The Agent Skills Directory

[COMMAND_EXECUTION]: The skill requires administrative privileges via sudo to perform high-risk operations such as managing system services with launchctl and granting binary capabilities with setcap on Linux.\n- [EXTERNAL_DOWNLOADS]: Users are instructed to build a custom binary using xcaddy, which downloads the caddy-dns/vercel plugin from a remote GitHub repository during the build process.\n- [CREDENTIALS_UNSAFE]: The documentation suggests storing sensitive Vercel API tokens (vcp_*) in environment files or LaunchDaemon plist files, which remain in plaintext on the local filesystem.\n- [PROMPT_INJECTION]: The skill creates an attack surface for indirect prompt injection by modifying local configuration files based on untrusted user inputs.\n
Ingestion points: ~/.config/caddy/Caddyfile and ~/.openclaw/openclaw.json (described in SKILL.md and OPENCLAW.md).\n
Boundary markers: None present; configuration blocks are manually edited or appended.\n
Capability inventory: The skill uses caddy reload, launchctl, and systemctl to apply these configuration changes to the system.\n
Sanitization: No evidence of input validation or escaping for variables like YOUR_DOMAIN or APPNAME before writing to config files.

caddy