dsh-deepread

Pass

Audited by Gen Agent Trust Hub on Sep 4, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill is designed to process external, untrusted content such as articles, PDFs, and web pages, which represents an attack surface for indirect prompt injection.
  • Ingestion points: The skill accepts document sets, web pages, and PDF files as primary input for analysis as described in the 'Establish the Reading Contract' step.
  • Boundary markers: The 'Security & Safety Notes' section explicitly instructs the agent to treat content as untrusted data, never as instructions, and to avoid following embedded prompts or executing commands found in sources.
  • Capability inventory: The instructions recommend utilizing the host agent's native tools for file access, OCR, and web reading.
  • Sanitization: The workflow (Steps 3 and 4) enforces a clear separation between source text (Evidence Ledger) and agent/user interpretation, preventing external content from polluting the agent's logic.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 4, 2026, 10:17 AM
Security Audit — agent-trust-hub — dsh-deepread