auto-bug-fixer
Pass
Audited by Gen Agent Trust Hub on Sep 20, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill instructs the agent to execute shell-level Git commands (
git bisect start,git bisect bad,git bisect good) to automate root cause analysis. - [EXTERNAL_DOWNLOADS]: The skill's output templates include instructions for installing external Python packages (
mindstudio-probe==8.3.0,tb_graph_ascend). These packages appear to be official resources related to the vendor's ecosystem. - [INDIRECT_PROMPT_INJECTION]: The skill possesses a broad attack surface as it processes untrusted user-supplied data such as
error_phenomenonandreproduce_stepsto generate code patches and test cases. - Ingestion points: Processes external error logs, user-provided reproduction steps, and code change information in
SKILL.md. - Boundary markers: The instructions provide a rigid structured output format, which helps maintain context but does not fully sanitize adversarial inputs.
- Capability inventory: The skill is authorized to perform Git operations and generate code/tests, which could be misused if malicious instructions are embedded in the error logs it analyzes.
- Sanitization: No specific sanitization or filtering logic is defined for the input data.
Audit Metadata