Skills
Navigate
skills/mukul975/anthropic-cybersecurity-skills/analyzing-pdf-malware-with-pdfid/Gen Agent Trust Hub

analyzing-pdf-malware-with-pdfid

Pass

Audited by Gen Agent Trust Hub on Sep 17, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONDYNAMIC_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted PDF content and presents it to the AI agent, creating a surface for injection. \n
  • Ingestion points: External PDF files are processed via scripts/agent.py and tools like pdf-parser. \n
  • Boundary markers: No explicit boundaries are used to distinguish untrusted PDF content from agent instructions. \n
  • Capability inventory: The skill can execute shell commands, perform file I/O, and emulate shellcode. \n
  • Sanitization: Content is extracted and displayed to the agent context without sanitization or instruction-filtering. \n- [COMMAND_EXECUTION]: The workflow relies on executing several external utilities via the shell. \n
  • Evidence: SKILL.md and related documentation specify the use of pdfid, pdf-parser, and scdbg via command-line execution and Python subprocess.run. \n- [DYNAMIC_EXECUTION]: The skill involves the dynamic extraction and emulation of binary payloads. \n
  • Evidence: It includes logic for carving PE files (MZ header detection) and emulating x86 shellcode using the speakeasy framework.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 17, 2026, 09:56 PM