malware-analysis
Pass
Audited by Gen Agent Trust Hub on Feb 17, 2026
Risk Level: SAFEPROMPT_INJECTIONDATA_EXFILTRATION
Full Analysis
- Indirect Prompt Injection (LOW): The skill extracts strings and metadata from untrusted malware samples and prepares them for LLM analysis, creating a surface for indirect prompt injection. \n
- Ingestion points: File content in
scripts/generate_report.pyis processed viaanalyze_fileandextract_from_filecalls. \n - Boundary markers: The formatted output in
format_as_textuses standard ASCII separators (e.g.,===,---) but lacks explicit protective instructions for the LLM to ignore embedded commands. \n - Capability inventory: The script performs local file writes (
Path.write_text) and initiates network API calls. \n - Sanitization: IOC extraction utilizes
defang=True, but other extracted strings and metadata are not sanitized for LLM safety. \n- Network Operations (LOW): The script (via theMalwareTriagemodule) queries external threat intelligence APIs including VirusTotal and AbuseIPDB. While these activities are inherent to the primary function of the skill, they represent communication with non-whitelisted external domains.
Audit Metadata