feat: update CodeMapper model to Claude Sonnet 4.6 and adjust routing configurations for improved performance

Author: Smith

CodeMapper-subagent.agent.md

@@ -2,7 +2,7 @@
 description: Explore the codebase to find relevant files, usages, dependencies, and context for a given research goal or problem statement.
 argument-hint: Find files, usages, dependencies, and context related to: <research goal or problem statement>
 tools: ['search', 'usages', 'problems', 'changes', 'testFailure']
-model: GPT-5.4 mini (copilot)
+model: Claude Sonnet 4.6 (copilot)
 model_role: fast-readonly
 ---
 You are CodeMapper-subagent, a read-only discovery agent.

docs/agent-engineering/MODEL-ROUTING.md

@@ -43,7 +43,7 @@ While internal subagent dispatch resolves models dynamically, top-level user ent
 
 - `Planner` is pinned to `GPT-5.5` so plan quality, decomposition, and risk framing use the strongest available planning model. The planner fallback ordering is governed by `governance/model-routing.json` via `roles.capable-planner` (primary: `GPT-5.5 (copilot)`, fallback 1: `Claude Opus 4.7 (copilot)`, fallback 2: `GPT-5.4 mini (copilot)`). This workflow preserves the scalar `model:` frontmatter in `Planner.agent.md` and does not imply list-valued frontmatter is required.
 - `CodeReviewer`, `PlanAuditor`, and `AssumptionVerifier` rely on premium adversarial reviewers (`Claude Opus 4.7`) for direct invocation and for `LARGE` or high-risk orchestrated dispatch; ordinary `TRIVIAL`, `SMALL`, and `MEDIUM` orchestrated dispatch is tier-aware as described in [Fallback semantics](#fallback-semantics).
-- `ExecutabilityVerifier`, `Orchestrator`, and the implementation agents typically resolve to cheaper defaults (`Claude Sonnet 4.6`, `GPT-5.4`, `GPT-5.4 mini`, `Gemini 3.1 Pro`) to contain premium usage.
+- `CodeMapper` is pinned to `Claude Sonnet 4.6` for capable, cost-effective codebase discovery (1× premium multiplier). `ExecutabilityVerifier`, `Orchestrator`, and the implementation agents typically resolve to contained defaults (`Claude Sonnet 4.6`, `GPT-5.4`, `GPT-5.4 mini`, `Gemini 3.1 Pro`) to balance capability and premium usage.
 
 This yields a pragmatic split:
 

evals/scenarios/orchestrator-model-resolution.json

@@ -81,6 +81,20 @@
           "live_runtime_assertion": false
         }
       },
+      {
+        "case_id": "fast-readonly-codemapper-sonnet",
+        "research_agent": "CodeMapper-subagent",
+        "complexity_tier": "MEDIUM",
+        "role": "fast-readonly",
+        "resolution_mode": "inherit_from default",
+        "dispatch_context": "Planner or Orchestrator research delegation to CodeMapper resolves through governance/model-routing.json agent_role_index.CodeMapper-subagent -> roles.fast-readonly and must pass the resolved primary model explicitly to agent/runSubagent.",
+        "reference_expectation": {
+          "resolved_primary_model": "Claude Sonnet 4.6 (copilot)",
+          "resolution_source": "governance/model-routing.json roles.fast-readonly.primary",
+          "agentName_field_used": "agentName",
+          "live_runtime_assertion": false
+        }
+      },
       {
         "case_id": "executability-verifier-review-readonly-sonnet",
         "executor_agent": "ExecutabilityVerifier-subagent",
@@ -102,6 +116,6 @@
     "offline_harness_observes_live_runSubagent_model_parameters": false,
     "live_runtime_behavior_out_of_scope": true,
     "behavioral_runtime_assertions_present": false,
-    "reference_cases_documented": 6
+    "reference_cases_documented": 7
   }
 }

evals/tests/drift-detection.test.mjs

@@ -504,12 +504,12 @@ console.log('\n=== Check #6 — by_tier matrix shape ===');
           SMALL:   { inherit_from: 'default' },
           MEDIUM: {
             primary: 'GPT-4o (copilot)',
-            fallbacks: ['Claude Sonnet 4.5 (copilot)'],
+            fallbacks: ['Claude Sonnet 4.6 (copilot)'],
             cost_tier: 'medium',
             latency_tier: 'medium',
           },
           LARGE: {
-            primary: 'Claude Sonnet 4.5 (copilot)',
+            primary: 'Claude Sonnet 4.6 (copilot)',
             fallbacks: ['GPT-4o (copilot)'],
             cost_tier: 'high',
             latency_tier: 'slow',

evals/tests/orchestration-handoff-contract.test.mjs

@@ -71,6 +71,7 @@ const _capablePlanner         = resolveRoleModel('capable-planner', 'MEDIUM');
 const capablePlannerPrimary   = _capablePlanner.primary;        // GPT-5.5 (copilot)
 const capablePlannerFallback0 = _capablePlanner.fallbacks[0];   // Claude Opus 4.7 (copilot)
 const capablePlannerFallback1 = _capablePlanner.fallbacks[1];   // GPT-5.4 mini (copilot)
+const fastReadonlyPrimary = resolveRoleModel(modelRouting.agent_role_index['CodeMapper-subagent'], 'MEDIUM').primary;
 
 let passed = 0;
 let failed = 0;
@@ -817,6 +818,20 @@ check(
   evCase?.reference_expectation?.live_runtime_assertion === false
 );
 
+const codeMapperCase = allCases.find(c => c.case_id === 'fast-readonly-codemapper-sonnet');
+check(
+  'Model resolution scenario: fast-readonly-codemapper-sonnet case exists',
+  codeMapperCase !== undefined
+);
+check(
+  `Model resolution scenario: CodeMapper resolves through fast-readonly to ${fastReadonlyPrimary}`,  // Claude Sonnet 4.6 (copilot)
+  codeMapperCase?.reference_expectation?.resolved_primary_model === fastReadonlyPrimary
+);
+check(
+  'Model resolution scenario: CodeMapper live_runtime_assertion is false',
+  codeMapperCase?.reference_expectation?.live_runtime_assertion === false
+);
+
 const largeHighRiskCase = allCases.find(c => c.case_id === 'capable-reviewer-large-high-risk-override');
 check(
   'Model resolution scenario: capable-reviewer-large-high-risk-override case exists',

governance/model-routing.json

@@ -161,12 +161,12 @@
       }
     },
     "fast-readonly": {
-      "primary": "GPT-5.4 mini (copilot)",
-      "fallbacks": ["GPT-5.4 (copilot)", "Claude Sonnet 4.6 (copilot)"],
-      "cost_tier": "low",
-      "latency_tier": "fast",
+      "primary": "Claude Sonnet 4.6 (copilot)",
+      "fallbacks": ["GPT-5.4 mini (copilot)", "GPT-5.4 (copilot)"],
+      "cost_tier": "medium",
+      "latency_tier": "medium",
       "consumers": ["CodeMapper-subagent.agent.md"],
-      "reasoning_effort_hint": "low",
+      "reasoning_effort_hint": "medium",
       "by_tier": {
         "TRIVIAL": { "inherit_from": "default" },
         "SMALL":   { "inherit_from": "default" },

plans/project-context.md

@@ -6,7 +6,7 @@ The following agents are available for Orchestrator phase dispatch. The `executo
 
 | Agent | Role | Primary Use Case | Model Recommendation |
 | --- | --- | --- | --- |
-| CodeMapper-subagent | Read-only discovery | Codebase exploration, file mapping | Fast model (mini) |
+| CodeMapper-subagent | Read-only discovery | Codebase exploration, file mapping | Capable read-only model (Sonnet 4.6) |
 | Researcher-subagent | Research & evidence | Deep investigation, evidence extraction | Fast model (mini) |
 | CoreImplementer-subagent | Backend implementation | Code creation, modification, testing | Capable model (Sonnet) |
 | UIImplementer-subagent | UI implementation | Components, styling, accessibility | Capable model (Sonnet) |