diff --git a/CLAUDE.md b/CLAUDE.md
new file mode 100644
index 0000000..48e34df
--- /dev/null
+++ b/CLAUDE.md
@@ -0,0 +1,45 @@
+# CLAUDE.md
+
+## Repository Purpose
+
+Centralized CodeRabbit AI code review configuration. Contains a single `coderabbit.yaml`
+referenced by other repositories via `remote_config`. No source code — configuration only.
+
+## Validation Commands
+
+```bash
+# Run all checks (YAML lint, schema validation, editorconfig)
+pre-commit run --all-files
+
+# Individual checks
+yamllint --strict .                # YAML linting
+check-jsonschema --schemafile \
+  'https://storage.googleapis.com/coderabbit_public_assets/schema.v2.json' \
+  coderabbit.yaml                  # Schema validation
+```
+
+CI runs `pre-commit/action@v3.0.1` on push/PR to `main`.
+
+## Architecture
+
+```
+coderabbit.yaml          # Main config — schema.v2.json validated
+.yamllint.yaml           # yamllint rules (120 char lines, strict)
+.pre-commit-config.yaml  # Pre-commit hooks: check-yaml, yamllint, schema, editorconfig
+.editorconfig            # 2-space indent, UTF-8, LF line endings
+.github/workflows/       # CI validation via pre-commit
+```
+
+Other repos consume this config by adding to their `.coderabbit.yaml`:
+```yaml
+remote_config:
+  url: "https://raw.githubusercontent.com/ivuorinen/coderabbit/refs/heads/main/coderabbit.yaml"
+```
+
+## Conventions
+
+- YAML files start with `---` document marker
+- 2-space indentation, UTF-8, LF line endings
+- Max line length: 120 characters
+- Schema reference: `# yaml-language-server: $schema=https://storage.googleapis.com/coderabbit_public_assets/schema.v2.json`
+- Always validate against the official CodeRabbit schema before committing
diff --git a/coderabbit.yaml b/coderabbit.yaml
index ee6e0fc..a8a0ee8 100644
--- a/coderabbit.yaml
+++ b/coderabbit.yaml
@@ -84,6 +84,14 @@ reviews:
         Do not enforce current coding standards or suggest updates to align
         with current practices. Focus only on grammar, spelling, and
         readability if changes are being reviewed.
+    - path: ".serena/**"
+      instructions: >-
+        .serena directory contains Serena MCP tool configuration and
+        AI-generated project memories. These files are maintained by
+        tooling and do not represent hand-written source code.
+        Do not enforce coding standards or suggest refactoring. Focus
+        only on grammar, spelling, and readability if memory files
+        are being reviewed. Do not review cache files.
     - path: "**/*.{test,spec}.*"
       instructions: >-
         These are test files. Focus on test coverage completeness,