diff --git a/.coderabbit.yaml b/.coderabbit.yaml
index f2e7050609e0..66ea63486404 100644
--- a/.coderabbit.yaml
+++ b/.coderabbit.yaml
@@ -50,6 +50,100 @@ reviews:
 
     # Test fixtures and data
     - "!**/testdata/**"
+  path_instructions:
+    # --- Rules derived from PR review patterns ---
+    # Generated by teams:coderabbit-rules-from-pr-reviews (2026-03-27)
+    # Analyzed 100 recent merged PRs, 480 human review comments from 24 reviewers
+
+    - path: "api/**/*_types.go"
+      instructions: |
+        Do not use +kubebuilder:validation:Pattern for string validation. Use CEL
+        XValidation rules instead with a human-readable message explaining the
+        expected format. For example, instead of:
+          +kubebuilder:validation:Pattern=`^arn:`
+        Use:
+          +kubebuilder:validation:XValidation:rule="self.matches('^arn:...')",message="must be a valid ARN in the format arn:<partition>:..."
+
+        Enum values in kubebuilder:validation:Enum markers must use PascalCase
+        (e.g., Standard, IO1, GP2) following Kubernetes API conventions. Avoid
+        lowercase (e.g., "enabled"/"disabled") or camelCase enum values.
+        Also prefer specific descriptive terms over generic Enable/Disable
+        (e.g., HTTPEndpoint/None instead of Enabled/Disabled).
+
+        Every API field with constrained values (enums, formatted strings, ranges)
+        must document in the godoc comment:
+        1. The list of valid values and what each means to an end user
+        2. The expected format for string fields (e.g., "must be a valid IPv4 or IPv6 address")
+        3. The behavior when the field is omitted (e.g., "When omitted, the platform
+           is left to choose a reasonable default, which is subject to change over time.
+           The current default is ...")
+        Do not assume users will read kubebuilder markers — the godoc should be
+        self-contained.
+
+        Follow OpenShift API conventions for pointer usage:
+        - Do NOT use pointers for struct fields when the zero value is not a valid
+          user choice. Use value types with omitzero instead.
+        - Do NOT use pointers for enum fields unless empty string "" is a valid enum
+          value.
+        - Do NOT use pointers for numeric fields when the minimum value is > 0.
+        - Use omitzero (not omitempty) for struct value types — omitempty does
+          nothing for non-pointer structs.
+        - Only use pointers when you need to distinguish between "not set" and
+          "explicitly set to zero value" and both are valid user choices.
+        See: https://github.com/openshift/enhancements/blob/master/dev-guide/api-conventions.md
+
+        All string fields in API types must have both MinLength and MaxLength
+        kubebuilder validation markers unless there is a documented reason not to.
+        Common values:
+        - IPv4/IPv6 addresses: MinLength=3, MaxLength=39
+        - DNS names: MinLength=1, MaxLength=253
+        - Kubernetes resource names: MinLength=1, MaxLength=253
+        Empty string is rarely a valid value — if not, set MinLength=1 at minimum.
+
+        When a struct has mutually exclusive sub-structs (e.g., platform-specific
+        config), use a properly marked discriminated union:
+        1. Add +union marker to the parent struct
+        2. Add +unionDiscriminator to the discriminator field
+        3. Add +unionMember to each member field
+        4. Add bidirectional CEL validation that enforces "required when discriminator
+           matches AND forbidden otherwise":
+           rule="self.type == 'Foo' ? has(self.foo) : !has(self.foo)"
+        5. Consider placing the union in a child struct to keep the parent clean for
+           future non-union fields.
+        Do not use separate has() rules without the forbidden-otherwise check.
+
+        Do not import and embed upstream Kubernetes types (e.g.,
+        corev1.LocalObjectReference, corev1.SecretReference) in CRD API types.
+        Create your own minimal reference types instead. Upstream types can change
+        unexpectedly or add fields you don't want to support, breaking your API
+        contract. Only import upstream types when passing them through verbatim to
+        another Kubernetes API.
+
+        Do not use +patchStrategy or +patchMergeKey markers on CRD types — they
+        have no effect. CRDs use Server-Side Apply (SSA) with +listType=map and
+        +listMapKey markers instead for merge semantics on list fields.
+
+    - path: "**/*_test.go"
+      instructions: |
+        Test case names should follow the Gherkin convention pattern:
+        "When <condition>, it should <expected behavior>".
+        For example: "When cluster has no nodes, it should clear NodesInfo status"
+        This makes test output self-documenting and easier to understand in CI logs.
+
+    - path: "**/*.go"
+      instructions: |
+        When writing data to files or performing operations where order matters,
+        prefer slices of structs over maps to ensure deterministic iteration order.
+        Go map iteration is randomized and can cause flaky tests or inconsistent
+        output.
+
+        New exported functions and non-trivial logic changes should include unit
+        tests. Tests should cover:
+        1. Happy path
+        2. Error/edge cases (malformed input, empty values, nil)
+        3. Platform-specific behavior for all relevant platforms (e.g., if adding
+           AWS support, consider whether GCP/Azure also need test coverage)
+        Check that corresponding _test.go files are included in the PR.
   tools:
     golangci-lint:
       enabled: true