fix(codex): read text-block custom tool output

src/cli.test.ts

@@ -5,6 +5,7 @@ import { describe, it, expect } from 'vitest'
 import { failClosedResponse, run, type RunResult, type Vendor } from './cli.js'
 import type { Config } from './config.js'
 import type { Agent, SessionEvent } from './types.js'
+import { enforceTdd } from './rules/enforce-tdd.js'
 import { enforceFilenameCasing } from './rules/enforce-filename-casing.js'
 import { forbidContentPattern } from './rules/forbid-content-pattern.js'
 import type { FileContent, Rule } from './rules/contract.js'
@@ -150,6 +151,43 @@ describe('cli', () => {
     expect(captured?.some((e) => e.kind === 'command')).toBe(true)
   })
 
+  it('passes Codex Vitest text-block output to enforceTdd as red evidence', async () => {
+    let capturedPrompt = ''
+    const redAwareAgent: Agent = {
+      reason: (prompt) => {
+        capturedPrompt = prompt
+        const observedRed =
+          prompt.includes('AssertionError: expected undefined to be 5') &&
+          /Test Files\s+1 failed/.test(prompt)
+        return Promise.resolve({
+          kind: observedRed ? 'pass' : 'violation',
+          reason: observedRed ? '' : 'Vitest red evidence was not observed',
+        })
+      },
+    }
+    const payload = JSON.stringify({
+      transcript_path: 'test/fixtures/transcripts/codex-tdd-test-failed.jsonl',
+      cwd: '/workspaces/probity',
+      tool_name: 'apply_patch',
+      tool_input: {
+        command:
+          '*** Begin Patch\n' +
+          '*** Add File: src/calculator.ts\n' +
+          '+export const add = (a: number, b: number) => a + b\n' +
+          '*** End Patch\n',
+      },
+    })
+
+    const { response } = await setup({
+      vendor: 'codex',
+      payload,
+      config: { rules: [enforceTdd({ fastPath: false })], ai: redAwareAgent },
+    })
+
+    expect(capturedPrompt).toContain('npx vitest run src/calculator.test.ts')
+    expect(response).toBe('')
+  })
+
   it('threads a ctx.readFile capability into the RuleContext for write actions', async () => {
     let captured: FileContent | undefined
     const captureRule: Rule = async (_action, ctx) => {

src/vendors/codex/transcript.test.ts

@@ -41,6 +41,18 @@ describe('codex transcript', () => {
     }
   })
 
+  it('reads custom tool output from legacy strings and current text blocks', async () => {
+    const events = await readTranscript(
+      'test/fixtures/transcripts/codex-custom-tool-output-formats.jsonl',
+    )
+
+    const actions = events.filter((event) => event.kind === 'action')
+    expect(actions.map((event) => event.output)).toEqual([
+      'legacy output',
+      'current output\nsecond block',
+    ])
+  })
+
   it('parses JSON-encoded function_call arguments into objects at the boundary', async () => {
     const events = await readTranscript(
       'test/fixtures/transcripts/codex-basic.jsonl',

src/vendors/codex/transcript.ts

@@ -42,12 +42,23 @@ const CustomToolCallSchema = z.object({
   }),
 })
 
+const CustomToolOutputSchema = z
+  .union([
+    z.string(),
+    z.array(z.object({ type: z.literal('text'), text: z.string() })),
+  ])
+  .transform((output) =>
+    typeof output === 'string'
+      ? output
+      : output.map((block) => block.text).join('\n'),
+  )
+
 const CustomToolCallOutputSchema = z.object({
   type: z.literal('response_item'),
   payload: z.object({
     type: z.literal('custom_tool_call_output'),
     call_id: z.string(),
-    output: z.string(),
+    output: CustomToolOutputSchema,
   }),
 })
 

test/fixtures/transcripts/codex-custom-tool-output-formats.jsonl

@@ -0,0 +1,4 @@
+{"type":"response_item","payload":{"type":"custom_tool_call","name":"legacy_tool","input":"legacy input","call_id":"call_legacy"}}
+{"type":"response_item","payload":{"type":"custom_tool_call_output","call_id":"call_legacy","output":"legacy output"}}
+{"type":"response_item","payload":{"type":"custom_tool_call","name":"current_tool","input":"current input","call_id":"call_current"}}
+{"type":"response_item","payload":{"type":"custom_tool_call_output","call_id":"call_current","output":[{"type":"text","text":"current output"},{"type":"text","text":"second block"}]}}

test/fixtures/transcripts/codex-tdd-test-failed.jsonl

@@ -1,5 +1,5 @@
 {"timestamp":"2026-01-01T00:00:00.000Z","type":"response_item","payload":{"type":"message","role":"user","content":[{"text":"Write a simple add function in src/calculator.ts using TDD."}]}}
 {"timestamp":"2026-01-01T00:00:01.000Z","type":"response_item","payload":{"type":"custom_tool_call","status":"completed","call_id":"call_test_write","name":"apply_patch","input":"*** Begin Patch\n*** Add File: /workspaces/probity/src/calculator.test.ts\n+import { describe, expect, it } from 'vitest'\n+\n+import { add } from './calculator.js'\n+\n+describe('calculator', () => {\n+  it('adds two numbers', () => {\n+    expect(add(2, 3)).toBe(5)\n+  })\n+})\n*** End Patch\n"}}
 {"timestamp":"2026-01-01T00:00:01.500Z","type":"response_item","payload":{"type":"custom_tool_call_output","call_id":"call_test_write","output":"Success."}}
-{"timestamp":"2026-01-01T00:00:02.000Z","type":"response_item","payload":{"type":"function_call","name":"shell","arguments":"{\"command\":\"npx vitest run src/calculator.test.ts\"}","call_id":"call_test_run"}}
-{"timestamp":"2026-01-01T00:00:03.000Z","type":"response_item","payload":{"type":"function_call_output","call_id":"call_test_run","output":" FAIL  src/calculator.test.ts > calculator > adds two numbers\nAssertionError: expected undefined to be 5 // Object.is equality\n\n- Expected: 5\n+ Received: undefined\n\n at src/calculator.test.ts:6:25\n\nTest Files  1 failed (1)\n     Tests  1 failed (1)"}}
+{"timestamp":"2026-01-01T00:00:02.000Z","type":"response_item","payload":{"type":"custom_tool_call","name":"exec_command","input":"{\"cmd\":\"npx vitest run src/calculator.test.ts\"}","call_id":"call_test_run"}}
+{"timestamp":"2026-01-01T00:00:03.000Z","type":"response_item","payload":{"type":"custom_tool_call_output","call_id":"call_test_run","output":[{"type":"text","text":" FAIL  src/calculator.test.ts > calculator > adds two numbers\nAssertionError: expected undefined to be 5 // Object.is equality\n\n- Expected: 5\n+ Received: undefined\n\n at src/calculator.test.ts:6:25"},{"type":"text","text":"Test Files  1 failed (1)\n     Tests  1 failed (1)"}]}}

test/integration/enforce-tdd-codex.test.ts

@@ -49,7 +49,7 @@ const it = baseTest
 describe.concurrent(
   'enforce-tdd + codex',
   () => {
-    it('allows clean TDD with minimal implementation', async ({
+    it('uses Vitest text-block output as red evidence for minimal implementation', async ({
       runScenario,
     }) => {
       const result = await runScenario({