Support COPILOT_MODEL=auto in AWF validation and api-proxy resolution

containers/api-proxy/guards/ai-credits-guard.js

@@ -142,13 +142,15 @@ function resolveModelPricing(model, state = aiCreditsState) {
  * Only rejects when maxAiCredits is active and no default pricing is configured.
  *
  * @param {string} model
+ * @param {string} [provider]
  * @returns {{ rejected: boolean, model: string, error: object } | null}
  */
-function checkUnknownModelRejection(model) {
+function checkUnknownModelRejection(model, provider = undefined) {
   const config = getAiCreditsConfig();
   if (!config.max) return null; // guard not active, don't reject
   if (!model) return null; // no model in request body, can't check
   if (config.defaultPricing) return null; // has fallback, don't reject
+  if (provider === PROVIDER_COPILOT && model.toLowerCase() === 'auto') return null;
 
   const pricing = resolveModelPricing(model);
   if (pricing) return null; // model resolved, don't reject

containers/api-proxy/guards/ai-credits-guard.test.js

@@ -8,6 +8,7 @@ const {
   canonicalizeModel,
   resetAiCreditsGuardForTests,
 } = require('./ai-credits-guard');
+const { PROVIDER_COPILOT, PROVIDER_OPENAI } = require('../provider-names');
 const { collectLogOutput } = require('../test-helpers/log-test-helpers');
 
 describe('ai-credits-guard', () => {
@@ -468,5 +469,13 @@ describe('ai-credits-guard', () => {
       const sonnet5 = checkUnknownModelRejection('claude-sonnet-5');
       expect(sonnet5).toBeNull();
     });
+
+    it('allows the Copilot auto selector without default pricing', () => {
+      process.env.AWF_MAX_AI_CREDITS = '10';
+      resetAiCreditsGuardForTests();
+
+      expect(checkUnknownModelRejection('auto', PROVIDER_COPILOT)).toBeNull();
+      expect(checkUnknownModelRejection('auto', PROVIDER_OPENAI)).not.toBeNull();
+    });
   });
 });

containers/api-proxy/guards/common-guard-checks.js

@@ -38,9 +38,10 @@
  * @param {Function} deps.buildModelPolicyError
  * @param {string|null} model - Model name extracted from the request, or null
  *   to skip model-specific guards.
+ * @param {string|null} [provider=null] - Provider handling the request.
  * @returns {Array<object>} Array of guard descriptor objects.
  */
-function buildCommonGuardChecks(deps, model) {
+function buildCommonGuardChecks(deps, model, provider = null) {
   const {
     getEffectiveTokenBlockState,
     buildEffectiveTokenLimitError,
@@ -164,7 +165,7 @@ function buildCommonGuardChecks(deps, model) {
         }),
       },
       {
-        block: checkUnknownModelRejection(model),
+        block: checkUnknownModelRejection(model, provider),
         isBlocked: block => !!block,
         statusCode: 400,
         eventName: 'unknown_model_ai_credits',

containers/api-proxy/model-resolver.js

@@ -270,6 +270,18 @@ function resolveModel(requestedModel, aliases, availableModels, currentProvider,
   const key = requestedModel.toLowerCase();
   const fallbackConfig = normalizeFallbackConfig(modelFallbackConfig);
 
+  if (currentProvider === 'copilot' && key === 'auto') {
+    log.push('[model-resolver] special pass-through: "auto"');
+    return {
+      resolvedModel: requestedModel,
+      candidates: [requestedModel],
+      log,
+      fallback: fallbackConfig.enabled
+        ? { activated: false, selection_method: 'middle_power_median', reason: 'direct_match' }
+        : undefined,
+    };
+  }
+
   // Loop detection
   if (chain.includes(key)) {
     log.push(`[model-resolver] loop detected: "${requestedModel}" already in chain [${chain.join(' → ')}]`);

containers/api-proxy/model-resolver.test.js

@@ -127,6 +127,13 @@ describe('resolveModel', () => {
     expect(result.resolvedModel).toBe('gpt-4o');
   });
 
+  it('should treat the Copilot auto model as a pass-through', () => {
+    const result = resolveModel('auto', aliases, availableModels, 'copilot');
+    expect(result).not.toBeNull();
+    expect(result.resolvedModel).toBe('auto');
+    expect(result.fallback.activated).toBe(false);
+  });
+
   it('should be case-insensitive for alias lookup', () => {
     const result = resolveModel('SONNET', aliases, availableModels, 'copilot');
     expect(result).not.toBeNull();
@@ -358,6 +365,12 @@ describe('rewriteModelInBody', () => {
     expect(result).toBeNull(); // No rewrite needed
   });
 
+  it('should not rewrite the Copilot auto model', () => {
+    const body = Buffer.from(JSON.stringify({ model: 'auto', messages: [] }));
+    const result = rewriteModelInBody(body, 'copilot', aliases, availableModels);
+    expect(result).toBeNull();
+  });
+
   it('should return null for non-JSON body', () => {
     const body = Buffer.from('not json');
     const result = rewriteModelInBody(body, 'copilot', aliases, availableModels);

containers/api-proxy/proxy-guards.js

@@ -172,7 +172,7 @@ function enforceGuards({ body, provider, req, res, requestId, startTime, span, i
     checkUnknownModelRejection,
     getModelPolicyBlockState,
     buildModelPolicyError,
-  }, model);
+  }, model, provider);
 
   for (const guard of guardChecks) {
     if (!guard.isBlocked(guard.block)) continue;

containers/api-proxy/server.startup-model-validation.test.js

@@ -67,6 +67,16 @@ describe('validateRequestedModel', () => {
     }));
   });
 
+  it('treats the Copilot auto model as available without requiring it in provider model lists', () => {
+    process.env.AWF_REQUESTED_MODEL = 'auto';
+    cachedModels.copilot = ['gpt-4o', 'gpt-4o-mini'];
+    validateRequestedModel();
+    expect(logRequest).toHaveBeenCalledWith('info', 'model_validation', expect.objectContaining({
+      requested_model: 'auto',
+      resolved_via: 'direct',
+    }));
+  });
+
   it('searches across multiple providers', () => {
     process.env.AWF_REQUESTED_MODEL = 'claude-sonnet-4-5';
     cachedModels.copilot = ['gpt-4o'];

containers/api-proxy/server.token-guards.test.js

@@ -284,6 +284,19 @@ describe('proxyRequest max-ai-credits guard', () => {
     return makeReqFactory('/v1/chat/completions', headers);
   }
 
+  function makeModelReq(body, headers = {}) {
+    const req = makeReq(headers);
+    const bodyBuf = Buffer.from(body);
+    const originalEmit = req.emit.bind(req);
+    req.emit = function(event, ...args) {
+      if (event === 'end') {
+        originalEmit('data', bodyBuf);
+      }
+      return originalEmit(event, ...args);
+    };
+    return req;
+  }
+
   beforeEach(() => {
     process.env.AWF_MAX_AI_CREDITS = '0.1';
     delete process.env.AWF_MAX_EFFECTIVE_TOKENS;
@@ -327,6 +340,20 @@ describe('proxyRequest max-ai-credits guard', () => {
     expect(payload.error.max_ai_credits).toBe(0.1);
     expect(payload.error.total_ai_credits).toBeGreaterThanOrEqual(0.1);
   });
+
+  it('allows Copilot auto requests through the ai credits request guard', async () => {
+    const upstreamRequest = makeProxyReq();
+    const httpsRequestSpy = jest.spyOn(https, 'request').mockImplementation(() => upstreamRequest);
+
+    const req = makeModelReq(JSON.stringify({ model: 'auto', messages: [] }));
+    const res = makeRes();
+    proxyRequest(req, res, 'api.githubcopilot.com', { Authorization: '******' }, 'copilot');
+    req.emit('end');
+    await flushPromises();
+
+    expect(httpsRequestSpy).toHaveBeenCalledTimes(1);
+    expect(res.writeHead).not.toHaveBeenCalledWith(400, expect.anything());
+  });
 });
 
 describe('proxyRequest permission-denied guard', () => {

containers/api-proxy/websocket-guards.js

@@ -17,7 +17,7 @@ const HTTP_STATUS_LINES = {
 function enforceWebSocketGuards({ socket, logRequest, requestId, provider }, guardDeps) {
   // WebSocket upgrade requests have no JSON body, so model-specific guards
   // receive null and are skipped (their getters return null for null models).
-  const guardChecks = buildCommonGuardChecks(guardDeps, null);
+  const guardChecks = buildCommonGuardChecks(guardDeps, null, provider);
 
   for (const guard of guardChecks) {
     if (!guard.isBlocked(guard.block)) continue;

src/commands/validators/config-assembly-model-detection.test.ts

@@ -225,6 +225,27 @@ describe('config-assembly', () => {
       );
     });
 
+    it('should allow COPILOT_MODEL=auto', () => {
+      mockBuildConfigOnce({
+        copilotGithubToken: 'github_pat_testtoken',
+      });
+
+      const agentOptions = createMinimalAgentOptions();
+      agentOptions.additionalEnv = { COPILOT_MODEL: 'auto' };
+
+      expect(() => {
+        assembleAndValidateConfig(
+          {},
+          'echo test',
+          createMinimalLogAndLimits(),
+          createMinimalNetworkOptions(),
+          agentOptions,
+        );
+      }).not.toThrow();
+
+      expect(logger.error).not.toHaveBeenCalled();
+    });
+
     it('should allow COPILOT_MODEL that matches a runtime alias key and resolves to a valid concrete model', () => {
       mockBuildConfigOnce({
         copilotGithubToken: 'github_pat_testtoken',

src/copilot-model.test.ts

@@ -16,6 +16,11 @@ describe('validateCopilotModel', () => {
     expect(result).toEqual({ valid: true, resolvedModel: 'gpt-5.3-codex' });
   });
 
+  it('accepts the Copilot auto model', () => {
+    const result = validateCopilotModel(' auto ');
+    expect(result).toEqual({ valid: true, resolvedModel: 'auto' });
+  });
+
   it.each([
     'gpt-4.5',
     'gpt-5.1',

src/copilot-model.ts

@@ -42,6 +42,7 @@ const RETIRED_COPILOT_MODEL_ALIASES: Record<string, string> = {
  * explicitly listed as a non-CLI model in the test's exclusion set).
  */
 const SUPPORTED_COPILOT_MODELS = new Set([
+  'auto',
   'gpt-4',
   'gpt-4.1',
   'gpt-4.5',