OpenAgentsControl/plugins/claude-code/skills/task-breakdown/SKILL.md

---

name: task-breakdown description: Use when a feature touches 4 or more files, involves multiple components, or has subtasks that could run in parallel. context: fork

agent: task-manager

Task Breakdown

Overview

Break down complex features into atomic, verifiable subtasks with dependency tracking. Each subtask gets its own JSON file with clear acceptance criteria and deliverables.

Announce at start: "I'm using the task-breakdown skill to create an execution plan for [feature]."

The Process

Step 1: Analyze Feature

Identify these elements:

• Core objective and scope
• Technical risks and dependencies
• Natural task boundaries
• Tasks that can run in parallel

Step 2: Create Task Plan

Write .tmp/tasks/{feature}/task.json:

{
  "id": "jwt-auth",
  "name": "JWT Authentication System",
  "status": "active",
  "objective": "Implement JWT-based authentication with refresh tokens",
  "context_files": [
    ".opencode/context/core/standards/code-quality.md",
    ".opencode/context/core/standards/security-patterns.md"
  ],
  "reference_files": [
    "src/middleware/auth.middleware.ts"
  ],
  "exit_criteria": [
    "All tests passing",
    "JWT tokens signed with RS256"
  ],
  "subtask_count": 3,
  "completed_count": 0,
  "created_at": "2026-02-16T02:00:00Z"
}

Rules:

• Feature ID: kebab-case
• Objective: max 200 chars
• context_files: standards/conventions ONLY
• reference_files: project source files ONLY
• Exit criteria: binary pass/fail

Step 3: Generate Subtasks

Write .tmp/tasks/{feature}/subtask_01.json, subtask_02.json, etc:

{
  "id": "jwt-auth-01",
  "seq": "01",
  "title": "Create JWT service with token generation",
  "status": "pending",
  "depends_on": [],
  "parallel": true,
  "suggested_agent": "CoderAgent",
  "context_files": [
    ".opencode/context/core/standards/security-patterns.md"
  ],
  "reference_files": [],
  "acceptance_criteria": [
    "JWT tokens signed with RS256 algorithm",
    "Access tokens expire in 15 minutes"
  ],
  "deliverables": [
    "src/auth/jwt.service.ts",
    "src/auth/jwt.service.test.ts"
  ]
}

Rules:

• Sequential numbering: 01, 02, 03...
• Atomic tasks: completable in 1-2 hours
• Dependencies: map via depends_on array
• Parallel tasks: set parallel: true for isolated work
• Agent assignment: CoderAgent, TestEngineer, CodeReviewer, OpenFrontendSpecialist
• Acceptance criteria: binary pass/fail only
• Deliverables: specific file paths or endpoints

Step 4: Validate Structure

Verify:

• ✅ All JSON files valid
• ✅ Dependency references exist
• ✅ Context files separate from reference files
• ✅ Acceptance criteria are binary
• ✅ Deliverables are specific

Step 5: Return Summary

## Tasks Created

Location: .tmp/tasks/jwt-auth/
Files: task.json + 3 subtasks

Subtasks:
- 01: Create JWT service (parallel: true, agent: CoderAgent)
- 02: Create password hashing util (parallel: true, agent: CoderAgent)
- 03: Integrate middleware (parallel: false, agent: CoderAgent)

Next Steps:
- Execute subtasks in dependency order
- Tasks 01 and 02 can run in parallel
- Task 03 depends on completion of 01 and 02

Red Flags

If you think any of these, STOP and re-read this skill:

• "I can just implement it directly, it's not that complex"
• "The breakdown will take longer than just doing it"
• "I already know what needs to be done"
• "There's only really one task here"

Common Rationalizations

Excuse	Reality
"It's only 3-4 files, I don't need a breakdown"	3-4 files = multiple subtasks with dependencies. Skipping tracking means losing progress on failure.
"I'll track it in my head"	Subagents don't share memory. JSON files are the only reliable state.
"The tasks are obvious, no need to document them"	Obvious tasks still need acceptance criteria. "Done" without binary criteria is not done.
"Parallel execution isn't worth it for this"	Parallel tasks cut execution time in half. The JSON overhead is 2 minutes. The time saving is 20+.

Remember

• Each subtask completable in 1-2 hours (atomic)
• context_files = standards ONLY, reference_files = source code ONLY
• Acceptance criteria must be binary (pass/fail)
• Mark isolated tasks as parallel: true
• Assign appropriate agent for each subtask
• Deliverables must be specific file paths

Related

• context-discovery
• code-execution
• parallel-execution

---

Task: Break down this feature into atomic subtasks: $ARGUMENTS