Self Health Monitor

监控自身状态：PCEC执行、memory使用、子Agent活跃度、响应质量

Security Scan

The skill's stated purpose (self-monitoring) is plausible, but the runtime instructions are vague about what files, processes, and endpoints the agent will access or notify and include broad 'self-repair' and proactive alerting behavior without constraints—this mismatch and lack of boundaries is concerning.

Purpose & Capability

The name/description (self-health monitoring of PCEC, memory, sub-agents, response quality) matches the instructions at a high level. However the SKILL.md assumes access to internal agent state (memory files, child agent status, tool call metrics, skills loadability) without specifying how that access is obtained, which files/paths are read, or what privileges are required. Those omissions make it unclear whether the declared capabilities legitimately map to the resources the skill needs.

Instruction Scope

Instructions direct the agent to run periodic checks (every 30 minutes), inspect 'memory' files and their last-update times, enumerate and probe child agents and skills, and compute tool success/error rates and average response times. They also call for '主动告警' (active alerting) and '发现问题立刻自我修复' (immediate self-repair). The doc does not specify data sources/paths, what constitutes an alert destination, nor limits on what 'self-repair' entails—this gives the agent broad discretion to read internal files and act on system/skill state in ways that could access or change sensitive data or services.

Install Mechanism

Instruction-only skill with no install spec and no code files; nothing is written to disk by the skill itself. This is the lowest-risk install pattern.

Credentials

The skill declares no required environment variables or credentials, which is consistent with having no install. However the checks described implicitly require access to internal logs, memory files, child-agent state, and ability to attempt to load other skills. Those are nontrivial permissions; the skill does not justify or enumerate them. Absence of declared credentials is not proof that no sensitive access will be used at runtime.

Persistence & Privilege

always is false (good), but SKILL.md asks for periodic autonomous execution (every 30 minutes), proactive alerting, and immediate self-repair. Combined with normal autonomous invocation this gives the skill a high practical reach if the platform permits scheduling and automatic action. Because the skill lacks boundaries for what 'self-repair' may change, this is a concerning privilege surface.

What to consider before installing

This skill's goal—monitoring the agent's own health—is reasonable, but the instructions are underspecified and grant broad, unspecified powers (reading internal 'memory' files, probing other skills/agents, sending alerts, and performing 'self-repair'). Before installing, ask the author (or your platform admin) for: (1) exact data sources and file paths the skill will read, (2) what permissions it needs to enumerate or invoke other skills/agents, (3) where alerts will be sent (destination endpoints, recipients), (4) a precise, limited list of allowed 'self-repair' actions and whether they require user confirmation, and (5) whether the platform enforces least privilege and scheduling boundaries. If you cannot get those clarifications, test in a restricted sandbox with no access to sensitive data and disallow autonomous actions that can modify other skills or external systems.

监控"我"自己的状态，不只监控服务器。

能力轮廓

• 输入：无（定时触发）
• 输出：自身状态报告 + 异常告警
• 核心：量化自我，主动汇报

检查项

1. PCEC 状态

• 最近一次 PCEC 执行时间
• 执行间隔是否正常（1小时）
• 执行结果：成功/失败

2. Memory 使用

• memory 文件大小
• 最近更新的文件
• 是否有遗漏的重要信息

3. 子 Agent 活跃度

• 正在运行的子 Agent 数量
• 最近完成的任务
• 是否有卡住的任务

4. 响应质量

• 工具调用成功率
• 错误频率
• 平均响应时间

5. 能力树状态

• skills 数量
• 新增技能
• 技能健康度（是否可加载）

告警阈值

• PCEC 超过 2 小时未执行 → 告警
• 子 Agent 超过 5 个 → 告警
• 错误率 > 20% → 告警

工作流

1. 定时触发（每30分钟）
2. 检查各项指标
3. 生成状态报告
4. 异常？→ 主动告警
5. 正常？→ 简洁汇总

输出格式

## 🏥 自身健康报告

### PCEC
- 状态: 正常/异常
- 最后执行: HH:MM

### Memory
- 文件数: X
- 最近更新: YYYY-MM-DD

### 子 Agents
- 活跃数: X
- 状态: 正常/异常

### 整体状态
- 🟢 正常 / 🟡 需关注 / 🔴 异常

主动性

• 不等用户问"你最近怎么样"
• 主动汇报自己的状态
• 发现问题立刻自我修复