---

name: status description: Report AgentGuards guardrail status. Use when the user asks whether AgentGuards is active, healthy, or correctly configured in Codex.

AgentGuards status

Check the real state yourself — run these, don't hand them to the user. Never print the API key: the commands below only ever pass it along, and you report at most its ag_ prefix.

1. If the AgentGuards CLI is installed, use it. It checks the key, the service and each agent's install in one go. Run agentguards doctor, or ~/.agentguards/bin/agentguards doctor if it is not on PATH (on Windows, $HOME\.agentguards\bin\agentguards.exe doctor). If that ran, report its output and stop here.

2. Otherwise, find the key. The hooks use the first of: the AGENTGUARDS_API_KEY environment variable, ~/.codex/agentguards_token, then ~/.agentguards/credentials.json (saved by agentguards login). Say which one is set — not the value.

3. Check the service is reachable (macOS/Linux):

   url="${AGENTGUARDS_URL:-https://prod.agentguards.co}"
   curl -s -o /dev/null -w '%{http_code}\n' "$url/health"
   

   200 means it is up. A connection failure means the hooks fail closed and block until it is back (unless AGENTGUARDS_FAIL_OPEN=true).

4. Confirm the key is accepted, not just present (same shell, so $url is set) (macOS/Linux):

   key="${AGENTGUARDS_API_KEY:-$(cat ~/.codex/agentguards_token 2>/dev/null)}"
   key="${key:-$(python3 -c "import json,os;print(json.load(open(os.path.expanduser('~/.agentguards/credentials.json')))['api_key'])" 2>/dev/null)}"
   curl -s -o /dev/null -w '%{http_code}\n' "$url/v1/guardrails/evaluate-input" \
     -H "X-API-Key: $key" -H 'Content-Type: application/json' -d '{"text":"status check"}'
   

   200 means it is accepted. 401 means the key is wrong or was revoked at https://agentguards.co/dashboard/keys. On Windows, do the same with Invoke-WebRequest, reading the key from the same three places.

5. Report plainly: where the key comes from, the two status codes, the fail mode (fail-closed unless AGENTGUARDS_FAIL_OPEN=true), and what the hooks cover — every prompt, every shell command, fetched content from shell commands, and apply_patch edits. Don't speculate beyond what the checks showed. If the key is missing or rejected, use the setup skill.