The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill documents an installation method that downloads a script from a remote URL and executes it immediately via bash. Evidence: curl -fsSL https://openclaw.ai/install.sh | bash in SKILL.md and references/deployment.md.
[EXTERNAL_DOWNLOADS]: The skill mentions downloading the openclaw CLI and various channel plugins from npm and the project's registry. Evidence: npm install -g openclaw@latest and clawhub install openclaw-genie.
[COMMAND_EXECUTION]: The skill describes an exec tool allowing the agent to run arbitrary shell commands on the gateway, a sandbox, or a connected node. Evidence: Documentation for the exec tool in references/tools.md.
[INDIRECT_PROMPT_INJECTION]: The skill documents a gateway that processes untrusted messages from external platforms like Discord, WhatsApp, and Telegram, which could contain malicious instructions. 1. Ingestion points: Messages from 22+ messaging platforms described in references/channels.md. 2. Boundary markers: None explicitly provided in the instructions. 3. Capability inventory: The agent has access to exec (shell), browser (automation), and message (messaging) tools. 4. Sanitization: The skill mentions a sandbox mode in openclaw.json to isolate untrusted sessions.

openclaw-genie