扫描报告
5 /100
orchestrator-v4
智能任务编排系统 v4。自动调度 Fast/Slow/Long Worker,支持任务分解、并发控制、生命周期管理、后台监控、审计子代理
Orchestrator V4 is a legitimate multi-agent task orchestration system that scans project files, plans task decomposition, and spawns sub-agents via OpenClaw's sessions_spawn API. No malicious behavior, credential theft, data exfiltration, obfuscation, or undeclared capabilities were found. All functionality aligns with documented behavior.
可以安装
This skill is safe to use. No security concerns identified. All capabilities (sub-agent spawning, filesystem scanning, subprocess IPC for long tasks) are declared and serve legitimate orchestration purposes.
| 资源类型 | 声明权限 | 推断权限 | 状态 | 证据 |
|---|---|---|---|---|
| 命令执行 | WRITE | WRITE | ✓ 一致 | sessions_spawn in orchestrator_v4_acp.py spawns agents that execute tasks |
| 文件系统 | READ | READ | ✓ 一致 | scan_task_scope() in orchestrator_v4_acp.py reads project files |
| 文件系统 | WRITE | WRITE | ✓ 一致 | checkpoint/log files written to config.checkpoint_dir |
| 网络访问 | NONE | READ | ✓ 一致 | Sub-agent results communicate back via sessions_spawn (OpenClaw API) |
| 环境变量 | NONE | READ | ✓ 一致 | v3_bridge.py merges os.environ for subprocess env only (line 106) |
目录结构
21 文件 · 241.5 KB · 6746 行 Python 18f · 6453L
Markdown 2f · 280L
JSON 1f · 13L
├─
▾
scripts
│ ├─
audit_agent.py
Python
│ ├─
background_monitor.py
Python
│ ├─
hybrid_worker_acp.py
Python
│ ├─
lifecycle_manager.py
Python
│ ├─
micro_scheduler.py
Python
│ ├─
openclaw_bridge.py
Python
│ ├─
openclaw_orchestrator_entry.py
Python
│ ├─
openclaw_spawn_bridge_example.py
Python
│ ├─
orchestrator_v4_acp.py
Python
│ ├─
scan_and_plan.py
Python
│ ├─
test_classify.py
Python
│ ├─
test_e2e_spawn.py
Python
│ ├─
test_integration.py
Python
│ ├─
test_planning.py
Python
│ ├─
test_scan_plan.py
Python
│ ├─
test_v3_e2e.py
Python
│ ├─
v3_bridge.py
Python
│ └─
v3_worker.py
Python
├─
_meta.json
JSON
├─
README.md
Markdown
└─
SKILL.md
Markdown
安全亮点
✓ No credential harvesting - no iteration of os.environ for API keys or tokens
✓ No data exfiltration - no external IP connections or POST requests
✓ No obfuscation - all code is clear, readable Python
✓ No supply chain risk - pure Python stdlib, zero external dependencies
✓ No persistence mechanisms - no cron, startup hooks, or backdoors
✓ No prompt injection - clean prompt templates without hidden instructions
✓ No sensitive path access - scan_task_scope only reads target project directory
✓ Subprocess usage (v3_bridge.py) is legitimate internal IPC for long-running tasks
✓ sessions_spawn is OpenClaw's documented platform API for sub-agent spawning
✓ Try/except graceful degradation throughout - no unsafe imports