resonance-backend
Pass
Audited by Gen Agent Trust Hub on Feb 17, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION] (LOW): Indirect Prompt Injection surface. The skill ingests file content via
read_fileand possesses capabilities to execute commands and write files, creating a potential vector if malicious instructions are present in processed data.\n - Ingestion points:
read_fileis used for codebase analysis.\n - Boundary markers: Absent; no specific delimiters or instructions to ignore commands within file content are provided to the agent.\n
- Capability inventory:
run_command,write_file,edit_file, andread_file.\n - Sanitization: Absent; the skill focuses on validating the application's external inputs but does not sanitize the content it reads from the local environment.
Audit Metadata