zynkr-support

Pass

Audited by Gen Agent Trust Hub on Aug 15, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection due to its core functionality of processing untrusted external data.
  • Ingestion points: Untrusted data enters the agent context through Gmail message bodies and thread content retrieved via the google-workspace MCP server in SKILL.md (Step 2).
  • Boundary markers: The skill lacks explicit boundary markers or delimiters (e.g., XML tags or distinct markers) to separate incoming email content from instructional prompts. While it uses [[NEEDS PETER]] blocks for draft labeling, these do not isolate the input data during analysis.
  • Capability inventory: The skill possesses significant capabilities, including reading full email threads and creating new email drafts (mcp__google-workspace__draft_gmail_message), which could be abused if an attacker sends a malicious payload via email.
  • Sanitization: There is no evidence of sanitization, filtering, or instruction-override protection applied to the content parsed from the Gmail threads.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 15, 2026, 03:10 AM
Security Audit — agent-trust-hub — zynkr-support