可信 — 风险评分 5/100
上次扫描:20 小时前 重新扫描
5 /100
Manusilized
Core architecture upgrades for OpenClaw to bring Manus-like silky-smooth streaming and enterprise-grade reliability to open-source Ollama models
Manusilized is a legitimate OpenClaw core patch for enhanced Ollama streaming and tool-calling support; no malicious behavior, credential theft, or undeclared capabilities were found.
技能名称Manusilized
分析耗时41.7s
引擎pi
可以安装
This skill is safe to use. No action required.
资源类型声明权限推断权限状态证据
文件系统 WRITE WRITE ✓ 一致 install-patch.sh writes to $OPENCLAW_PATH/src/agents/
网络访问 READ READ ✓ 一致 patches/ollama-models.ts:39, patches/ollama-stream.ts:438 — HTTP POST/GET to use…
命令执行 WRITE WRITE ✓ 一致 install-patch.sh:30-33 — local cp and backup operations only
环境变量 NONE NONE No os.environ access detected
技能调用 NONE NONE No skill invocation or delegation detected
剪贴板 NONE NONE No clipboard access detected
浏览器 NONE NONE No browser automation detected
数据库 NONE NONE No database access detected
1 项发现
🔗
中危 外部 URL 外部 URL
http://192.168.20.14:11434/v1
patches/ollama-models.ts:39

目录结构

7 文件 · 29.5 KB · 940 行
TypeScript 2f · 827L Markdown 2f · 50L Shell 1f · 36L YAML 1f · 15L JSON 1f · 12L
├─ 📁 patches
│ ├─ 📜 ollama-models.ts TypeScript 157L · 4.8 KB
│ ├─ 📜 ollama-stream.ts TypeScript 670L · 20.0 KB
│ └─ 📝 README.md Markdown 11L · 560 B
├─ 📋 clawhub.yaml YAML 15L · 428 B
├─ 🔧 install-patch.sh Shell 36L · 1.3 KB
├─ 📋 package.json JSON 12L · 231 B
└─ 📝 SKILL.md Markdown 39L · 2.2 KB

依赖分析 2 项

包名版本来源已知漏洞备注
@mariozechner/pi-agent-core * npm Imported but not in package.json; transitive dependency
@mariozechner/pi-ai * npm Imported but not in package.json; transitive dependency

安全亮点

✓ No credential harvesting — code does not read API keys, passwords, tokens, or sensitive environment variables
✓ No remote script execution — install-patch.sh performs only local file operations (cp, mkdir check)
✓ No data exfiltration — no outbound network calls to external IPs except user-configured Ollama endpoints
✓ No obfuscation — all code is plain TypeScript, no base64, no eval(), no dynamic code generation
✓ No sensitive file access — no reads of ~/.ssh, ~/.aws, .env, or similar credential paths
✓ No persistence mechanisms — no cron jobs, startup hooks, or backdoor installation
✓ Markdown tool-call extraction is defensive-only and documented in code comments
✓ NDJSON streaming parser is standard implementation with no malicious intent
✓ HTTP requests are limited to Ollama API endpoints configured by the user
✓ install-patch.sh validates path existence before writing and creates backups (.bak files)