run-quality-gate
Fail
Audited by Gen Agent Trust Hub on Jul 18, 2026
Risk Level: HIGHCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill's implementation for the 'custom' gate maps the user-supplied 'custom_command' input directly to the Bash tool. This allows for arbitrary command execution on the host environment as there is no defined whitelist, sanitization, or restriction on the provided string.\n- [PROMPT_INJECTION]: The skill presents a significant attack surface for indirect prompt injection. If the 'custom_command' field is populated by an agent based on untrusted data (e.g., from a code repository, CI configuration, or PR description), an attacker can execute malicious scripts.\n
- Ingestion points: The 'custom_command' property in the input JSON defined in SKILL.md.\n
- Boundary markers: Absent. The skill instructions do not provide delimiters or 'ignore' instructions to separate the command logic from potentially malicious data.\n
- Capability inventory: The skill utilizes the 'Bash' tool for command execution and the 'Read' tool for file access across its operational logic.\n
- Sanitization: Absent. While the process step mentions validating the command, no specific logic, regex, or constraints are provided, leaving the execution vulnerable to injection.
Recommendations
- AI detected serious security threats
Audit Metadata