Pass
Audited by Gen Agent Trust Hub on Sep 26, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted data from external PDF documents, which serves as an attack surface for instructions meant to influence agent behavior.
- Ingestion points: The skill uses
pypdf.PdfReaderin several scripts (e.g.,scripts/extract_form_field_info.py,scripts/fill_fillable_fields.py) andpdfplumber.openinSKILL.mdto read content from user-provided PDF files. - Capability inventory: The skill possesses capabilities to write files (
writer.write), execute shell commands (qpdf,pdftk), and generate images from PDF pages. - Boundary markers: While
forms.mdestablishes a structuredfields.jsonworkflow to handle form data, the agent still processes raw text extracted from the document body. - Sanitization: There is no evidence of explicit sanitization or filtering of text extracted from PDFs before it is processed by the AI agent context.
- [DYNAMIC_EXECUTION]: The script
scripts/fill_fillable_fields.pyperforms runtime modification of thepypdflibrary, known as monkeypatching. - Evidence: The function
monkeypatch_pydpf_methodredefinesDictionaryObject.get_inheritedto correct a specific bug in selection list handling withinpypdfversion 5.7.0. - [COMMAND_EXECUTION]: The skill instructions facilitate the execution of various command-line utilities for PDF processing.
- Evidence:
SKILL.mdandreference.mdprovide usage examples forqpdf,pdftk, andpoppler-utils(pdftotext,pdftoppm,pdfimages) which require shell execution privileges.
Audit Metadata