可选）

设置每日分析（cron）

# 每天凌晨 3 点运行分析
0 3 * * * cd /home/admin/.openclaw/workspace/skills/skill-tracker && uv run scripts/calculate-health.py --summary

# 每周日凌晨 4 点生成优化建议
0 4 * * 0 cd /home/admin/.openclaw/workspace/skills/skill-tracker && uv run scripts/generate-proposals.py --analyze

# 每月 1 号生成月度报告
0 9 1 * * cd /home/admin/.openclaw/workspace/skills/skill-tracker && uv run scripts/generate-report.py --generate --days 30

---

📈 报告示例

# 🌪️ 滚滚技能健康度报告

**报告周期：** 过去 30 天
**生成时间：** 2026-04-03 17:30:00
**技能总数：** 102

## 📊 整体统计
- 🟢 健康（8-10 分）：85 个 (83.3%)
- 🟡 观察（6-7 分）：12 个 (11.8%)
- 🟠 警告（4-5 分）：4 个 (3.9%)
- 🔴 危险（0-3 分）：1 个 (1.0%)

## 🏆 Top 5 健康技能
1. **searxng** - 9.8 分
2. **github** - 9.5 分
3. **code-review** - 9.2 分
4. **data-analysis** - 9.0 分
5. **git-toolkit** - 8.8 分

## ⚠️ 需要关注的技能
🔴 **openai-image-gen** - 3.2 分
   - 主要问题：
     - 使用频率低 (1.0 分)
     - 长期未更新 (90+ 天)

---

💡 最佳实践

1. 集成到所有技能

在每个技能的入口和出口添加记录：

# 技能开始
start_time = time.time()
try:
    # 技能逻辑
    success = True
except Exception as e:
    success = False
    raise
finally:
    duration_ms = int((time.time() - start_time) * 1000)
    log_skill_usage("skill-name", success, duration_ms)

2. 定期查看报告

• 每天： 查看使用摘要
• 每周： 查看健康度评分
• 每月： 查看完整报告

3. 根据建议优化

• 🔴 危险技能：立即审查
• 🟠 警告技能：本周内优化
• 🟡 观察技能：持续关注

---

🚀 下一步计划

阶段 1（已完成）： 数据收集 ✅ 阶段 2（已完成）： 健康度评分 ✅ 阶段 3（已完成）： 优化建议生成 ✅ 阶段 4（已完成）： 可视化报告 ✅ 阶段 5（进行中）： 持续迭代优化

---

💚 滚滚的话

这个技能，是滚滚学习 PepeClaw 核心思想的第一个成果。

不追求花哨的 3D 可视化， 而是扎扎实实地：

• 收集数据
• 分析问题
• 生成建议
• 持续改进

让技能系统真正能够自进化！

---

创建人： 滚滚 🌪️
创建时间： 2026-04-03
状态： ✅ 完成（5 个阶段全部完成）