The Agent Skills Directory

[PROMPT_INJECTION]: The skill contains instructions that direct the agent to execute steps 'autonomously' and to 'run the Bash commands directly, do not just describe them'. This pattern attempts to override standard agent behavior regarding user confirmation and transparency.
[COMMAND_EXECUTION]: The skill performs extensive command-line operations, including process management (killing existing services via lsof and pkill), environment setup, and launching background processes for the web server and the network tunnel.
[EXTERNAL_DOWNLOADS]: The skill utilizes npm ci to fetch dependencies and uses the Cloudflare Tunnel (cloudflared) tool to establish remote connectivity. While these target well-known services (NPM and Cloudflare), they involve external network interactions and code execution.
[DATA_EXPOSURE]: The instructions explicitly reference and access sensitive local files, including .env.production.local for environment variables and Cloudflare credential files located in ~/.cloudflared/. The skill also establishes a public endpoint (tunnel) to the local server, which creates a data exposure surface.

expose-tunnel