OLS-3588: Add reasoningConfig field to Agent CRD

.golangci-kal.yml

@@ -21,6 +21,7 @@ linters:
             disable:
               - statusoptional # Legacy, not recommended
               - nonpointerstructs # Intended for native types, not CRD types
+              - nomaps # ReasoningConfig is a deliberate freeform map (spec rule 18)
           lintersConfig:
             conditions:
               isFirstField: Warn

api/v1alpha1/agent_types.go

@@ -17,6 +17,7 @@ limitations under the License.
 package v1alpha1
 
 import (
+	apiextensionsv1 "k8s.io/apiextensions-apiserver/pkg/apis/apiextensions/v1"
 	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
 )
 
@@ -81,6 +82,20 @@ type AgentSpec struct {
 	// +kubebuilder:validation:Minimum=1
 	// +kubebuilder:validation:Maximum=500
 	MaxTurns int32 `json:"maxTurns,omitempty"`
+
+	// reasoningConfig is a freeform map of provider- and model-specific
+	// reasoning parameters. The exact keys and values depend on the provider
+	// and model — consult the provider's SDK documentation for supported
+	// parameters (e.g., Claude: thinking/budget_tokens, Gemini: thinking_budget/
+	// thinking_level, OpenAI: reasoning.effort/reasoning.summary).
+	// The operator serializes this map as the LIGHTSPEED_REASONING_CONFIG JSON
+	// env var on the sandbox pod without validation — the sandbox and upstream
+	// SDK/API validate at invocation time. Invalid keys are ignored by the
+	// adapter; invalid values on recognized keys are rejected by the SDK/API.
+	// When omitted, the env var is not set and the sandbox uses SDK defaults.
+	// +optional
+	// +kubebuilder:validation:MinProperties=1
+	ReasoningConfig map[string]apiextensionsv1.JSON `json:"reasoningConfig,omitempty"`
 }
 
 // +kubebuilder:object:root=true
@@ -99,7 +114,7 @@ type AgentSpec struct {
 // The "default" agent must exist; "smart" and "fast" are optional (the
 // operator auto-links to "default" if absent).
 //
-// Example — a high-capability agent tier:
+// Example — a high-capability agent tier with extended thinking:
 //
 //	apiVersion: agentic.openshift.io/v1alpha1
 //	kind: Agent
@@ -113,6 +128,9 @@ type AgentSpec struct {
 //	    analysisSeconds: 300
 //	    executionSeconds: 600
 //	  maxTurns: 200
+//	  reasoningConfig:
+//	    thinking: "enabled"
+//	    effort: "high"
 //
 // Example — a fast, cost-efficient agent tier:
 //

config/crd/bases/agentic.openshift.io_agents.yaml

@@ -36,11 +36,12 @@ spec:
           name per step.\n\nAgent is cluster-scoped. The metadata.name serves as the
           tier identifier.\nThe \"default\" agent must exist; \"smart\" and \"fast\"
           are optional (the\noperator auto-links to \"default\" if absent).\n\nExample
-          — a high-capability agent tier:\n\n\tapiVersion: agentic.openshift.io/v1alpha1\n\tkind:
-          Agent\n\tmetadata:\n\t  name: smart\n\tspec:\n\t  llmProvider:\n\t    name:
-          vertex-ai\n\t  model: claude-opus-4-6\n\t  timeouts:\n\t    analysisSeconds:
-          300\n\t    executionSeconds: 600\n\t  maxTurns: 200\n\nExample — a fast,
-          cost-efficient agent tier:\n\n\tapiVersion: agentic.openshift.io/v1alpha1\n\tkind:
+          — a high-capability agent tier with extended thinking:\n\n\tapiVersion:
+          agentic.openshift.io/v1alpha1\n\tkind: Agent\n\tmetadata:\n\t  name: smart\n\tspec:\n\t
+          \ llmProvider:\n\t    name: vertex-ai\n\t  model: claude-opus-4-6\n\t  timeouts:\n\t
+          \   analysisSeconds: 300\n\t    executionSeconds: 600\n\t  maxTurns: 200\n\t
+          \ reasoningConfig:\n\t    thinking: \"enabled\"\n\t    effort: \"high\"\n\nExample
+          — a fast, cost-efficient agent tier:\n\n\tapiVersion: agentic.openshift.io/v1alpha1\n\tkind:
           Agent\n\tmetadata:\n\t  name: fast\n\tspec:\n\t  llmProvider:\n\t    name:
           vertex-ai\n\t  model: claude-haiku-4-5\n\t  timeouts:\n\t    analysisSeconds:
           120\n\t    executionSeconds: 300\n\t  maxTurns: 100"
@@ -108,6 +109,22 @@ spec:
                     only alphanumerics, dots, hyphens, underscores, slashes, colons,
                     and at-signs
                   rule: self.matches('^[a-zA-Z0-9][a-zA-Z0-9._\\-/:@]*$')
+              reasoningConfig:
+                additionalProperties:
+                  x-kubernetes-preserve-unknown-fields: true
+                description: |-
+                  reasoningConfig is a freeform map of provider- and model-specific
+                  reasoning parameters. The exact keys and values depend on the provider
+                  and model — consult the provider's SDK documentation for supported
+                  parameters (e.g., Claude: thinking/budget_tokens, Gemini: thinking_budget/
+                  thinking_level, OpenAI: reasoning.effort/reasoning.summary).
+                  The operator serializes this map as the LIGHTSPEED_REASONING_CONFIG JSON
+                  env var on the sandbox pod without validation — the sandbox and upstream
+                  SDK/API validate at invocation time. Invalid keys are ignored by the
+                  adapter; invalid values on recognized keys are rejected by the SDK/API.
+                  When omitted, the env var is not set and the sandbox uses SDK defaults.
+                minProperties: 1
+                type: object
               timeouts:
                 description: |-
                   timeouts configures per-step and per-turn timeout limits.