diff --git a/.coderabbit.yaml b/.coderabbit.yaml
index 57174d8..9eb5e72 100644
--- a/.coderabbit.yaml
+++ b/.coderabbit.yaml
@@ -9,6 +9,15 @@ reviews:
   auto_review:
     enabled: true
     drafts: false
+    # Documentation-only PRs are not reviewed (conventional "docs" titles).
+    ignore_title_keywords:
+      - "docs:"
+      - "docs("
+  # Documentation and specs are not reviewed.
+  path_filters:
+    - "!**/*.md"
+    - "!docs/**"
+    - "!openspec/**"
   path_instructions:
     - path: "apps/**/*.py"
       instructions: >-
diff --git a/.github/pull_request_template.md b/.github/pull_request_template.md
index 91357e2..cd1f9a3 100644
--- a/.github/pull_request_template.md
+++ b/.github/pull_request_template.md
@@ -15,25 +15,11 @@ profiles, screenshots containing credentials, or downloaded customer documents.
 
 State whether AI assisted this change and what the author verified.
 
-## CodeRabbit review
-
-Public repositories with fewer than 10 stars require a manual review request,
-even when automatic reviews are enabled. Bot-authored PRs may also be skipped.
-
-Once this PR is open and its changes are ready, comment `@coderabbitai review`
-on the PR to request an incremental review. Use `@coderabbitai full review`
-only when a complete new pass is needed. These commands are instructions for
-an actual PR comment; text in this PR description does not trigger them.
-
-- [ ] Request a CodeRabbit review on the final PR head.
-- [ ] Verify that the review completed and covers that head; a walkthrough,
-      summary, skipped notice or pending status alone is not review evidence.
-- [ ] Address confirmed findings and record the review link and reviewed head
-      SHA. If review cannot complete, record the reason for the maintainer.
-
-If more commits are pushed after the review, request another incremental pass.
-Do not treat CodeRabbit as a substitute for CI or the repository's required
-human/independent review. Maintainers decide whether to merge if CodeRabbit
-is unavailable.
+## Pullfrog review
+
+Pullfrog is the sole reviewer. Require a completed review of the final head,
+resolve actionable threads and keep all CI checks green. Ignore CodeRabbit and
+the known sticky `pullfrog-approval` status only. PRs that change only
+`*.md` files need CI only.
 
 Review evidence: <!-- completed review URL and reviewed commit SHA -->
diff --git a/.github/workflows/pullfrog.yml b/.github/workflows/pullfrog.yml
index 7795a7f..7a4c145 100644
--- a/.github/workflows/pullfrog.yml
+++ b/.github/workflows/pullfrog.yml
@@ -20,12 +20,33 @@ jobs:
     permissions:
       id-token: write
       contents: read
+      pull-requests: read
     steps:
+      # Local addition: documentation-only PRs (only *.md files, docs/ or
+      # openspec/, the pr-gate rule; doc/ assets still get a review) are not
+      # reviewed. Fails open: without a PR number or file list, the review runs.
+      - name: Skip documentation-only PRs
+        id: scope
+        env:
+          GH_TOKEN: ${{ github.token }}
+          RUN_NAME: ${{ inputs.name }}
+        run: |
+          review=true
+          pr=$(printf '%s' "$RUN_NAME" | sed -nE 's/^Review (new commits on )?#([0-9]+).*/\2/p')
+          if [ -n "$pr" ]; then
+            files=$(gh pr diff "$pr" --repo "$GITHUB_REPOSITORY" --name-only || true)
+            if [ -n "$files" ] && ! printf '%s\n' "$files" | grep -qvE '^(docs/|openspec/)|\.md$'; then
+              review=false
+            fi
+          fi
+          echo "review=$review" >> "$GITHUB_OUTPUT"
       - name: Checkout code
+        if: steps.scope.outputs.review == 'true'
         uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
         with:
           fetch-depth: 1
       - name: Run agent
+        if: steps.scope.outputs.review == 'true'
         uses: pullfrog/pullfrog@ae4145b3d72b3f4aa045e2352dc6b2c7538daeeb # deployed on flowcool/ghostfolio-ibkr-sync
         with:
           prompt: ${{ inputs.prompt }}