The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The remote-debug-server.js file provides code to instantiate a WebSocket server that processes an evaluate command. This command executes arbitrary JavaScript strings using the Runtime.evaluate method of the Node.js inspector module. The server defaults to listening on 0.0.0.0, which exposes the code execution interface to all network interfaces without authentication in the basic server implementation.
[DYNAMIC_EXECUTION]: The skill facilitates runtime code execution through both the remote debugger implementation and the use of the source-map-support package, which overrides the default stack trace generation logic to map compiled code back to source files at runtime.
[DATA_EXFILTRATION]: The tracing.js configuration captures http.request.body and user/session identifiers (req.user.id, req.session.id). This sensitive data is automatically transmitted to a remote Jaeger endpoint (JAEGER_ENDPOINT), which could result in the leakage of credentials or PII if the requests contain such information. Additionally, production-debug.js allows capturing per-request logs and returning them in response headers (X-Debug-Info).
[INDIRECT_PROMPT_INJECTION]: The skill processes external requirements via the $ARGUMENTS placeholder without validation or boundary markers. Given the skill's ability to generate network-exposed debug servers and data-capturing tracers, malicious input could lead the agent to produce insecure or compromised infrastructure code.
Ingestion points: Found in the Requirements section of SKILL.md.
Boundary markers: No delimiters or safety warnings are applied to the interpolated requirements.
Capability inventory: Includes remote command execution (via inspector), network data transmission (to Jaeger/Elasticsearch), and local file writing (for logging and profiling).
Sanitization: No evidence of input sanitization or filtering was found.

distributed-debugging-debug-trace