The Agent Skills Directory

Dynamic Execution: The skill executes inline Node.js scripts to perform non-trivial file system operations, such as monitoring log directories for changes and reading file content from specific byte offsets. While these scripts serve as utility functions for diagnostic purposes, the execution of logic assembled at runtime is a pattern that warrants review.
Indirect Prompt Injection: The skill monitors and classifies runtime logs generated by the application. Because these logs can capture data from external or untrusted sources, there is a potential attack surface for indirect prompt injection, where malicious instructions within logs could attempt to influence the agent's subsequent classification or repair actions. The skill includes a redaction gate and structured parsing logic to help mitigate this risk.
Command Execution: The skill performs various shell operations to support development workflows, including dependency restoration via npm ci, project-local port discovery, and starting the application server. The instructions emphasize safety practices, such as requiring user approval for process control and providing specific escaping guidance to the agent to prevent command injection.
Diagnostic Data Management: The skill maintains an audit trail of its actions and findings in a local .powernative/debug-app/ directory. It uses a specialized redaction script to sanitize sensitive information—such as tokens, secrets, and headers—before persisting any diagnostic summaries, which is a positive security practice.

debug-app