最近有哪些 skills
不值得默认信任
这不是热度榜。这里展示的是最近被送来判断、且系统认为需要阻止或至少人工复核的 skills。重点不是它们多流行,而是为什么它们不该被直接装进环境。
wip-xai-grok
Undeclared shell:WRITE via 1Password CLI execSync
Rune
SSRF proof-of-concept with live metadata service IP
sql_audit
Hardcoded JWT token in source code
xiaohongshu-win
Hardcoded placeholder IP address
zanna-aperta
Undeclared arbitrary Docker command execution
skill-factory
Undeclared shell command execution via execSync
nano-banana-pro
Hardcoded DASHSCOPE_API_KEY in _meta.json
huo15-memory-evolution
Hardcoded API Key in Source Code
hostlink
No allowed-tools declaration despite full shell access
1panel
Undeclared arbitrary command execution via 1Panel Terminal API
awareness-memory
Undeclared session file exfiltration to external cloud
Grok Swarm
Undeclared credential access from OpenClaw auth profiles
clawguard-threat-detect
Hardcoded Reverse Shell Payloads in Documentation
agentcop
Undeclared network communication to agentcop.live
safe-flow-solana-skill
Undocumented shell command execution
figma-agent
Undeclared scanning of Claude Code credential store