The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: Documentation in SKILL.md and references/deployment.md advocates for the installation of the platform via a remote shell script (https://openclaw.ai/install.sh) piped directly to bash. This is represented as a standard manual installation method.
[COMMAND_EXECUTION]: The skill provides numerous examples of shell commands for system administration and automation. This includes firewall management (ufw), SSH hardening (sed), service control (systemctl), and the use of shell utilities like tar and find within automation prompts for tasks such as backups and log compaction.
[PROMPT_INJECTION]: The skill describes a platform architecture that processes untrusted external data, creating a potential surface for indirect prompt injection.
Ingestion points: The documented system reads web content, messaging channel inputs (Telegram/WhatsApp), and data from external APIs like GitHub and Stripe.
Boundary markers: The provided prompt templates do not explicitly demonstrate the use of delimiters or instructions to ignore nested prompts within external data.
Capability inventory: Depending on configuration, agents may be granted access to shell execution (execute_command) and file system operations.
Sanitization: The references/security.md file provides detailed guidance on mitigation strategies, including 'strict' sandboxing modes, granular tool allow-lists, and Docker security optimizations to isolate agent environments.
[EXTERNAL_DOWNLOADS]: The skill references several well-known services and tools for integration, including Ollama, ElevenLabs (voice), Supabase (database), n8n (automation), and networking tools like Cloudflare and Tailscale.

openclaw-expert