argent-metro-debugger
Pass
Audited by Gen Agent Trust Hub on Sep 14, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill interacts with external application data during debugging sessions, creating a potential exposure surface for untrusted content.
- Ingestion points: Application runtime data, including console log files on disk, evaluation responses, and component tree structures are processed by the agent (detailed in SKILL.md).
- Boundary markers: The skill contains explicit documentation ('Golden Rule 4') warning the agent that logs and app content are untrusted data and should never be interpreted as instructions.
- Capability inventory: Includes capabilities to execute JavaScript code within the target application context using
debugger-evaluateand search local files usingGrep. - Sanitization: Relies on clear behavioral instructions within the prompt guiding the agent to ignore embedded instructions and avoid copying or exposing discovered credentials.
Audit Metadata