---
description: Report AgentGuards guardrail status for a self-hosted appliance. Use when the user runs /agentguards:status or asks whether AgentGuards is active, healthy, or correctly configured.
---

# AgentGuards status — self-hosted

There is no `health_check` MCP tool in this variant (see the `guardrails`
skill for why). Check status directly instead:

1. **Confirm both required variables are set** in the environment:
   `AGENTGUARDS_URL` and `AGENTGUARDS_API_KEY`. If either is missing, stop here
   and point the user at the `setup` skill — the hooks fail closed
   ("Not configured") without them, which blocks every prompt.

2. **Check the appliance is reachable**, using the actual configured value —
   never hardcode a URL:
   ```
   curl -s -o /dev/null -w '%{http_code}\n' "$AGENTGUARDS_URL/health"
   ```
   `200` means the appliance is up. A connection failure or a certificate error
   means the hooks will fail closed too — see the `setup` skill's TLS step.

3. **Confirm the key is accepted**, not just present:
   ```
   curl -s -o /dev/null -w '%{http_code}\n' "$AGENTGUARDS_URL/v1/guardrails/evaluate-input" \
     -H "X-API-Key: $AGENTGUARDS_API_KEY" -H 'Content-Type: application/json' \
     -d '{"text":"status check"}'
   ```
   `200` means it's accepted. `401` means the key is wrong or was revoked on
   the appliance's own Keys page.

4. **Report plainly**: which of the two variables are set, the `/health`
   status code, and the evaluate-input status code. Do not speculate beyond
   what these three checks actually showed.