diff --git a/.github/workflows/notify-main-failure.yml b/.github/workflows/notify-main-failure.yml
index cabe027..f3e0321 100644
--- a/.github/workflows/notify-main-failure.yml
+++ b/.github/workflows/notify-main-failure.yml
@@ -114,9 +114,36 @@ jobs:
           WORKFLOW_REF: ${{ github.workflow_ref }}
           BRANCH: ${{ github.ref_name }}
           RUN_ID: ${{ github.run_id }}
+          RUN_ATTEMPT: ${{ github.run_attempt }}
         run: |
           set -uo pipefail
 
+          # A RE-RUN is the most common way a failure gets fixed, and it was the
+          # one case this could never report. Attempts share a run id, so the
+          # failing attempt is the very run the history lookup below excludes as
+          # "the current one" — it would then find some older, green run and
+          # conclude nothing had broken. Check the previous attempt first.
+          if [ "${RUN_ATTEMPT:-1}" -gt 1 ]; then
+            PREV_ATTEMPT=$((RUN_ATTEMPT - 1))
+            if LAST=$(gh api "repos/${REPO}/actions/runs/${RUN_ID}/attempts/${PREV_ATTEMPT}" \
+                        --jq '.conclusion // ""' 2>&1); then
+              case "${LAST}" in
+                failure|timed_out|startup_failure)
+                  echo "should_post=true" >> "$GITHUB_OUTPUT"
+                  echo "prev_conclusion=${LAST}" >> "$GITHUB_OUTPUT"
+                  echo "Attempt ${PREV_ATTEMPT} concluded '${LAST}', this attempt recovered it."
+                  exit 0
+                  ;;
+                *)
+                  echo "should_post=false" >> "$GITHUB_OUTPUT"
+                  echo "Attempt ${PREV_ATTEMPT} concluded '${LAST:-unknown}', not a recovery."
+                  exit 0
+                  ;;
+              esac
+            fi
+            echo "Could not read attempt ${PREV_ATTEMPT}, falling back to run history: ${LAST}"
+          fi
+
           # Match the workflow FILE rather than its display name, which a
           # `run-name:` override changes; fall back to the name if the ref does
           # not resolve to a path inside this repo.
diff --git a/.github/workflows/workflow-lint.yml b/.github/workflows/workflow-lint.yml
index 0dad854..dbc11f1 100644
--- a/.github/workflows/workflow-lint.yml
+++ b/.github/workflows/workflow-lint.yml
@@ -151,13 +151,20 @@ jobs:
           GH_TOKEN: ${{ github.token }}
         run: |
           set -uo pipefail
+          # The shared policy travels with the scripts, so every repo audits under
+          # the same rules and a decision recorded once is not re-litigated per
+          # repo. A repo with its own .github/zizmor.yml still wins, since zizmor
+          # prefers a repo-local config over --config.
+          CFG=(--config .ci-shared/zizmor.yml)
+          [ -f .github/zizmor.yml ] && CFG=()
+
           if [ "$BLOCKING" = "true" ]; then
-            uvx "zizmor@${VERSION}" --persona="$PERSONA" .github/workflows
+            uvx "zizmor@${VERSION}" "${CFG[@]}" --persona="$PERSONA" .github/workflows
           else
             # Advisory: report and move on. The findings land in the log and the
             # step summary either way, so turning this into a gate later is a
             # one-line input change rather than a discovery exercise.
-            uvx "zizmor@${VERSION}" --persona="$PERSONA" .github/workflows | tee zizmor.out || true
+            uvx "zizmor@${VERSION}" "${CFG[@]}" --persona="$PERSONA" .github/workflows | tee zizmor.out || true
             {
               echo "### zizmor (advisory)"
               echo
diff --git a/zizmor.yml b/zizmor.yml
new file mode 100644
index 0000000..2b5aeed
--- /dev/null
+++ b/zizmor.yml
@@ -0,0 +1,31 @@
+# Shared zizmor policy for every repo that calls workflow-lint.yml.
+#
+# One file rather than seven, for the same reason the pipelines were unified: a
+# per-repo copy drifts, and the drift is invisible until the signal is wrong.
+#
+# ONE audit is turned down here, and only partly, because it flags a DECISION
+# rather than a defect. Everything else stays on: if a finding survives this file
+# it is worth reading, which is the only way the report keeps meaning anything.
+#
+# Deliberately NOT suppressed, though it is the largest remaining group:
+# `secrets-inherit`. `secrets: inherit` is our standard everywhere, so it will
+# report on nearly every caller — but narrowing those to explicit secret lists is
+# a legitimate hardening step nobody has decided against yet, and silencing it
+# here would quietly make that decision. Suppress it when someone chooses to.
+
+rules:
+  # In-org references float on `@main` deliberately (decision 11.BH). 101
+  # composite-action call sites already do, and keeping 150+ SHA pins bumped as
+  # the reusables evolve costs more than it buys — so the control moved to the
+  # source instead, where it is one gate rather than 150: `main` in
+  # mindsdb/github-actions requires a code-owner review, dismisses stale
+  # approvals, and admits no bypass actors.
+  #
+  # THIRD-PARTY actions get no such exemption. A tag there is mutable and an
+  # upstream account compromise reaches our pipelines with the caller's secrets in
+  # scope, which is a real finding and tracked as ENG-1053 hardening item 1.
+  unpinned-uses:
+    config:
+      policies:
+        mindsdb/*: ref-pin
+        "*": hash-pin