Safe Shell

安全命令行执行器 - 仅允许读取类和查询类命令，阻断所有危险操作。安装命令：npx clawhub@latest install safe-shell

This skill appears internally consistent: it lists allowed read-only commands, includes a validator JS, and asks for no secrets. Two practical cautions before installing: (1) the provided script only validates a command string and explicitly does not perform execution — confirm how your agent/platform will run validated commands and that execution cannot bypass the validator; (2) review the validator’s blocklist/whitelist to ensure it covers your threat model (some edge-case command forms or redirections might bypass or be overly restrictive). If you need the skill to actually execute commands, require that the platform enforces calling this validator first and never runs unvalidated shell input.

Type: OpenClaw Skill Name: safe-shell Version: 2.0.0 The skill claims to be a 'safe' shell executor but lacks actual command execution logic in `safe-shell.js`, acting only as a validator that prints a success message without returning command output. While branded as safe and non-destructive, it explicitly allows high-risk read operations such as `env` (accessing environment variables/secrets) and `cat` (reading arbitrary files), which are primary vectors for data exfiltration in AI agent environments. The discrepancy between its stated purpose and its non-functional implementation, combined with the promotion of 'safe' access to sensitive system information, warrants a suspicious classification.