ci(town-crier): config-error fails loud, bus-hiccup fails open (review feedback)

Author: jasperboerhof

.github/workflows/announce-pr.yml

@@ -7,13 +7,15 @@
 # Produces BOTH sides of a request's lifecycle, so the bus never drifts from GitHub:
 #   - announce: when "Agent Review Requested" lands on a PR, tell the crier once.
 #   - resolve:  when that PR closes/merges, retire its thread — otherwise a landed PR
-#               sits "open" on the bus forever (there is no GitHub->bus merge sync) and
-#               joined harnesses keep getting offered already-merged work.
+#               sits "open" on the bus forever (there is no GitHub->bus merge sync).
 # Joined harnesses pick up open requests from the bus — this workflow does NOT poll or review.
 #
-# The bus is fail-open: a town-crier hiccup (cold start, transient 5xx) must never red a
-# PR's checks, so both jobs use continue-on-error + a curl --max-time. Neither job touches
-# the GITHUB_TOKEN (they auth to the bus with TOWN_CRIER_TOKEN), so permissions are dropped.
+# Failure policy (two distinct modes, so a real problem is never silently masked):
+#   - MISSING provisioning (no TOWN_CRIER_URL/TOKEN) is a config error -> fail LOUD (red).
+#   - A bus HICCUP (cold start / transient 5xx / timeout) -> ::warning:: + stay GREEN (fail-open;
+#     a coordination-layer blip must never red a contributor's PR checks).
+# Neither job uses the GITHUB_TOKEN (they auth to the bus with TOWN_CRIER_TOKEN), so permissions
+# are dropped to nothing.
 name: town-crier producer (announce + resolve)
 
 permissions: {}
@@ -28,7 +30,6 @@ jobs:
     runs-on: ubuntu-latest
     steps:
       - name: Announce to the crier
-        continue-on-error: true
         env:
           CRIER_URL: ${{ vars.TOWN_CRIER_URL }}
           CRIER_TOKEN: ${{ secrets.TOWN_CRIER_TOKEN }}
@@ -37,7 +38,13 @@ jobs:
           TITLE: ${{ github.event.pull_request.title }}
           REQUESTER: ${{ github.event.pull_request.user.login }}
         run: |
+          # Missing provisioning is a config error — fail LOUD so it can't pass silently.
+          if [ -z "$CRIER_URL" ] || [ -z "$CRIER_TOKEN" ]; then
+            echo "::error::town-crier not provisioned — set the TOWN_CRIER_URL variable + TOWN_CRIER_TOKEN secret"
+            exit 1
+          fi
           # jq builds the JSON so a PR title with quotes can't break the payload.
+          # A bus hiccup is not the PR's fault — degrade to a warning, keep the check green.
           curl -fsS --max-time 10 -X POST "$CRIER_URL/announce" \
             -H "Authorization: Bearer $CRIER_TOKEN" \
             -H "Content-Type: application/json" \
@@ -46,22 +53,27 @@ jobs:
               --arg repo "$REPO" \
               --arg title "$TITLE" \
               --arg requester "$REQUESTER" \
-              '{pr_url:$pr, repo:$repo, title:$title, requester:$requester}')"
+              '{pr_url:$pr, repo:$repo, title:$title, requester:$requester}')" \
+            || echo "::warning::town-crier announce failed (transient bus issue?) — not blocking the PR"
 
   resolve:
     if: github.event.action == 'closed' && contains(github.event.pull_request.labels.*.name, 'Agent Review Requested')
     runs-on: ubuntu-latest
     steps:
       - name: Resolve on the crier
-        continue-on-error: true
         env:
           CRIER_URL: ${{ vars.TOWN_CRIER_URL }}
           CRIER_TOKEN: ${{ secrets.TOWN_CRIER_TOKEN }}
           PR_URL: ${{ github.event.pull_request.html_url }}
           MERGED: ${{ github.event.pull_request.merged }}
         run: |
+          if [ -z "$CRIER_URL" ] || [ -z "$CRIER_TOKEN" ]; then
+            echo "::error::town-crier not provisioned — set the TOWN_CRIER_URL variable + TOWN_CRIER_TOKEN secret"
+            exit 1
+          fi
           NOTE=$([ "$MERGED" = "true" ] && echo "merged" || echo "closed without merge")
           curl -fsS --max-time 10 -X POST "$CRIER_URL/resolve" \
             -H "Authorization: Bearer $CRIER_TOKEN" \
             -H "Content-Type: application/json" \
-            -d "$(jq -n --arg pr "$PR_URL" --arg note "$NOTE" '{pr_url:$pr, note:$note}')"
+            -d "$(jq -n --arg pr "$PR_URL" --arg note "$NOTE" '{pr_url:$pr, note:$note}')" \
+            || echo "::warning::town-crier resolve failed (transient bus issue?) — not blocking the PR"