The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes a local Node.js script (light-webserver.js) to manage a preview environment. It uses bash for directory setup and server lifecycle management. These operations include defensive checks for symlinks and file ownership before execution.
[DATA_EXFILTRATION]: The prototype server binds to the loopback interface (127.0.0.1) by default. API endpoints for the annotation loop are protected by a randomly generated session token to prevent unauthorized access from other local processes.
[INDIRECT_PROMPT_INJECTION]: The skill handles user-provided annotations (comments and CSS selectors) through a web interface. The documentation in references/annotation-loop.md explicitly instructs the agent to treat these inputs as untrusted and to never execute them as commands, correctly identifying and mitigating this attack surface.
[DYNAMIC_EXECUTION]: The light-webserver.js script uses child_process methods (spawn, execFileSync) exclusively for process management (e.g., checking if a server PID is alive). The PIDs are validated as integers before use, preventing command injection.

ce-prototype