grey-haven-smart-debugging
Warn
Audited by Gen Agent Trust Hub on Feb 17, 2026
Risk Level: MEDIUMEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS] (MEDIUM): The file
reference/error-patterns-database.mdcontains a template that explicitly instructs the agent to runpip install package_name. This encourages the installation of unverified third-party dependencies during the debugging process. - [COMMAND_EXECUTION] (LOW): The
SKILL.mdconfiguration enables theBashtool, which permits arbitrary command execution on the host. While relevant to debugging, it significantly expands the attack surface. - [PROMPT_INJECTION] (LOW): The skill is susceptible to Indirect Prompt Injection (Category 8) because it processes untrusted data (stack traces, error logs) and has access to high-privilege tools.
- Ingestion points: Error logs and stack traces provided by the user or read via
GrepandReadtools. - Boundary markers: Absent. There are no instructions to the agent to ignore embedded commands within the analyzed logs.
- Capability inventory: Includes
Bash,Write, andMultiEditfor code modification and system access. - Sanitization: No sanitization of ingested trace data is mentioned.
Audit Metadata