sme-agent-protocol

Pass

Audited by Gen Agent Trust Hub on Mar 1, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The protocol establishes an indirect prompt injection surface by requiring the agent to process data from the local filesystem and external URLs.
  • Ingestion points: The protocol mandates the use of the Read tool for local files and codebase content, and WebFetch or firecrawl for external web documentation.
  • Boundary markers: The protocol does not specify the use of delimiters or 'ignore embedded instructions' warnings when handling content from these sources.
  • Capability inventory: The protocol assumes or recommends access to Read, Grep, Glob, WebFetch, firecrawl, LSP, and Bash (specifically for read-only commands).
  • Sanitization: There are no instructions for sanitizing, escaping, or validating the data retrieved from external sources before it is analyzed or included in responses.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 1, 2026, 07:44 PM