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。
figma-agent
Undeclared scanning of Claude Code credential store
flyai-transit-tour
Undeclared shell execution in workflow
Memory Workflow
Undeclared LLM data transmission
Novai360 智能市场分析
Undeclared network access to third-party API
doctor-check
API key validation method unspecified
xclaw-skill
Undocumented private key storage in plaintext
memory-compactor
Documentation-only skill with unverifiable behavior
onetrust
Third-party credential proxy without transparency
blood-pressure-therapy
Undeclared external URL references
PathClaw
Hardcoded External IP Address
authlock
Shell command injection vulnerability in --exec
claw-wallet
Unsigned closed-source binary execution without integrity verification
edge
Undeclared shell execution via npx spawn
skill-state-manager
Credential Harvesting Framework
video-to-text
Undeclared subprocess execution via execSync
youdaonote
Dangerous curl|bash installation pattern documented