The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The RemoteDebugServer implementation in 'resources/implementation-playbook.md' creates a WebSocket server (port 9230) that accepts 'evaluate' messages and executes them using the Node.js inspector's Runtime.evaluate method. The implementation contains no authentication or authorization logic, allowing any user with network access to the port to execute arbitrary JavaScript code in the process context.
[COMMAND_EXECUTION]: The provided Dockerfile in 'resources/implementation-playbook.md' sets the environment variable NODE_OPTIONS='--inspect=0.0.0.0:9229'. This configuration binds the Node.js debugger to all available network interfaces, which allows unauthorized remote attachment and command execution if the container is deployed in an environment without strict firewalling.
[DATA_EXFILTRATION]: The skill's implementation playbook defines multiple HTTP endpoints, such as '/debug/heap' and '/debug/profile', that trigger heap snapshots and CPU profiling. These snapshots capture the entire contents of the process memory, which can contain sensitive data including session tokens, environment variables, and personally identifiable information (PII).
[DATA_EXFILTRATION]: The distributed tracing configuration in 'resources/implementation-playbook.md' includes hooks that capture request and response bodies ('http.request.body') and user IDs ('user.id'). While the skill mentions redacting PII, the default code patterns provided actively collect and send this sensitive data to external collectors like Jaeger or Elasticsearch.

distributed-debugging-debug-trace