feat(#622): structure the ci discord alert and document the alarms format

.github/workflows/main.yml

@@ -411,9 +411,12 @@ jobs:
             exit 0
           fi
           subject="$(printf '%s' "${COMMIT_MSG}" | head -n 1)"
+          # Canonical alarms embed: [source] severity — what fired, one link to
+          # the run, and the critical hue shared with the Axiom notifiers.
           payload="$(jq -n \
-            --arg title "main pipeline failed: ${subject}" \
+            --arg title "[CI] critical — main pipeline failed" \
             --arg url "${RUN_URL}" \
-            --arg desc "${FAILED}" \
-            '{embeds: [{title: $title, url: $url, description: $desc, color: 15548997}]}')"
+            --arg desc "${subject}" \
+            --arg failed "${FAILED}" \
+            '{embeds: [{title: $title, url: $url, description: $desc, color: 15026253, fields: [{name: "Failed jobs", value: $failed}]}]}')"
           curl -fsS -H 'content-type: application/json' -d "${payload}" "${WEBHOOK_URL}"

apps/bugsink/README.md

@@ -46,6 +46,10 @@ Discord messaging-service config pointing at the channel's incoming webhook (the
 webhook added in its settings. No outgoing email is configured — alerting is webhook-only, and
 per-project alert toggles live in each Bugsink project's settings.
 
+Bugsink's Discord messaging service renders a fixed body the app doesn't template, so these alerts
+keep their own shape rather than the structured embed the CI path posts to the same channel
+(`docs/architecture/platform/observability.md` § Alarms channel).
+
 ## Housekeeping
 
 The `.github/workflows/bugsink-vacuum.yml` workflow runs `bugsink-manage vacuum_files` monthly (and

docs/architecture/platform/observability.md

@@ -178,3 +178,11 @@ The `vers verification lag` threshold monitor watches `vers.verification.lag` an
 `vers alarms`. It alerts on no data as well as on the threshold. The gauge exports from
 `service-replay`'s always-warm machine ([deployment](./deployment.md)), so a silent dataset means
 the exporter or the process around it is down, never a healthy quiet system.
+
+## Alarms channel
+
+Axiom monitors, the CI pipeline, and Bugsink post to one Discord channel. The CI `alert` job posts a
+structured embed — a `[CI] critical — …` title, the failing run's link, and a red severity colour
+(`#e5484d`, decimal `15026253`) — in `.github/workflows/main.yml`. Axiom and Bugsink post their
+tools' stock formats: Axiom's custom-webhook notifier, the one templated body it offers, is not
+enabled on the plan, and Bugsink's Discord messaging service exposes no templating.