bug:

[AllyourBaseBelongToUs]

🐛 Bug Report: JSON Schema Validation Error During Task Expansion

Summary

Task expansion fails with JSON schema validation error when using AI to generate subtasks. The schema is missing 'dependencies' in the required fields for subtask items.

Description

When attempting to expand tasks using tm expand <task_id> --research, the operation fails with a 400 Bad Request error indicating that the response schema validation is incomplete.

Error Message:

Invalid schema for response_format 'codex_output_schema': In context=('properties', 'subtasks', 'items'), 'required' is required to be supplied and to be an array including every key in properties. Missing 'dependencies'.

Steps to Reproduce

1. Initialize a Task Master project with tasks
2. Run complexity analysis: tm analyze-complexity --research
3. Attempt to expand a high-complexity task (score 8+): tm expand 9 --research --force
4. Observe the schema validation error during AI subtask generation

Expected Behavior

• Task expansion should succeed
• AI-generated subtasks should include proper dependency arrays
• Schema validation should pass for valid subtask structures

Actual Behavior

• Expansion fails with JSON schema validation error
• Error indicates 'dependencies' field is missing from required schema properties
• Codex CLI exits with code 1, max retries reached

Environment

• Task Master Version: 0.40.0
• Platform: Windows 11
• AI Provider: codex-cli (Codex)
• Model: gpt-5.1-codex-max
• Research Mode: Enabled

Additional Context

• Tasks 1-8 were successfully expanded (manually created/expanded)
• Tasks 9-10 fail during AI expansion (complexity scores: 8)
• Existing tasks.json has proper dependencies arrays in subtasks
• Issue appears to be in the JSON schema used for validating AI-generated responses

Full Terminal Output Context

Task 8 Expansion Success:

╭────────────────── 💡 Telemetry ──────────────────╮
│                                                  │
│   AI Usage Summary:                              │
│     Command: expand-task                         │
│     Provider: claude-code                        │
│     Model: opus                                  │
│     Tokens: 49463 (Input: 48174, Output: 1289)   │
│     Est. Cost: $0.000000                         │
│                                                  │
╰──────────────────────────────────────────────────╯
[INFO] Successfully expanded task 8.

Task 9 Expansion Failure Sequence:

⠋ Expanding task 9...
[INFO] Reading tasks from [PROJECT_PATH]\.taskmaster\tasks\tasks.json
[INFO] Expanding task 9: Implement MCP Server with Tool and Resource Providers with research
[INFO] Looking for complexity report at: [PROJECT_PATH]\.taskmaster\reports\task-complexity-report.json
[INFO] Found complexity analysis for task 9: Score 8
[INFO] Using subtask count from complexity report: 8
[INFO] Using expansion prompt from complexity report for task 9.
⠧ Generating 8 subtasks...
[ERROR] [codex-cli] Error event: unexpected status 400 Bad Request: {
  "error": {
    "message": "Invalid schema for response_format 'codex_output_schema': In context=('properties', 'subtasks', 'items'), 'required' is required to be supplied and to be an array including every key in properties. Missing 'dependencies'.",
    "type": "invalid_request_error",
    "param": "text.format.schema",
    "code": "invalid_json_schema"
  }
}
[ERROR] [codex-cli] Turn failed: unexpected status 400 Bad Request: [same error]
[ERROR] Codex CLI object generation failed: Codex CLI exited with code 1
[WARN] Attempt 1 failed for role research (generateObject / codex-cli): Codex CLI API error during object generation
[ERROR] Something went wrong on the provider side. Max retries reached for role research (generateObject / codex-cli).
[ERROR] Service call failed for role research (Provider: codex-cli, Model: gpt-5.1-codex-max): Codex CLI API error during object generation

Impact

• High-complexity tasks cannot be automatically expanded
• Blocks TDD autopilot workflow for complex modules
• Manual subtask creation works but defeats automation benefits

Error Response Details

Failed Model: gpt-5.1-codex-max

Full Error Response:

{
  "error": {
    "message": "Invalid schema for response_format 'codex_output_schema': In context=('properties', 'subtasks', 'items'), 'required' is required to be supplied and to be an array including every key in properties. Missing 'dependencies'.",
    "type": "invalid_request_error",
    "param": "text.format.schema",
    "code": "invalid_json_schema"
  }
}

Process Exit: Code 1 after 8727ms
Max Retries: Reached for research role

Possible Solution

Update the codex_output_schema to include 'dependencies' in the required array for subtask items:

{
  "subtasks": {
    "items": {
      "properties": {
        "id": {"type": "string"},
        "title": {"type": "string"},
        "description": {"type": "string"},
        "dependencies": {"type": "array"},
        "details": {"type": "string"},
        "status": {"type": "string"},
        "testStrategy": {"type": "string"}
      },
      "required": ["id", "title", "description", "dependencies", "details", "status", "testStrategy"]
    }
  }
}

Severity: High - Blocks automated task expansion for complex tasks
Priority: High - Core functionality broken for research-enhanced expansion

• Task Master version:

0.40.0

• Operating system:

Windows

• IDE:

Cursor

---

Fallback to claude code worked and task-master expand_all proceeded through all tasks

[coderabbitai]

📝 CodeRabbit Plan Mode

Generate an implementation plan and prompts that you can use with your favorite coding agent.

✅ Successfully generated implementation plan.

💡 Use @coderabbitai <feedback> to replan

Examples

• Example 1
• Example 2

---

🔗 Similar Issues

Related Issues

• Invalid JSON Schema for Codex CLI Provider: Missing additionalProperties: false in subtasks schema #1522
• bug: Task expansion fails with JSON parsing error when using claude-code provider #1174
• bug: issue with the task-master AI response parsing in claude-code #1223
• bug: Error parsing subtasks #182

🔗 Related PRs

#1000 - fix(ai-validation): comprehensive fixes for AI response validation issues [merged]
#1079 - feat: improve scope up and down command & parse-prd improvements [merged]
#1094 - fix(expand-task): include parent task context in complexity report variant [merged]
#1191 - fix: resolve cross-level dependency bug in add-dependency command [merged]
#1287 - fix: expand_all now uses complexity analysis recommendations [merged]

👤 Suggested Assignees

• Crunchyman-ralph
• AllyourBaseBelongToUs
• Kl-11
• PabloAraya6
• carnett-orchestrai

---

🧪 Issue enrichment is currently in open beta.

You can configure auto-planning by selecting labels in the issue_enrichment configuration.

To disable automatic issue enrichment, add the following to your .coderabbit.yaml:

issue_enrichment:
  auto_enrich:
    enabled: false

💬 Have feedback or questions? Drop into our discord or schedule a call!

[Crunchyman-ralph]

@AllyourBaseBelongToUs isn't this a duplicate of your existing issue, it seems like they are related, its just related to the json schema that we have to tweak to play nice with codex basically

[coderabbitai]

Coding Plan

Approach

Ticket Summary

The task expansion command fails with a JSON schema validation error when using the codex-cli provider (OpenAI's gpt-5.1-codex-max model). OpenAI's strict mode requires all fields to be listed under "Required" in the function calling schema, but the current Zod schema uses .default() modifiers on the dependencies field, which makes it appear as optional in the generated JSON Schema and causes OpenAI to reject it.

Observations

Codebase Summary

The Task Master project uses Zod schemas defined in src/schemas/ to validate AI-generated responses. The core subtask schema is defined in src/schemas/base-schemas.js and imported by src/schemas/expand-task.js for task expansion operations. These Zod schemas are passed directly to the Vercel AI SDK's generateObject function in src/ai-providers/base-provider.js, which converts them to JSON Schema format. The codex-cli provider operates with needsExplicitJsonSchema = false, using OpenAI's native structured output mode which enforces strict schema validation. The codebase already handles all subtask fields defensively with null coalescing operators and array defaults at usage points, so making fields explicitly required won't break existing functionality.

Assumptions

Assumption 1: Handling of default values in OpenAI strict mode

Options Considered:

• Remove .default() from dependencies only and keep other fields unchanged
• Remove all .default() modifiers from all optional fields in SubtaskSchema
• Replace .nullable().default(null) patterns with empty string defaults

Chosen Option: Remove .default() from the dependencies field specifically, and change testStrategy from .nullable().default(null) to use an empty string default without nullable. The status field should retain its .default('pending') as it represents a valid initial state.

Rationale: This approach minimizes changes while ensuring OpenAI strict mode compatibility. Based on the exploration findings, the codebase already handles undefined values gracefully at all usage points, so making dependencies explicitly required will not break functionality.

Assumption 2: Scope of schema fixes beyond expand-task

Options Considered:

• Fix only the expand-task schema to resolve the immediate ticket
• Fix all schemas that use SubtaskSchema or BaseTaskSchema to prevent future issues
• Audit and fix the entire schema directory for OpenAI compatibility

Chosen Option: Fix both base-schemas.js (which affects expand-task, update-tasks, and other commands) comprehensively to ensure consistent schema structure across all AI-powered commands.

Rationale: This prevents the same error from occurring in other commands that use OpenAI's strict mode. The update-tasks command also uses generateObjectService with OpenAI and would encounter the same validation issues if not addressed.

💡 User Tips

Regenerate the plan with different choices with @coderabbitai <feedback>.

Implementation Steps

Phase 1: Update SubtaskSchema for OpenAI Strict Mode Compatibility

Fix the core subtask schema definition to ensure all properties are included in the required fields array when converted to JSON Schema format.

Task 1: Remove .default() from dependencies field

Modify src/schemas/base-schemas.js to make the dependencies field explicitly required without a default value.

• Locate the SubtaskSchema definition (around line 27-35)
• Change dependencies: z.array(z.number().int()).default([]) to dependencies: z.array(z.number().int())
• This ensures the dependencies field appears in the JSON Schema's required array for OpenAI validation

Task 2: Fix testStrategy field nullable pattern

Update the testStrategy field to remove the problematic .nullable().default(null) pattern.

• In the same SubtaskSchema definition in src/schemas/base-schemas.js
• Change testStrategy: z.string().nullable().default(null) to testStrategy: z.string().default('')
• This maintains backward compatibility while ensuring the field is always present with a valid string value (empty string when not specified)

Task 3: Evaluate status field default behavior

Determine if the status field's default should be retained or removed based on OpenAI strict mode requirements.

• Test whether status: z.enum(['pending', 'done', 'completed']).default('pending') causes schema validation issues
• If the .default() modifier on enum fields is acceptable to OpenAI, retain it as it represents a valid initial state
• If it causes validation errors, remove .default('pending') and ensure all AI prompts explicitly instruct the model to generate status values

🤖 Prompt for AI agents

Update the SubtaskSchema in `src/schemas/base-schemas.js` to ensure OpenAI
strict mode compatibility by making all fields explicitly required.

Remove `.default([])` from the dependencies field - change to
`z.array(z.number().int())`
Fix testStrategy field - change from `.nullable().default(null)` to
`.default('')`
Evaluate the status field's `.default('pending')` - test if enum defaults are
compatible with OpenAI strict mode, and remove if necessary

Phase 2: Fix BaseTaskSchema Nullable Patterns

Address similar schema issues in the base task schema that would affect other commands using task objects.

Update BaseTaskSchema nullable fields

Fix the .nullable().default(null) patterns in BaseTaskSchema in src/schemas/base-schemas.js.

• Identify all fields in BaseTaskSchema using .nullable().default(null) pattern (likely on string fields like description or details)
• Replace with appropriate non-nullable defaults:
  • For text fields that can be empty: change to .default('')
  • For fields that should always have content: remove .nullable() and .default() to make truly required
• Ensure consistency with how these fields are used throughout the codebase (exploration shows defensive null checks are already in place)

🤖 Prompt for AI agents

Fix BaseTaskSchema in `src/schemas/base-schemas.js` to remove problematic
nullable patterns for OpenAI compatibility.

Identify all fields using `.nullable().default(null)` pattern
Replace with appropriate defaults:
- Text fields that can be empty: change to `.default('')`
- Fields requiring content: remove `.nullable()` and `.default()` to make
required
Verify consistency with existing codebase usage patterns

Phase 3: Fix UpdateTasksSchema for Consistent Behavior

Ensure the update-tasks command schema follows the same strict mode compatibility patterns.

Update UpdatedTaskSchema subtasks field

Modify src/schemas/update-tasks.js to fix the nullable array pattern.

• Locate the UpdatedTaskSchema definition (around line 5)
• Change subtasks: z.array(SubtaskSchema).nullable().default(null) to subtasks: z.array(SubtaskSchema).optional()
• This allows the subtasks field to be omitted entirely during updates (since not all updates modify subtasks) while avoiding the problematic .nullable().default(null) chain
• Verify that the update-tasks command logic handles optional subtasks correctly

🤖 Prompt for AI agents

Update the UpdatedTaskSchema in `src/schemas/update-tasks.js` for OpenAI strict
mode compatibility.

Change the subtasks field from `.nullable().default(null)` to `.optional()`
This allows the field to be omitted during partial updates while avoiding the
problematic nullable pattern
Verify the update-tasks command logic handles optional subtasks correctly

Phase 4: Validation and Testing

Verify that the schema changes resolve the OpenAI validation error and maintain compatibility with other providers.

Task 1: Test schema changes with codex-cli provider

Validate the fix against the specific failure scenario from the ticket.

• Run tm expand <task_id> --research --force using codex-cli provider on a high-complexity task
• Verify that the JSON schema validation error no longer occurs
• Confirm that generated subtasks include all required fields: id, title, description, dependencies (even if empty array), details, status, and testStrategy (even if empty string)

Task 2: Verify claude-code provider compatibility

Ensure the schema changes don't break the fallback provider that was working.

• Run the same expand command using claude-code provider
• Confirm that task expansion still succeeds with the updated schemas
• Validate that the generated subtask structure matches expectations

Task 3: Test update-tasks command with both providers

Validate that the update-tasks schema changes work correctly.

• Execute tm update-tasks command with both codex-cli and claude-code providers
• Verify that partial updates (without subtasks modifications) still work correctly
• Confirm that full updates (including subtasks) function as expected

🤖 Prompt for AI agents

Validate all schema changes resolve the OpenAI validation error while
maintaining compatibility across providers.

Test with codex-cli provider: run `tm expand <task_id> --research --force` and
verify no validation errors occur
Test with claude-code provider: run same command and confirm task expansion
still works
Test update-tasks command with both providers: verify partial and full updates
function correctly

---

🚀 Next Steps

🤖 All AI agent prompts combined

Task: 1

Update the SubtaskSchema in `src/schemas/base-schemas.js` to ensure OpenAI
strict mode compatibility by making all fields explicitly required.

Remove `.default([])` from the dependencies field - change to
`z.array(z.number().int())`
Fix testStrategy field - change from `.nullable().default(null)` to
`.default('')`
Evaluate the status field's `.default('pending')` - test if enum defaults are
compatible with OpenAI strict mode, and remove if necessary
===============================================================================

Task: 2

Fix BaseTaskSchema in `src/schemas/base-schemas.js` to remove problematic
nullable patterns for OpenAI compatibility.

Identify all fields using `.nullable().default(null)` pattern
Replace with appropriate defaults:
- Text fields that can be empty: change to `.default('')`
- Fields requiring content: remove `.nullable()` and `.default()` to make
required
Verify consistency with existing codebase usage patterns
===============================================================================

Task: 3

Update the UpdatedTaskSchema in `src/schemas/update-tasks.js` for OpenAI strict
mode compatibility.

Change the subtasks field from `.nullable().default(null)` to `.optional()`
This allows the field to be omitted during partial updates while avoiding the
problematic nullable pattern
Verify the update-tasks command logic handles optional subtasks correctly
===============================================================================

Task: 4

Validate all schema changes resolve the OpenAI validation error while
maintaining compatibility across providers.

Test with codex-cli provider: run `tm expand <task_id> --research --force` and
verify no validation errors occur
Test with claude-code provider: run same command and confirm task expansion
still works
Test update-tasks command with both providers: verify partial and full updates
function correctly

💡 Iterate on the plan with: @coderabbitai <feedback>

Example Feedback
- @coderabbitai You can skip phase 3. Add a simple unit test case for phase 2.
- @coderabbitai For assumption 1 go ahead with option 3 and replan.

---

💬 Have feedback or questions? Drop into our discord or schedule a call!

[AllyourBaseBelongToUs]

@AllyourBaseBelongToUs isn't this a duplicate of your existing issue, it seems like they are related, its just related to the json schema that we have to tweak to play nice with codex basically

they are related cause a similar file needs to be edited yeah, but had checked before that its not the same thing

cause the error is not the same either

one was expanding a task the other one was updating subtasks

though the odd thing is, it sometimes works in the MCP but most of the times fails and goes straight to fallback

[AllyourBaseBelongToUs]

Here for example, same attempt, update subtask and it works with codex 5.1 max

then the next 20 attempts doesnt with the API Error

confusing...

"""

"data": {
"message": "Successfully updated subtask with ID 5.5",
"subtaskId": "5.5",
"parentId": "5",
"subtask": {
"id": 5,
"title": "Implement Section D parser for FILE blocks with numbered step extraction",
"description": "Parse FILE sections into FileImplementation objects with extracted implementation steps",
"dependencies": [
1
],
"details": "Create parseFileImplementations() method using state machine: SEEKING_FILE_HEADER, PARSING_FILENAME, EXTRACTING_STEPS. Identify FILE: headers with regex. Extract file paths and extensions. Parse numbered steps (1., 2., etc.) with sub-step detection (1.1, 1.2). Handle missing numbers, duplicate numbers, and orphaned content. Implement step merging for broken lines.\n<info added on 2025-12-19T18:52:09.655Z>\nImplemented parseFileImplementations in src/parser/section-extractor.ts to extract FILE blocks with verb-aware numbered steps, returning FileImplementation objects with fileName, purpose, and steps array; all 7 related tests passing.\n</info added on 2025-12-19T18:52:09.655Z>",
"status": "done",
"testStrategy": "Test with malformed FILE headers, broken numbering, multi-step implementations, and content without numbers.",
"parentId": "undefined",
"updatedAt": "2025-12-19T18:50:11.528Z"
},
"tasksPath": "c:PathToProject\.taskmaster\tasks\tasks.json",
"useResearch": false,
"telemetryData": {
"timestamp": "2025-12-19T18:52:09.653Z",
"userId": "XXX",
"commandName": "update-subtask",
"modelUsed": "gpt-5.1-codex-max",
"providerName": "codex-cli",
"inputTokens": 44537,
"outputTokens": 1936,
"totalTokens": 46473,
"totalCost": 0,
"currency": "USD",
"isUnknownCost": false
},
"tagInfo": {
"currentTag": "master",
"availableTags": [
"master"
]
}
},
"version": {
"version": "0.40.0",
"name": "task-master-ai"
},
"tag": "master"
}

"""

[Crunchyman-ralph]

Yup, but what I mean is that it a more "general/global schema issue"

[AllyourBaseBelongToUs]

yeah, should i keep finding and posting this in issues or rather send via mail or similar?

[Crunchyman-ralph]

Keep shooting them here, though I feel like this PR #1523 handles most cases, let me know or create more issues if you run into anything else. Appreciate you @AllyourBaseBelongToUs