The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill provides instructions and automation for downloading various developer tools and OS-level dependencies from well-known and trusted sources, including GitHub, Nodesource (Node.js), Docker, GitLab, and official vendor domains like Claude.ai and ChatGPT.com.
[REMOTE_CODE_EXECUTION]: Several setup runbooks utilize standard 'curl | bash' patterns to install critical toolchain components such as Node.js, fnm (Fast Node Manager), the Claude Code CLI, and the Codex CLI. These operations are performed against official vendor infrastructure and are typical for system administration tasks.
[COMMAND_EXECUTION]: The provided shell scripts (e.g., seed.sh, assistant-kill.sh, assistant-maintenance.sh) and markdown runbooks execute a wide range of system commands to manage services, configure firewalls (UFW), handle rootless containers (Podman), and perform maintenance. These are appropriate given the skill's purpose as a server administration and assistant deployment guide.
[PRIVILEGE_ESCALATION]: The server setup guides provide clear instructions for human administrators to configure passwordless sudo for an admin account and perform necessary root-level configurations. Additionally, the skill implements a 'security hardening' phase that uses root privileges to lock down the assistant's service account, restricting its ability to modify its own configuration or system files.
[PERSISTENCE]: The skill automates the creation and management of systemd user-level services to ensure the AI assistant (OpenClaw) remains running across sessions. It also configures Linux 'lingering' for the service account to maintain background processes without an active login session.
[INDIRECT_PROMPT_INJECTION]: The assistant (OpenClaw) is designed to process untrusted input from external communication platforms (Slack and Discord). This ingestion of external data constitutes a known attack surface. The skill proactively addresses this by providing the agent with explicit 'Red Line' instructions to never exfiltrate credentials and requiring it to follow a pre-defined 'playbook' for all delegated tasks.
[DYNAMIC_EXECUTION]: Administrative shell scripts use Node.js one-liners for safe JSON processing and file manipulation during the 'seeding' and 'maintenance' phases of the deployment.

alignfirst-setup-guide