code-forensics-weapon
Pass
Audited by Gen Agent Trust Hub on Jul 25, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection through the ingestion and processing of untrusted external data.
- Ingestion points: Untrusted data enters the context via email archives (scripts/parse_emails.py), PDF invoices (scripts/parse_invoices.py), and git repository logs (scripts/parse_git_log.py).
- Boundary markers: The report templates (e.g., scripts/build_master_report.js, scripts/build_attorney_memo.js) interpolate extracted text into document placeholders without explicit boundary markers or instructions for downstream consumers to ignore embedded commands.
- Capability inventory: The skill utilizes local file system access for reading source materials and writing deliverables, and it invokes the 'soffice' command for PDF conversion.
- Sanitization: The skill performs basic sanitization, such as stripping HTML script/style tags using BeautifulSoup in scripts/parse_emails.py, but it does not filter or validate the content of all extracted strings against injection patterns.
Audit Metadata