invoice-organizer

Pass

Audited by Gen Agent Trust Hub on Sep 15, 2026

Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill uses local shell commands (find, mkdir, cp, mv) to search for documents, create file structures, and relocate or duplicate files according to the organization plan. These operations are driven by user-initiated directories.
  • [INDIRECT_PROMPT_INJECTION]: The skill exposes a potential attack surface for indirect prompt injection because it processes untrusted third-party textual data.
  • Ingestion points: The skill ingests raw data from unverified PDF text and image OCR inputs during the processing step in SKILL.md (Step 2: Extract Information from Each File).
  • Boundary markers: There are no boundary markers or instructions to isolate the parsed contents from the core prompt logic, meaning a malicious payload embedded within an invoice text could influence the model's behavior.
  • Capability inventory: The system can perform filesystem modifications (mkdir, cp, mv) and create files (CSV summary output).
  • Sanitization: The instructions do not prescribe validation or sanitization rules for extracted values (such as vendor names or descriptions) before utilizing them to build paths or filenames, creating a risk of filename manipulation or CSV injection.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 15, 2026, 07:52 AM
Security Audit — agent-trust-hub — invoice-organizer