The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: Fetches installation scripts and configuration data from several well-known and trusted services, including NodeSource, Cloudflare, Tailscale, Docker, and the official Signal repository.
[REMOTE_CODE_EXECUTION]: Executes deployment and setup scripts from official sources such as openclaw.ai, nodesource.com, tailscale.com, cloudflare.com, and docker.com. These are standard installers for the infrastructure components required by the assistant.
[COMMAND_EXECUTION]: Instructs the agent to perform extensive shell operations for server provisioning, OS configuration, and software installation across all reference files.
[PRIVILEGE_ESCALATION]: Requires the use of 'sudo' to perform administrative tasks, including package management, systemd service configuration, and firewall rule adjustments.
[PERSISTENCE]: Configures systemd service units for OpenClaw, Ollama, and Cloudflare Tunnel to ensure services remain active across reboots, and establishes crontab entries for automated data backups.
[INDIRECT_PROMPT_INJECTION]: The skill facilitates the processing of untrusted data from multiple messaging platforms, creating a potential attack surface for indirect prompt injection.
Ingestion points: Ingests message streams from Signal, WhatsApp, Telegram, Discord, and Slack as documented in references/channel-setup.md.
Boundary markers: The skill acknowledges in references/security-hardening.md that prompt injection is an unsolved risk and provides guidance on implementing tool-use policies to mitigate it.
Capability inventory: The setup involves broad shell access, file system write permissions for agent workspaces, and network capabilities for LLM API interaction and messaging gateway connectivity.
Sanitization: Recommends implementing Docker sandboxing to isolate tool execution and using sensitive log redaction features to protect data privacy.

ac-openclaw