Skip to content
Ganglia

zone_guard

zone_guard is a reflex, not a rule. It traces the call graph ahead of every AI tool call, detects when a proposed action would touch a marked node, and intercepts — before the AI writes a single character.

How it works

Section titled “How it works”
  1. You mark nodes as danger zones: processPayment, AuthGuard, db.migrate
  2. When the AI calls any tool that would affect those nodes, zone_guard fires first
  3. It surfaces the blast radius and asks for explicit permission
  4. The AI cannot proceed without your approval

The interception happens at the MCP layer — before execution, not after.

Marking danger zones

Section titled “Marking danger zones”
# Via natural language to your AI agent:
"Mark processPayment as a danger zone"


# Or directly:
zone_guard(action="mark", node="processPayment", reason="payment critical path")

Managing zones

Section titled “Managing zones”
zone_guard(action="list")           → all marked nodes
zone_guard(action="unmark", node="processPayment")
zone_guard(action="status")         → zones loaded, paths monitored, intercepts today

When an intercept fires

Section titled “When an intercept fires”
⚠ zone_guard intercept


The proposed action touches 2 danger zone nodes:
  • processPayment  (CRITICAL — payment execution)
  • AuthGuard       (HIGH — all authenticated routes)


Blast radius: 12 functions, 3 API routes
Intercept depth: 3 hops


Approve this action? [yes/no]

Use cases

Section titled “Use cases”
  • Payment and billing code — never touch without explicit sign-off
  • Auth and session logic — high blast radius, easy to break silently
  • Database migrations — irreversible, must be reviewed
  • Third-party integrations — API calls with real-world side effects