todo-tracking

All tasks live in .claude/branches/<slugified-branch>/tasks/ as individual files with ordering support.

One location. One format. Always on a branch. Branch names slugified (/ → -).

Resources

Resource	Purpose
`references/branch-tasks.md`	Complete task system documentation
`templates/task-file-template.md`	Individual task file format
`templates/plan-template.md`	Plan document format

<quick_start>

Quick Start

File Location

text

.claude/branches/<branch>/tasks/
├── 001-pending-p1-setup-create-structure.md
├── 002-pending-p1-setup-install-deps.md
├── 010-pending-p1-us1-create-user-model.md
├── 011-complete-p1-us1-implement-auth.md
├── 050-pending-p1-found-fix-type-error.md  # From /crew:check
└── 099-pending-p3-polish-update-docs.md

File Naming

text

{order}-{status}-{priority}-{story}-{slug}.md

order: 001, 002... (3 digits)
status: pending, in_progress, complete
priority: p1, p2, p3
story: setup, found, us1, us2, polish

Task File Structure

markdown

---
status: pending
priority: p1
story: us1
parallel: true
file_path: src/models/user.ts
depends_on: []
---

# T010: Create User model

## Description

[What and why]

## Acceptance Criteria

- [ ] **Given** X, **When** Y, **Then** Z

## File Path

`src/models/user.ts`

## Work Log

### [DATE] - Created

**By:** /crew:design

</quick_start>

<parallel_strategy>

Parallel Agent Strategy

One task = One agent. Launch many in parallel.

javascript

// Launch all parallel tasks in SINGLE message
Task({
  subagent_type: "general-purpose",
  prompt: "T001...",
  description: "T001",
  run_in_background: true,
});
Task({
  subagent_type: "general-purpose",
  prompt: "T002...",
  description: "T002",
  run_in_background: true,
});
Task({
  subagent_type: "general-purpose",
  prompt: "T003...",
  description: "T003",
  run_in_background: true,
});

// Collect results
TaskOutput({ task_id: "t001-id", block: true });
TaskOutput({ task_id: "t002-id", block: true });
TaskOutput({ task_id: "t003-id", block: true });

// Launch next batch...

Rules:

Each task handles ONE small piece of work
Only launch tasks with parallel: true
Batch and collect before next batch
Small scope = faster completion = less context exhaustion

</parallel_strategy>

<command_integration>

Command Integration

Command	Action
`/crew:design`	Creates plan + individual task files
`/crew:build`	Reads tasks, launches parallel agents by batch
`/crew:check`	Adds finding files as new tasks
`/crew:fix`	Works through pending tasks

</command_integration>

Validation Checklist

Tasks in .claude/branches/<slugified-branch>/tasks/ (feat/foo → feat-foo)
Filenames follow ordering convention
Status matches filename AND frontmatter
Each task has acceptance criteria
Parallel tasks marked appropriately
Work log updated on completion

Search AI Tools

Install this agent skill to your Project

SKILL.md