source-code-scanning
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill framework involves scanning and analyzing untrusted source code and project metadata. This creates a surface for indirect prompt injection where malicious instructions embedded in the files being reviewed could attempt to influence the agent's behavior. \n
- Ingestion points: The agent reads source code, configuration files, and package manifests from user-provided repositories.\n
- Boundary markers: The instructions do not specify the use of delimiters or protective framing to isolate scanned code from the agent's instructions.\n
- Capability inventory: The agent has access to numerous CLI tools and security scanners capable of executing commands and file operations.\n
- Sanitization: No sanitization of the ingested code content is performed before analysis.\n- [EXTERNAL_DOWNLOADS]: The skill workflow recommends installing and utilizing well-known security tools from official registries. These tools, including Semgrep, Bandit, and TruffleHog, are standard industry scanners.\n- [COMMAND_EXECUTION]: The audit process requires the execution of multiple shell commands for enumerating project files and running security analysis tools.
Audit Metadata