debugging
Pass
Audited by Gen Agent Trust Hub on Nov 2, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data such as logs, error stacks, API responses, and captured request replays in
SKILL.mdandVERIFICATION.md. This creates a surface for indirect prompt injection where malicious instructions could be embedded in the data being debugged. - Ingestion points: External data enters through reproducers, error logs, stack traces, and API/UI evidence.
- Boundary markers: The skill instructs to identify invariants at boundaries and compare working vs. failing paths to help isolate unexpected behavior.
- Capability inventory: The skill uses project-specific tools for testing and measurement, which potentially involve file system access and execution of test suites.
- Sanitization: The instructions explicitly require redacting secrets from logs, commands, and artifacts before presentation to the user or agent context.
Audit Metadata