anthropic-watch

A tiny Cloudflare Worker that polls status.anthropic.com every 2 minutes and DMs Wesley on Slack via alerthub-ingress when the Anthropic API degrades.

The whole point of this Worker is Anthropic-outage Disaster Recovery: when Claude API is degraded, mayors may go silent. The notify path here uses Gemini 2.5 Flash via OpenRouter to summarise — no Claude / Anthropic SDK anywhere in the hot path — so the alert lands even when Anthropic is down.

Architecture

       ┌─────────────────────────────────────┐
       │  Cron Trigger (*/2 * * * *)         │
       └────────────────┬────────────────────┘
                        ▼
        ┌────────────────────────────────────┐
        │ status.anthropic.com summary.json  │
        └────────────────┬───────────────────┘
                        ▼
                  diff vs KV
              (last_indicator,
               last_seen_incidents)
                        │
              transition? ─── no ──▶ update KV, exit
                        │
                       yes
                        ▼
              Gemini 2.5 Flash via
              OpenRouter (terse JSON
              {title, body})
                        │
                        ▼
         HMAC-signed POST /alert
              alerthub-ingress
              (proactive:anthropic-watch)
                        │
                        ▼
         Slack DM to Wesley via
              alerthub-fanout

Producer identity

This Worker is registered with Alert Hub as a proactive producer:

Field	Value
producer_type	proactive
producer_id	anthropic-watch
HMAC env var	HMAC_PROACTIVE_ANTHROPIC_WATCH
Origin header	proactive:anthropic-watch

Signing matches verifyProducerHmac in alert-hub (src/lib/hmac.ts):

X-Gastown-Origin:    proactive:anthropic-watch
X-Gastown-Timestamp: <ISO-8601>
X-Gastown-Signature: hex sha256( HMAC(secret, `${timestamp}\n${body}`) )

Kill switch

ANTHROPIC_WATCH_ENABLED must be "true" or the cron handler early-returns. Defaults to "false" so first-deploy is safe. Flip it via:

echo true | wrangler secret put ANTHROPIC_WATCH_ENABLED

Secrets

Set once with wrangler secret put:

Secret	Source
OPENROUTER_API_KEY	Doppler gastown/prd
HMAC_PROACTIVE_ANTHROPIC_WATCH	Mint fresh; mirror to alerthub-ingress
ANTHROPIC_WATCH_ENABLED	"true" after smoke test

KV namespace

wrangler kv namespace create ANTHROPIC_WATCH_STATE
# Paste the id into wrangler.toml under [[kv_namespaces]].

Smoke test

1. Deploy with ANTHROPIC_WATCH_ENABLED=false.
2. Seed a fake "degraded" baseline so the next tick reads a transition:

   wrangler kv key put --binding=STATE watch_state_v1 \
     '{"last_indicator":"none","last_seen_incident_ids":["FAKE-SEED-ID"],"last_tick_at":"2026-05-12T00:00:00Z"}'
   echo true | wrangler secret put ANTHROPIC_WATCH_ENABLED
   curl https://anthropic-watch.<account>.workers.dev/tick

3. Confirm Slack DM lands within 30s.
4. Reset KV to actual current status (or just let next cron tick overwrite).

Deploys

Production deploys run via .github/workflows/deploy.yml on push to main. First-deploy bootstrap can be a single direct wrangler deploy (one-time exception per Wesley's no-hot-architecture-changes rule). All subsequent deploys go through the GitHub Actions pipeline.

License

MIT — see LICENSE.