Pass
Audited by Gen Agent Trust Hub on Mar 10, 2026
Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTIONREMOTE_CODE_EXECUTION
Full Analysis
- [REMOTE_CODE_EXECUTION]: The script
scripts/fill_fillable_fields.pymodifies the internal behavior of thepypdflibrary at runtime. - Evidence: The
monkeypatch_pydpf_methodfunction reassigns theDictionaryObject.get_inheritedmethod to a local implementation to adjust how PDF form field attributes are retrieved. - [PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection as it processes content from untrusted external PDF documents.
- Ingestion points: PDF files read and processed through the
pypdfandpdfplumberlibraries as described inSKILL.mdandforms.md. - Boundary markers: None; the skill does not instruct the agent to use delimiters or ignore instructions found within the processed PDF content.
- Capability inventory: The skill has the capability to read/write files and provides instructions for executing external command-line tools.
- Sanitization: No sanitization or validation of the text extracted from PDF files is performed before processing.
- [PROMPT_INJECTION]: There is a discrepancy in the metadata regarding the skill's authorship and licensing.
- Evidence: The skill metadata identifies "Agilkannan" as the author, whereas the
LICENSE.txtfile attributes ownership to "Anthropic, PBC". - [COMMAND_EXECUTION]: The skill documentation provides instructions for the agent to execute various system-level command-line utilities.
- Evidence:
SKILL.mdandreference.mdcontain commands forqpdf,pdftotext,pdftoppm,pdfimages, and the ImageMagickmagicktool.
Audit Metadata