The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The RemoteDebugServer class in resources/implementation-playbook.md implements a WebSocket handler that accepts an evaluate command. This command passes arbitrary input directly to inspector.Session().post('Runtime.evaluate', ...) with includeCommandLineAPI: true, enabling remote attackers to execute arbitrary JavaScript within the process.
[COMMAND_EXECUTION]: The provided Dockerfile configuration in resources/implementation-playbook.md exposes the Node.js debugger on all network interfaces (--inspect=0.0.0.0:9229) and installs low-level system tools (tcpdump, strace, gdb) which can be leveraged for advanced exploitation if the container is compromised.
[DATA_EXFILTRATION]: The TracingSystem implementation in resources/implementation-playbook.md uses OpenTelemetry hooks to capture full HTTP request bodies (span.setAttribute('http.request.body', JSON.stringify(request.body))). This pattern frequently leads to the leakage of PII, passwords, and session tokens into distributed tracing backends like Jaeger.
[CREDENTIALS_UNSAFE]: The ProductionDebugger class implements a backdoor mechanism that enables debugging features based on a static header token (x-debug-token). This represents a significant security risk if the token is leaked or if the IP-based whitelist is misconfigured.

distributed-debugging-debug-trace