Published guides for HOL Guard and plugin-scanner, pulled from the AI Plugin Scanner library docs so the product page stays aligned with the canonical documentation source.
Published guides for HOL Guard and plugin-scanner, pulled from the AI Plugin Scanner library docs so the product page stays aligned with the canonical documentation source.
Use this guide when Gemini CLI is loading local extensions or embedded MCP server declarations and you want Guard to block changed extension surfaces before launch.
blocked changes go to the local approval center until Gemini exposes a richer native approval model
What Guard treats as an artifact
Guard treats both of these as relevant launch inputs:
the extension manifest itself
any embedded MCP server declarations inside that manifest
That matters because a Gemini extension can change publisher metadata, context wiring, and runtime server commands in the same file.
Install and launch flow
hol-guard detect gemini --json
hol-guard install gemini
hol-guard run gemini --dry-run
hol-guard run gemini
hol-guard approvals
Recommended validation loop
Use a workspace or home directory with a real Gemini extension installed, then confirm:
Guard detects the extension manifest
embedded MCP server entries are listed in the detection output
the first dry run records a receipt
changed extension metadata routes to the approval center when policy blocks
Best fit
Gemini is a good Guard fit when you want:
extension-level trust before launch
publisher and context metadata recorded alongside MCP server definitions
a local approval-center path while Gemini-native approvals are still limited
Next guides
Gemini harness
Use this guide when Gemini CLI is loading local extensions or embedded MCP server declarations and you want Guard to block changed extension surfaces before launch.