pdfcraft-linked-agenda

Pass

Audited by Gen Agent Trust Hub on Nov 4, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill operates by searching for specific text within user-provided PDF files using the text_find tool. The output of these searches (text matches and coordinates) is then used to populate parameters for subsequent tool calls such as link_add. This ingestion of untrusted external data creates a potential surface for indirect prompt injection if the PDF content contains instructions designed to manipulate the agent's logic during the workflow execution.
  • [COMMAND_EXECUTION]: The skill includes a local utility script scripts/write_script.py which the agent is instructed to execute using Python. This script is used to 'compile' a JSON recipe into a sequence of tool calls for the pdfcraft-cli. A review of the script's source code confirms it performs only safe data transformation tasks using the Python standard library (argparse, json, pathlib) and does not execute arbitrary system commands or perform network operations.
  • [EXTERNAL_DOWNLOADS]: In references/workflow-sources.json, the skill references documentation and license files from the py-pdf/pypdf GitHub repository. These references are used solely for workflow alignment and do not involve the execution of remote code or the installation of unverified dependencies.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 4, 2026, 06:47 PM