kill_switch

Global emergency stop. If triggered, the pipeline denies requests before they reach downstream modules.

What it mitigates

• Runaway agent behavior during incidents
• Delayed human response when immediate halt is needed
• Unsafe execution while policy is being updated

Recommended defaults

Minimal config

moduleConfig:
  kill_switch:
    enabled: true
    envVar: RADIUS_KILL_SWITCH
    filePath: ./.radius/KILL_SWITCH
    denyPhases:
      - pre_request
      - pre_tool
    reason: "emergency kill switch active"

Operational pattern

1. Set env var (RADIUS_KILL_SWITCH=1) or create trigger file.
2. Requests are denied deterministically.
3. Remove trigger and rerun npx agentradius doctor.

Use this as incident containment, not as a permanent policy mode.

Related modules

• audit
• approval_gate