Turn StatusCake into a verified alerting flow with Hermes

This is a submission for the Hermes Agent Challenge

What I Built

Most monitoring setups have the same weak spot.

Detection is easy. Decision-making is not.

StatusCake is good at telling you that something might be wrong. What happens next is where things usually get messy. One alert goes straight to a chat room. Another wakes the wrong person. A third turns out to be missed and never verified because the site had a brief wobble and recovered before anyone looked.

Hermes is useful in that gap.

Instead of treating StatusCake as the whole incident workflow, you can put Hermes behind the webhook and use it as a small verification and routing layer. The pattern is simple:

• StatusCake detects the issue
• Hermes receives the webhook
• Hermes logs the raw request
• Hermes checks the target from its own network perspective
• Hermes decides whether the alert is actually worth escalating
• Hermes sends the notification and keeps a local event history

That gives you something better than "dump every alert into a channel" without dragging in a full incident-management platform.

The short version

StatusCake detects. Hermes verifies, routes, emails, and remembers.

That is the whole idea.

Why bother adding Hermes at all?

Because most teams do not want every uptime alert treated the same way.

A down alert at 2:14pm during business hours is not the same as a down alert at 2:14am. A monitor flapping for ten seconds is not always the same as a real outage. And when something goes wrong, you usually want a record of what arrived, what Hermes saw, and who got notified.

Hermes gives you a good place to add that logic.

In this setup it does four jobs:

1. It receives the webhook in a deterministic way.
2. It verifies whether the site is actually down from where Hermes is running.
3. It applies simple time-of-day routing.
4. It keeps a JSONL incident history and can send a daily summary by email.

That is enough to make a cheap monitoring stack feel much more operationally sane.

The architecture

This is intentionally small.

StatusCake
  -> webhook POST
  -> local receiver beside Hermes
  -> raw request log
  -> verification step
  -> escalation decision
  -> immediate email notification
  -> JSONL event history
  -> daily summary email

There is no requirement to package this as a formal Hermes skill on day one. If your goal is to get a useful workflow running quickly, an ad hoc setup with a few scripts and a config file is enough.

That is the version I would recommend first.

Why webhook-first is the right move

You could try to build something around email parsing, but that is the wrong direction here.

Webhook delivery is cleaner for three reasons:

• StatusCake already supports it natively
• you keep the alert payload structured instead of scraping email text
• debugging gets much easier because you can log the exact request body and headers

In practice, that last bit matters more than people expect. When an alert does not behave the way you thought it would, the first question is usually: did StatusCake send what I think it sent?

If you keep a raw inbound log, you can answer that immediately.

The receiver can stay tiny

You do not need a huge service for this. A small local HTTP receiver is enough.

# scripts/statuscake_webhook_receiver.py
python3 scripts/statuscake_webhook_receiver.py \
  --config var/escalation-config.json \
  --host 127.0.0.1 \
  --port 8934

Expose that locally running receiver with whatever public entrypoint you trust for machine-to-machine POSTs. For demos, a Cloudflare quick tunnel is usually fine. The important part is that StatusCake can reach:

https://<public-host>/webhooks/statuscake-alerts

And before you test from StatusCake, check both health endpoints:

curl http://127.0.0.1:8934/health
curl https://<public-host>/health

One small gotcha: loading /webhooks/statuscake-alerts in a browser proves nothing. It is a POST route, not a page. Use /health for GET checks and an actual POST when you test the webhook path.

Log first, validate second

One of the best decisions in this design is to write the inbound request to disk before doing anything clever with it.

That gives you two useful artifacts:

• var/last-webhook-payload.json for quick inspection
• var/incoming-webhooks.jsonl as an append-only inbound ledger

That means you can answer the boring but important questions later:

• Did the webhook arrive?
• Was the payload malformed?
• Did the sender include the token?
• Did Hermes reject it, or was it never delivered?

Without that log, webhook debugging turns into folklore.

Verification is where this gets interesting

This is the part that makes Hermes more than a relay.

When a down alert comes in, Hermes does not need to panic immediately. It can probe the target itself and decide whether the outage looks real.

A very simple version is enough:

• hit the health endpoint if you have one
• hit the public URL
• count failures
• only escalate if the failure threshold is met

The config supports that directly:

{
  "timezone": "Europe/London",
  "probe_timeout_seconds": 8,
  "min_failed_probes": 1,
  "probe_urls": [],
  "notifications": {
    "immediate": [
      {
        "type": "email",
        "transport": "sendmail",
        "from": "statuscake-hermes@localhost",
        "to": ["[e-mail]"],
        "events": ["DOWN_CONFIRMED", "UP_CONFIRMED"],
        "subject_prefix": "[StatusCake]"
      }
    ]
  }
}

A detail worth calling out: leaving probe_urls empty is not a bug. In this setup, that tells Hermes to verify against the website_url coming from the StatusCake payload. That is a good default when you want the monitor to carry its own target.

If you do have a better health endpoint than the public homepage, use it.

Time-aware escalation without buying more software

A lot of teams want basic routing rules but do not want a full paging product yet.

That is fine. Hermes can do the simple version well.

{
  "escalation_schedule": {
    "windows": [
      {
        "name": "business-hours",
        "start_hour": 8,
        "end_hour": 18,
        "target": "primary-on-call",
        "channel": "discord:#ops-business-hours"
      },
      {
        "name": "out-of-hours",
        "start_hour": 18,
        "end_hour": 24,
        "target": "after-hours-escalation",
        "channel": "discord:#ops-after-hours"
      },
      {
        "name": "night-shift",
        "start_hour": 0,
        "end_hour": 8,
        "target": "night-duty-engineer",
        "channel": "discord:#ops-night"
      }
    ]
  }
}

That is enough to stop every alert from behaving like a fire alarm.

Immediate email is often enough

You do not need to overcomplicate delivery either.

For this build, immediate alert delivery is handled through sendmail, which keeps the integration dead simple on a machine that already knows how to send mail.

A confirmed DOWN_CONFIRMED or UP_CONFIRMED event triggers an email right away. In our case, that means messages go straight to [e-mail].

The CLI can also send a daily summary on demand:

python3 scripts/statuscake_alert_engine.py send-daily-summary-email \
  --config var/escalation-config.json \
  --print-result

And if you want the recap every morning, schedule the wrapper script:

python3 scripts/statuscake_daily_summary_email.py

In this project, that job is scheduled for 09:00 Europe/London each day.

The event store is deliberately boring

Every verified event gets appended to var/alerts.jsonl.

That sounds plain because it is plain. That is also why it is useful.

A JSONL file gives you:

• a durable incident ledger
• easy grepping and ad hoc analysis
• simple daily summaries
• a way to compare raw StatusCake signals with Hermes-confirmed incidents

You do not need a dashboard before you have a history.

Start with the history.

A quick local test loop

Once the receiver is up, you can test the full path locally with a form-encoded POST that looks like StatusCake.

curl -X POST http://127.0.0.1:8934/webhooks/statuscake-alerts \
  -H 'Content-Type: application/x-www-form-urlencoded' \
  --data 'website_name=Example+Storefront&website_url=https%3A%2F%2Fexample.com&status=down&check_rate=300&test_id=123456&alert_type=uptime'

The response tells you a lot in one go:

• whether Hermes accepted the payload
• whether verification passed
• which escalation window matched
• whether the event was stored
• whether an email notification succeeded

That is the kind of tight feedback loop you want when you are building monitoring workflows.

What this setup is good for

This pattern is a good fit if you want:

• fewer false positive escalations
• lightweight time-based routing
• a raw audit trail of webhook delivery
• immediate email notifications without a big notification stack
• simple daily incident recaps

Final thought

The useful part of this setup is not that Hermes can receive a webhook. Lots of tools can receive a webhook.

The useful part is that Hermes adds judgment after detection.

StatusCake tells you that something might be broken. Hermes can check whether it really looks broken, decide who should hear about it, send the email, and keep the record for tomorrow morning's summary.

That is a much better workflow than forwarding every alert and hoping the humans sort it out.