The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: Fetches the Tailscale installation script from the official tailscale.com domain.
[REMOTE_CODE_EXECUTION]: Executes the Tailscale installation script via shell piping (curl | sh). This is the standard, documented installation method for this well-known service.
[PRIVILEGE_ESCALATION]: Uses sudo for necessary system configurations, including managing the UFW firewall, starting Tailscale, and controlling systemd services for agent persistence.
[COMMAND_EXECUTION]: Provides instructions for standard development and system tools including git, npm, ufw, and systemctl. It also includes a shell script for monitoring API usage via curl requests to well-known AI provider endpoints.
[INDIRECT_PROMPT_INJECTION]: The skill defines a surface for processing external data by providing templates for the agent to rebuild local skills based on third-party descriptions.
Ingestion points: User-provided prompts in SKILL.md that describe external skill functionality.
Boundary markers: Recommends a dedicated "Security rules" section in the system prompt to enforce constraints.
Capability inventory: Configures permissions for shell_exec, file_write, and network_request in config.json.
Sanitization: Implements strict security through require_confirmation gates and instructions for manual code review.

openclaw-runbook-security