reading-invoice
Pass
Audited by Gen Agent Trust Hub on Feb 25, 2026
Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill executes the
shinkokuCLI tool to handle PDF files. It usesextract-textto pull data from text-based PDFs andto-imageto convert scanned PDFs into PNG format for OCR processing. These commands are directly related to the skill's primary function of invoice digitization. - [PROMPT_INJECTION]: The skill possesses a surface for indirect prompt injection as it processes untrusted content from external invoice files.
- Ingestion points: Text extracted from PDF files via the
shinkokuutility and image data processed through theReadtool. - Boundary markers: The skill defines structured output tags (
---INVOICE_DATA---), but does not utilize specific delimiters or defensive instructions to isolate untrusted input content during the analysis phase. - Capability inventory: Subprocess execution (
shinkoku), file system traversal (Glob), and file reading (Read). - Sanitization: No explicit sanitization, filtering, or escaping logic is applied to the data extracted from invoices before it is interpreted by the agent.
Audit Metadata