Rename output mode 'json' to 'structured_output'

Author: jahooma

backend/src/__tests__/loop-agent-steps.test.ts

@@ -24,9 +24,7 @@ import * as promptAgentStream from '../prompt-agent-stream'
 import * as requestContext from '../websockets/request-context'
 
 import type { AgentTemplate, StepGenerator } from '../templates/types'
-import type {
-  AgentState,
-} from '@codebuff/common/types/session-state'
+import type { AgentState } from '@codebuff/common/types/session-state'
 import type { WebSocket } from 'ws'
 
 describe('loopAgentSteps STEP behavior', () => {
@@ -151,7 +149,7 @@ describe('loopAgentSteps STEP behavior', () => {
       parentPrompt: 'Testing',
       model: 'claude-3-5-sonnet-20241022',
       inputSchema: {},
-      outputMode: 'json',
+      outputMode: 'structured_output',
       includeMessageHistory: true,
       toolNames: ['read_files', 'write_file', 'end_turn'],
       subagents: [],
@@ -180,26 +178,29 @@ describe('loopAgentSteps STEP behavior', () => {
     clearAgentGeneratorCache()
   })
 
-    llmCallCount = 0 // Reset LLM call count
+  llmCallCount = 0 // Reset LLM call count
   afterAll(() => {
     clearMockedModules()
   })
 
   it('should verify correct STEP behavior - LLM called once after STEP', async () => {
     // This test verifies that programmatic agents don't call the LLM,
     // and that STEP yielding works correctly without LLM involvement
-    
+
     let stepCount = 0
     const mockGenerator = (function* () {
       stepCount++
-      
+
       if (stepCount === 1) {
         // First call: Execute a tool, then STEP
         yield { toolName: 'read_files', args: { paths: ['file1.txt'] } }
         yield 'STEP' // Should pause here
       } else if (stepCount === 2) {
         // Second call: Should continue from here, not call LLM
-        yield { toolName: 'write_file', args: { path: 'output.txt', content: 'test' } }
+        yield {
+          toolName: 'write_file',
+          args: { path: 'output.txt', content: 'test' },
+        }
         yield { toolName: 'end_turn', args: {} }
       }
     })() as StepGenerator
@@ -213,47 +214,55 @@ describe('loopAgentSteps STEP behavior', () => {
     // Mock checkLiveUserInput to return true for multiple iterations
     let checkCallCount = 0
     const mockCheckLiveUserInput = require('@codebuff/backend/live-user-inputs')
-    spyOn(mockCheckLiveUserInput, 'checkLiveUserInput').mockImplementation(() => {
-      checkCallCount++
-      // Allow enough iterations to see the bug
-      return checkCallCount <= 3
-    })
+    spyOn(mockCheckLiveUserInput, 'checkLiveUserInput').mockImplementation(
+      () => {
+        checkCallCount++
+        // Allow enough iterations to see the bug
+        return checkCallCount <= 3
+      },
+    )
 
-    const result = await loopAgentSteps(new MockWebSocket() as unknown as WebSocket, {
-      userInputId: 'test-user-input',
-      agentType: 'test-agent',
-      agentState: mockAgentState,
-      prompt: 'Test prompt',
-      params: undefined,
-      fingerprintId: 'test-fingerprint',
-      fileContext: mockFileContext,
-      toolResults: [],
-      localAgentTemplates,
-      userId: TEST_USER_ID,
-      clientSessionId: 'test-session',
-      onResponseChunk: () => {},
-    })
+    const result = await loopAgentSteps(
+      new MockWebSocket() as unknown as WebSocket,
+      {
+        userInputId: 'test-user-input',
+        agentType: 'test-agent',
+        agentState: mockAgentState,
+        prompt: 'Test prompt',
+        params: undefined,
+        fingerprintId: 'test-fingerprint',
+        fileContext: mockFileContext,
+        toolResults: [],
+        localAgentTemplates,
+        userId: TEST_USER_ID,
+        clientSessionId: 'test-session',
+        onResponseChunk: () => {},
+      },
+    )
 
     console.log(`LLM calls made: ${llmCallCount}`)
     console.log(`Step count: ${stepCount}`)
-    
+
     // CORRECT BEHAVIOR: After STEP, LLM should be called once, then no more
     // The programmatic agent yields STEP, then LLM runs once
     expect(llmCallCount).toBe(1) // LLM called once after STEP
-    
+
     // The programmatic agent should have been called once (yielded STEP)
     expect(stepCount).toBe(1)
-    
+
     // After STEP, the LLM should run once, then the loop should continue correctly
   })
 
   it('should demonstrate correct behavior when programmatic agent completes without STEP', async () => {
     // This test shows that when a programmatic agent doesn't yield STEP,
     // it should complete without calling the LLM at all (since it ends with end_turn)
-    
+
     const mockGenerator = (function* () {
       yield { toolName: 'read_files', args: { paths: ['file1.txt'] } }
-      yield { toolName: 'write_file', args: { path: 'output.txt', content: 'test' } }
+      yield {
+        toolName: 'write_file',
+        args: { path: 'output.txt', content: 'test' },
+      }
       yield { toolName: 'end_turn', args: {} }
     })() as StepGenerator
 
@@ -263,20 +272,23 @@ describe('loopAgentSteps STEP behavior', () => {
       'test-agent': mockTemplate,
     }
 
-    const result = await loopAgentSteps(new MockWebSocket() as unknown as WebSocket, {
-      userInputId: 'test-user-input',
-      agentType: 'test-agent',
-      agentState: mockAgentState,
-      prompt: 'Test prompt',
-      params: undefined,
-      fingerprintId: 'test-fingerprint',
-      fileContext: mockFileContext,
-      toolResults: [],
-      localAgentTemplates,
-      userId: TEST_USER_ID,
-      clientSessionId: 'test-session',
-      onResponseChunk: () => {},
-    })
+    const result = await loopAgentSteps(
+      new MockWebSocket() as unknown as WebSocket,
+      {
+        userInputId: 'test-user-input',
+        agentType: 'test-agent',
+        agentState: mockAgentState,
+        prompt: 'Test prompt',
+        params: undefined,
+        fingerprintId: 'test-fingerprint',
+        fileContext: mockFileContext,
+        toolResults: [],
+        localAgentTemplates,
+        userId: TEST_USER_ID,
+        clientSessionId: 'test-session',
+        onResponseChunk: () => {},
+      },
+    )
 
     // Should NOT call LLM since the programmatic agent ended with end_turn
     expect(llmCallCount).toBe(0)

backend/src/__tests__/run-agent-step-tools.test.ts

@@ -307,7 +307,7 @@ describe('runAgentStep - set_output tool', () => {
       parentPrompt: 'Testing handleSteps functionality',
       model: 'claude-3-5-sonnet-20241022',
       inputSchema: {},
-      outputMode: 'json' as const,
+      outputMode: 'structured_output' as const,
       includeMessageHistory: true,
       toolNames: ['read_files', 'end_turn'],
       subagents: [],
@@ -424,7 +424,7 @@ describe('runAgentStep - set_output tool', () => {
       parentPrompt: 'Deletes assistant messages',
       model: 'claude-3-5-sonnet-20241022',
       inputSchema: {},
-      outputMode: 'json' as const,
+      outputMode: 'structured_output' as const,
       includeMessageHistory: true,
       toolNames: ['set_messages', 'end_turn'],
       subagents: [],
@@ -466,7 +466,7 @@ describe('runAgentStep - set_output tool', () => {
       parentPrompt: 'Parent agent that spawns inline agents',
       model: 'claude-3-5-sonnet-20241022',
       inputSchema: {},
-      outputMode: 'json' as const,
+      outputMode: 'structured_output' as const,
       includeMessageHistory: true,
       toolNames: ['spawn_agent_inline', 'end_turn'],
       subagents: ['message-deleter-agent'],

backend/src/__tests__/run-programmatic-step.test.ts

@@ -87,7 +87,7 @@ describe('runProgrammaticStep', () => {
       parentPrompt: 'Testing',
       model: 'claude-3-5-sonnet-20241022',
       inputSchema: {},
-      outputMode: 'json',
+      outputMode: 'structured_output',
       includeMessageHistory: true,
       toolNames: ['read_files', 'write_file', 'end_turn'],
       subagents: [],
@@ -820,7 +820,7 @@ describe('runProgrammaticStep', () => {
       // Create template with outputSchema
       const schemaTemplate = {
         ...mockTemplate,
-        outputMode: 'json' as const,
+        outputMode: 'structured_output' as const,
         outputSchema: {
           type: 'object',
           properties: {
@@ -868,7 +868,7 @@ describe('runProgrammaticStep', () => {
       // Create template with strict outputSchema
       const schemaTemplate = {
         ...mockTemplate,
-        outputMode: 'json' as const,
+        outputMode: 'structured_output' as const,
         outputSchema: {
           type: 'object',
           properties: {
@@ -949,10 +949,10 @@ describe('runProgrammaticStep', () => {
       })
     })
 
-    it('should work with outputMode json but no outputSchema defined', async () => {
+    it('should work with outputMode structured_output but no outputSchema defined', async () => {
       const schemaWithoutSchemaTemplate = {
         ...mockTemplate,
-        outputMode: 'json' as const,
+        outputMode: 'structured_output' as const,
         outputSchema: undefined, // No schema defined
         toolNames: ['set_output', 'end_turn'],
       }
@@ -986,6 +986,7 @@ describe('runProgrammaticStep', () => {
       })
     })
   })
+
   describe('logging and context', () => {
     it('should log agent execution start', async () => {
       const mockGenerator = (function* () {

backend/src/__tests__/sandbox-generator.test.ts

@@ -35,7 +35,7 @@ describe('QuickJS Sandbox Generator', () => {
       displayName: 'Test VM Agent',
       parentPrompt: 'Test VM isolation',
       model: 'anthropic/claude-4-sonnet-20250522',
-      outputMode: 'json',
+      outputMode: 'structured_output',
       includeMessageHistory: false,
       toolNames: ['set_output'],
       subagents: [],

backend/src/templates/agents/agent-builder.ts

@@ -86,7 +86,7 @@ export const agentBuilder = (model: Model): Omit<AgentTemplate, 'id'> => {
         .passthrough()
         .optional(),
     },
-    outputMode: 'json',
+    outputMode: 'structured_output',
     includeMessageHistory: false,
     toolNames: [
       'write_file',

backend/src/templates/agents/base-agent-builder.ts

@@ -120,7 +120,7 @@ export const baseAgentBuilder = (
         .passthrough()
         .optional(),
     },
-    outputMode: 'json',
+    outputMode: 'structured_output',
     includeMessageHistory: false,
     toolNames: [
       'create_plan',

backend/src/templates/agents/file-explorer.ts

@@ -17,7 +17,7 @@ export const fileExplorer = {
   parentPrompt:
     'Spawns multiple file picker agents in parallel to comprehensively explore the codebase from different perspectives',
   model: 'anthropic/claude-4-sonnet-20250522',
-  outputMode: 'json',
+  outputMode: 'structured_output',
   includeMessageHistory: false,
   toolNames: ['spawn_agents', 'set_output'],
   subagents: ['file-picker'],

backend/src/templates/dynamic-agents.knowledge.md

@@ -198,7 +198,7 @@ yield { toolName: 'write_file', args: { path: '...', content: '...' } }
 // Search code
 yield { toolName: 'code_search', args: { pattern: '...' } }
 
-// Set final output (required for outputMode: 'json')
+// Set final output (required for outputMode: 'structured_output')
 yield { toolName: 'set_output', args: { result: {...} } }
 ```
 

backend/src/tools/handlers/tool/spawn-agents.ts

@@ -233,7 +233,7 @@ export const handleSpawnAgents = ((params: {
           }
           let report = ''
 
-          if (agentTemplate.outputMode === 'json') {
+          if (agentTemplate.outputMode === 'structured_output') {
             report = JSON.stringify(result.value.agentState.output, null, 2)
           } else if (agentTemplate.outputMode === 'last_message') {
             const { agentState } = result.value

common/src/__tests__/agent-validation.test.ts

@@ -69,7 +69,7 @@ describe('Agent Validation', () => {
       displayName: 'Test Agent',
       parentPrompt: 'Testing',
       model: 'claude-3-5-sonnet-20241022',
-      outputMode: 'json' as const,
+      outputMode: 'structured_output' as const,
       toolNames: ['set_output'],
       subagents: [],
       includeMessageHistory: true,
@@ -671,7 +671,7 @@ describe('Agent Validation', () => {
         displayName: 'Test Agent',
         parentPrompt: 'Testing handleSteps',
         model: 'claude-3-5-sonnet-20241022',
-        outputMode: 'json' as const,
+        outputMode: 'structured_output' as const,
         toolNames: ['set_output'],
         systemPrompt: 'You are a test agent',
         instructionsPrompt: 'Process: {prompt}',
@@ -746,7 +746,7 @@ describe('Agent Validation', () => {
         displayName: 'Test Agent',
         parentPrompt: 'Testing',
         model: 'claude-3-5-sonnet-20241022',
-        outputMode: 'json' as const,
+        outputMode: 'structured_output' as const,
         toolNames: ['end_turn'], // Missing set_output
         subagents: [],
         systemPrompt: 'Test',
@@ -775,7 +775,7 @@ describe('Agent Validation', () => {
         displayName: 'Test Agent',
         parentPrompt: 'Testing',
         model: 'claude-3-5-sonnet-20241022',
-        outputMode: 'last_message' as const, // Not json
+        outputMode: 'last_message' as const, // Not structured_output
         toolNames: ['end_turn', 'set_output'], // Has set_output
         subagents: [],
         systemPrompt: 'Test',
@@ -788,7 +788,7 @@ describe('Agent Validation', () => {
       if (!result.success) {
         const errorMessage = result.error.issues[0]?.message || ''
         expect(errorMessage).toContain(
-          "'set_output' tool requires outputMode to be 'json'",
+          "'set_output' tool requires outputMode to be 'structured_output'",
         )
       }
     })