receipt-tax-ocr
Pass
Audited by Gen Agent Trust Hub on Oct 8, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests data from OCR processes on user-provided images, which creates an attack surface where malicious text embedded in a receipt could attempt to influence the agent's behavior during renaming or memo generation.
- Ingestion points: Local image files in the current working directory processed via OCR (Step 2 in SKILL.md Procedure).
- Boundary markers: The skill uses instructional guidance to classify data and explicitly requires the agent to 'Ask the user, do not guess' when transaction details are unclear, which acts as a manual verification step.
- Capability inventory: The agent is authorized to rename files on the filesystem and create or update markdown (.md) memo files.
- Sanitization: There are no explicit instructions for technical sanitization, escaping, or filtering of the strings extracted via OCR before they are used to construct filenames or table entries.
Audit Metadata