pdfcraft-compress-pdf

Warn

Audited by Gen Agent Trust Hub on Nov 4, 2026

Risk Level: MEDIUMEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill configuration in references/craft-command-evidence.json specifies a binary download for the tool runtime from an external GitHub repository: https://github.com/storytold/pdfcraft/releases/download/v0.8.0/pdfcraft-0.8.0-windows-x64-portable.zip.
  • [REMOTE_CODE_EXECUTION]: The metadata identifies pdfcraft-cli.exe as the executable to be run. Since this is downloaded from an external source and subsequently executed to perform PDF tasks, it represents a remote code execution vector.
  • [COMMAND_EXECUTION]: The skill provides instructions in SKILL.md to install related components using the command npx skills add full-aigc-skills/pdfcraft-skills --skill pdfcraft-editor. This refers to a vendor-provided resource by the author full-aigc-skills.
  • [INDIRECT_PROMPT_INJECTION]: The skill processes PDF files which are untrusted external data. This creates a surface for indirect prompt injection if the PDF content contains malicious instructions designed to be interpreted by the AI agent.
  • Ingestion points: PDF file paths are ingested via the doc_open tool as described in SKILL.md and references/craft-methods.md.
  • Boundary markers: No explicit delimiters or instructions to ignore embedded content were found in the skill's logic to prevent the agent from following instructions found inside the PDFs.
  • Capability inventory: The skill possesses file system access capabilities (read and write) through tools like doc_open, doc_optimize, and page_render.
  • Sanitization: The instructions do not mention any sanitization or validation of the text or metadata content extracted from the PDF files before processing.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Nov 4, 2026, 06:47 PM