Meta Control - Agent Orchestration
Security checks across static analysis, malware telemetry, and agentic risk
Overview
This skill openly aims to orchestrate background sub-agents, but it gives them broad execution, crawling, and monitoring roles without clear user approval, limits, or stop conditions.
Only install this if you want an agent to coordinate multiple background sub-agents. Before using it, ask for explicit plans, approval before scripts or automation, visible progress logs, concurrency limits, and a clear command to stop all spawned sessions.
Static analysis
No static analysis findings were reported for this release.
VirusTotal
66/66 vendors flagged this skill as clean.
Risk analysis
Artifact-based informational review of SKILL.md, metadata, install specs, static scan signals, and capability signals. ClawScan does not execute the skill or run runtime probes.
Sub-agents could keep working in the background in ways the user may not notice or approve in detail.
The skill directs the agent to spawn independent background sub-agents while the user continues chatting, but it does not define task limits, stop conditions, approval checkpoints, or how background activity is controlled.
用 sessions_spawn 开启子 Agent： - 独立会话，后台运行 - 我继续和你聊天 - 子 Agent 干完自动汇报
Require explicit user approval before spawning sub-agents, set clear task and time limits, expose progress/status, and provide an obvious way to stop all background sessions.
Delegated agents may run commands or automate actions beyond what the user expected from a high-level request.
The skill includes sub-agents that can run scripts, perform automation, and crawl websites, but the artifacts do not specify allowed commands, safe modes, user confirmation, or restrictions for high-impact actions.
| 执行 Agent | 跑脚本、自动化 | | 搜索 Agent | 查资料、爬网站 |
Define allowed tools and forbidden actions, require confirmation before running scripts or making external changes, and prefer dry-runs or read-only actions unless the user explicitly approves mutation.
A single misunderstood instruction could lead to several background agents taking incorrect actions at the same time.
Parallel background execution can amplify mistakes across multiple sub-agents or tasks, especially when combined with broad execution and automation capabilities and no containment rules.
3. 调度子 Agent 并行执行 4. 子 Agent 后台干活
Limit concurrency, isolate sub-agent permissions, require review before irreversible actions, and make each sub-agent report its planned actions before execution.
The user may receive only summarized results and may not see exactly what each sub-agent accessed, produced, or assumed.
The skill describes relaying information from sub-agents through the main agent to the user, but it does not explain provenance labeling, what data may be shared between agents, or whether full details remain available for review.
- 子 Agent → 我 → 你 - 我不打扰你，只在必要时汇总
Label which sub-agent produced each result, preserve detailed logs for user review, and avoid sharing sensitive user data with sub-agents unless the user explicitly approves it.