polite-mcp-conventions

Pass

Audited by Gen Agent Trust Hub on Sep 6, 2026

Risk Level: SAFE
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill describes workflows involving the ingestion of external data during 'live spike/CDP exploration' and 'DOM inspection' of third-party websites. This represents a potential surface for indirect prompt injection.
  • Ingestion points: Browser-based exploration and DOM inspection of shopping and classifieds sites.
  • Boundary markers: The skill mandates the use of AskUserQuestion to confirm the first live mutating action in a session.
  • Capability inventory: Capabilities include browser automation (CDP exploration), writing to local files (PLAN.md), and interactive user prompting.
  • Sanitization: The workflow relies on explicit human-in-the-loop confirmation before high-stakes actions are executed.
  • [SAFE]: The instructions are explicitly designed to increase safety by requiring confirmation before any browser actions that modify a user's personal account, such as adding items to a shopping cart. The practice of logging research findings into PLAN.md is a standard development convention.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 6, 2026, 12:53 AM
Security Audit — agent-trust-hub — polite-mcp-conventions