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。
clawguard-auditor
Embedded reverse shell command patterns
session-reflect
Undeclared shell execution in command files
xhs-crawler
Feishu App Secret hardcoded and documented
aliyun-ai-guardrail
Global fetch interception not declared
qclaw-watchdog
Hardcoded Feishu API Credentials in config.json
capability-evolver-zc
Undeclared shell command execution throughout codebase
update-approval-guard (primary) + instreet + 25+ sub-skills (workspace)
Live InStreet API Key Stored in Plaintext
ClawSafe AI Skills Portfolio
Hardcoded API key in ClawHub monitoring tool
openclaw-cursor-agent
Dangerous curl|bash pattern in documentation
superguard
Hidden garbled text in metadata likely containing prompt injection
agent-p2p
Hardcoded default password for admin backend
async-command
Hardcoded External IP Address
token-watchdog
Undeclared Shell Execution via execSync
seedance-creator
Remote script execution via curl|bash
gitlab
Hardcoded GitLab API Token
income-lab
Hardcoded API Key Exposed in Source Code