The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The instructions guide the user to download shell scripts from a third-party repository and execute them immediately using a pipe to the bash interpreter, which bypasses verification and allows for arbitrary code execution.
Evidence: curl -fsSL https://raw.githubusercontent.com/miaoxworld/OpenClawInstaller/main/install.sh | bash in SKILL.md.
Evidence: curl -fsSL https://raw.githubusercontent.com/miaoxworld/OpenClawInstaller/main/config-menu.sh | bash in SKILL.md.
[EXTERNAL_DOWNLOADS]: The skill fetches installation resources from several external sources.
Fetches configuration and installation scripts from the miaoxworld GitHub repository.
Downloads the Node Version Manager (NVM) installer from the official nvm-sh GitHub repository.
Downloads the Ollama local AI installer from ollama.com.
[INDIRECT_PROMPT_INJECTION]: The skill enables an attack surface where instructions embedded in external message data could trigger sensitive agent capabilities.
Ingestion points: The skill integrates with Telegram, Discord, WhatsApp, Slack, WeChat, iMessage, and Feishu to process incoming message streams.
Boundary markers: No delimiters or protective instructions are provided to help the agent distinguish between message content and commands.
Capability inventory: The skill explicitly grants the agent the ability to execute system commands, perform file operations, and browse the web as part of its 'Remote Control' feature set.
Sanitization: There is no evidence of input validation or command sanitization for data received from messaging platforms.
[METADATA_POISONING]: There is a discrepancy between the official skill attribution and the authorship claimed within the documentation body.
Documentation in SKILL.md attributes the skill to 'ara.so
Hermes Skills collection', which differs from the recognized skill author.

openclaw-installer-deployment