Harden release workflow and regex against CodeQL findings

Address three code-scanning alerts on the PR:

- Critical (actions/untrusted-checkout): split release.yaml into a
  read-only `prepare` job that checks out and runs repo code, and a
  privileged `publish` job (contents:write + OIDC) that never checks out
  repo code — it tags, releases, and invokes purely through the GitHub
  and AWS APIs. Also assert head_branch == main.
- High x2 (py/polynomial-redos): rewrite the italic and link regexes in
  markdown_to_mrkdwn with possessive quantifiers and exclusive character
  classes so they run in linear time on adversarial input. Adds a
  regression test.

.github/workflows/release.yaml

@@ -1,36 +1,42 @@
 name: Release
 
 # Runs after a successful Deploy. When the top of CHANGELOG.md names a version
-# that has no tag yet, this tags it, publishes a GitHub Release with the notes,
+# that has no Release yet, this tags it, publishes a GitHub Release with the
-# and — for minor/major bumps only — invokes the release-notifier Lambda to
+# notes, and — for minor/major bumps only — invokes the release-notifier Lambda
-# announce it in Slack. Triggering on Deploy completion (not release:published /
+# to announce it in Slack. Triggering on Deploy completion (not release:published
-# tag push) is deliberate: GITHUB_TOKEN-created events do not start downstream
+# / tag push) is deliberate: GITHUB_TOKEN-created events do not start downstream
 # workflows, and gating on Deploy success means we never announce a version that
 # is not actually live.
+#
+# Two jobs by design (CodeQL actions/untrusted-checkout): the only job that
+# checks out and runs repo code (`prepare`) is read-only and unprivileged; the
+# job that holds write + OIDC (`publish`) never checks out repo code — it acts
+# purely through the GitHub and AWS APIs.
 on:
   workflow_run:
     workflows: ["Deploy"]
     types: [completed]
 
-permissions:
+# Serialize so back-to-back releases announce in order, never overlapping.
-  contents: write # create the tag + GitHub Release
-  id-token: write # OIDC to assume the notifier-invoke role
-
-# Serialize so back-to-back releases announce in order (queued Deploys -> queued
-# Releases), never overlapping.
 concurrency:
   group: release-announce
   cancel-in-progress: false
 
 jobs:
-  release:
+  prepare:
-    if: ${{ github.event.workflow_run.conclusion == 'success' }}
+    # Deploy only runs on push to main, so head_sha is always a trusted main
+    # commit; assert head_branch == main to make that boundary explicit.
+    if: ${{ github.event.workflow_run.conclusion == 'success' && github.event.workflow_run.head_branch == 'main' }}
     runs-on: ubuntu-latest
+    permissions:
+      contents: read
+    outputs:
+      release: ${{ steps.rel.outputs.release }}
+      version: ${{ steps.rel.outputs.version }}
+      kind: ${{ steps.rel.outputs.kind }}
     steps:
       - uses: actions/checkout@v6
         with:
-          # The exact commit Deploy deployed — NOT the branch HEAD, which a later
-          # merge may have already moved past.
           ref: ${{ github.event.workflow_run.head_sha }}
           fetch-depth: 0
           fetch-tags: true
@@ -41,6 +47,8 @@ jobs:
 
       - name: Determine release
         id: rel
+        env:
+          GH_TOKEN: ${{ github.token }}
         run: |
           TOP=$(python scripts/changelog_cli.py top-version CHANGELOG.md)
           if [ -z "$TOP" ]; then
@@ -52,52 +60,64 @@ jobs:
           PREV="${PREV:-v0.0.0}"
           KIND=$(python scripts/changelog_cli.py bump-kind CHANGELOG.md "$PREV")
 
-          TAG_EXISTS=false
-          git rev-parse "v$TOP" >/dev/null 2>&1 && TAG_EXISTS=true
           RELEASE_EXISTS=false
           gh release view "v$TOP" >/dev/null 2>&1 && RELEASE_EXISTS=true
 
-          echo "version=$TOP"        >> "$GITHUB_OUTPUT"
+          echo "version=$TOP" >> "$GITHUB_OUTPUT"
-          echo "kind=$KIND"          >> "$GITHUB_OUTPUT"
+          echo "kind=$KIND"   >> "$GITHUB_OUTPUT"
-          echo "tag_exists=$TAG_EXISTS"        >> "$GITHUB_OUTPUT"
+          # Gate the publish job on a clean bump whose Release isn't published yet.
-          echo "release_exists=$RELEASE_EXISTS" >> "$GITHUB_OUTPUT"
-          # "release" gates the rest: a clean bump whose Release isn't published yet.
           if [ "$KIND" != "none" ] && [ "$RELEASE_EXISTS" = "false" ]; then
             echo "release=true" >> "$GITHUB_OUTPUT"
           else
             echo "release=false" >> "$GITHUB_OUTPUT"
             echo "v$TOP: kind=$KIND release_exists=$RELEASE_EXISTS — no action."
           fi
-        env:
-          GH_TOKEN: ${{ github.token }}
 
-      # Each artifact is created independently and idempotently so a re-run after
+      - name: Build release notes
-      # a mid-job failure can finish the release rather than skip it forever.
-      - name: Create tag
-        if: ${{ steps.rel.outputs.release == 'true' && steps.rel.outputs.tag_exists == 'false' }}
-        run: |
-          V="v${{ steps.rel.outputs.version }}"
-          git tag -a "$V" -m "$V"
-          git push origin "$V"
-
-      - name: Build payload
         if: ${{ steps.rel.outputs.release == 'true' }}
         run: |
           python scripts/changelog_cli.py payload CHANGELOG.md "${{ steps.rel.outputs.version }}" > payload.json
           python -c "import json; print(json.load(open('payload.json'))['notes'])" > notes.md
 
+      - name: Upload notes artifact
+        if: ${{ steps.rel.outputs.release == 'true' }}
+        uses: actions/upload-artifact@v4
+        with:
+          name: release-notes
+          path: |
+            payload.json
+            notes.md
+          retention-days: 1
+
+  publish:
+    needs: prepare
+    if: ${{ needs.prepare.outputs.release == 'true' }}
+    runs-on: ubuntu-latest
+    permissions:
+      contents: write # create the tag + GitHub Release
+      id-token: write # OIDC to assume the notifier-invoke role
+    env:
+      VERSION: ${{ needs.prepare.outputs.version }}
+      KIND: ${{ needs.prepare.outputs.kind }}
+      HEAD_SHA: ${{ github.event.workflow_run.head_sha }}
+    steps:
+      - uses: actions/download-artifact@v4
+        with:
+          name: release-notes
+
       # Announce BEFORE publishing the Release: the Release is the durable "done"
-      # marker, so announcing first keeps the step retryable. Minor/major only,
+      # marker (prepare skips once it exists), so announcing first keeps this
-      # and only once the invoke-role variable has been bootstrapped (see README).
+      # retryable. Minor/major only, and only once the invoke-role variable has
+      # been bootstrapped (see README).
       - name: Configure AWS credentials
-        if: ${{ steps.rel.outputs.release == 'true' && steps.rel.outputs.kind != 'patch' && vars.RELEASE_NOTIFY_INVOKE_ROLE_ARN != '' }}
+        if: ${{ env.KIND != 'patch' && vars.RELEASE_NOTIFY_INVOKE_ROLE_ARN != '' }}
         uses: aws-actions/configure-aws-credentials@v6
         with:
           role-to-assume: ${{ vars.RELEASE_NOTIFY_INVOKE_ROLE_ARN }}
           aws-region: us-east-1
 
       - name: Announce in Slack
-        if: ${{ steps.rel.outputs.release == 'true' && steps.rel.outputs.kind != 'patch' && vars.RELEASE_NOTIFY_INVOKE_ROLE_ARN != '' }}
+        if: ${{ env.KIND != 'patch' && vars.RELEASE_NOTIFY_INVOKE_ROLE_ARN != '' }}
         run: |
           aws lambda invoke \
             --function-name afterhours-release-notifier \
@@ -108,18 +128,19 @@ jobs:
           if grep -q '"FunctionError"' invoke-meta.json; then
             echo "::error::release-notifier returned an error"; cat response.json; exit 1
           fi
-          echo "Announced v${{ steps.rel.outputs.version }}."
+          echo "Announced v${VERSION}."
 
       - name: Warn if announcement skipped (not bootstrapped)
-        if: ${{ steps.rel.outputs.release == 'true' && steps.rel.outputs.kind != 'patch' && vars.RELEASE_NOTIFY_INVOKE_ROLE_ARN == '' }}
+        if: ${{ env.KIND != 'patch' && vars.RELEASE_NOTIFY_INVOKE_ROLE_ARN == '' }}
-        run: echo "::warning::RELEASE_NOTIFY_INVOKE_ROLE_ARN is unset — tagged + released but did not announce. Set the repo variable from the stack output."
+        run: echo "::warning::RELEASE_NOTIFY_INVOKE_ROLE_ARN is unset — tagging + releasing but not announcing. Set the repo variable from the stack output."
 
+      # No checkout: gh creates the tag at HEAD_SHA and the Release together.
       - name: Publish GitHub Release
-        if: ${{ steps.rel.outputs.release == 'true' }}
-        run: |
-          gh release create "v${{ steps.rel.outputs.version }}" \
-            --title "v${{ steps.rel.outputs.version }}" \
-            --notes-file notes.md \
-            --target "${{ github.event.workflow_run.head_sha }}"
         env:
           GH_TOKEN: ${{ github.token }}
+        run: |
+          gh release create "v${VERSION}" \
+            --repo "${{ github.repository }}" \
+            --title "v${VERSION}" \
+            --notes-file notes.md \
+            --target "${HEAD_SHA}"

src/shared/shared/blocks.py

@@ -19,11 +19,15 @@ def markdown_to_mrkdwn(text: str) -> str:
     ``*italic*``, ``[text](url)``, ``- `` bullets), but Slack uses a different
     dialect (``*bold*``, ``_italic_``, ``<url|text>``, ``•`` bullets). Bold is
     swapped via a placeholder first so the italic pass cannot mangle it.
+
+    The italic and link patterns use possessive quantifiers (``++``) and exclusive
+    character classes so they run in linear time on adversarial input — no
+    catastrophic backtracking (py/polynomial-redos).
     """
-    text = re.sub(r"\*\*(.+?)\*\*", "\x00\\1\x00", text)  # bold -> placeholder
+    text = re.sub(r"\*\*([^\n]+?)\*\*", "\x00\\1\x00", text)  # bold -> placeholder
-    text = re.sub(r"(?<!\*)\*(?!\*)(.+?)(?<!\*)\*(?!\*)", r"_\1_", text)  # italic
+    text = re.sub(r"\*([^*\n]++)\*", r"_\1_", text)  # italic (single asterisks)
     text = text.replace("\x00", "*")  # placeholder -> Slack bold
-    text = re.sub(r"\[([^\]]+)\]\(([^)]+)\)", r"<\2|\1>", text)  # links
+    text = re.sub(r"\[([^\]]++)\]\(([^)\n]++)\)", r"<\2|\1>", text)  # links
     text = re.sub(r"(?m)^(\s*)[-*]\s+", r"\1•  ", text)  # bullets
     return text
 

tests/shared/test_blocks.py

@@ -178,6 +178,17 @@ class TestReleaseAnnouncement:
         assert "•  see <http://x|docs>" in out
         assert "•  next" in out
 
+    def test_adversarial_input_runs_in_linear_time(self):
+        # py/polynomial-redos regression: possessive quantifiers must keep these
+        # patterns from catastrophic backtracking. Pathological inputs that would
+        # hang a backtracking engine complete effectively instantly here.
+        import time
+
+        for evil in ("*" + "*a" * 4000, "[" + "[\\(" * 4000, "[" + "](" * 4000):
+            start = time.perf_counter()
+            markdown_to_mrkdwn(evil)
+            assert time.perf_counter() - start < 1.0
+
     def test_blocks_have_header_and_notes(self):
         blocks = build_release_announcement_blocks(
             "1.10.0", "**Release notes** now self-announce.", "June 11, 2026"