The Agent Skills Directory

[COMMAND_EXECUTION]: The reference guide contains numerous examples of commands requiring administrative privileges (sudo) to manage system-level services via systemctl, modify configuration files in protected directories, and manage cron jobs for the openclaw user.
[EXTERNAL_DOWNLOADS]: The documentation includes instructions for downloading and executing software from the primary project domain, including an installation script (https://openclaw.ai/install.sh) and various NPM packages.
[REMOTE_CODE_EXECUTION]: The skill details how to run remote MCP server packages using uvx and provides patterns for piping remote scripts directly to a shell for installation and updates.
[PROMPT_INJECTION]: The skill describes a gateway system that ingests untrusted data from multiple external sources (Telegram, Discord, Slack, WhatsApp, iMessage), creating a surface for indirect prompt injection.
Ingestion points: External messaging channels documented in references/channels.md.
Boundary markers: The provided documentation does not specify the use of delimiters or safety instructions to isolate untrusted user messages from the agent's instructions.
Capability inventory: The system is documented to have access to shell execution via the exec tool (as detailed in references/05-config-and-cli.md) and the ability to schedule and execute autonomous tasks via a cron system (documented in references/03-cron-heartbeat.md).
Sanitization: No specific sanitization or escaping mechanisms for inbound message content are described in the provided reference materials.

openclaw-guide