Agent Skill
2/7/2026

task-management

This skill should be used when starting complex multi-step work, when breaking down work into smaller pieces, or when needing to track progress on tasks. Provides guidance on creating fine-grained tasks with proper dependencies.

D
drmaciver
14GitHub Stars
1Views
npx skills add DRMacIver/claude-reliability

SKILL.md

Nametask-management
DescriptionThis skill should be used when starting complex multi-step work, when breaking down work into smaller pieces, or when needing to track progress on tasks. Provides guidance on creating fine-grained tasks with proper dependencies.

name: Task Management description: "This skill should be used when starting complex multi-step work, when breaking down work into smaller pieces, or when needing to track progress on tasks. Provides guidance on creating fine-grained tasks with proper dependencies."

Task Management Guide

Finding Work: work next

When you need to pick a work item to work on, use the work next command. It automatically:

  • Finds the highest priority unblocked items
  • Randomly picks one to avoid always doing the same type of work
  • Shows the item details so you can start immediately

Use this instead of manually scanning the work list.

# Start a work session
claude-reliability work next
# -> Returns a suggested work item with full details

# Then mark it in progress
claude-reliability work on <work-item-id>

Creating Work Items Effectively

Philosophy

Work items are cheap and exist to help you. Don't be afraid to create many small, fine-grained items. Breaking work into small pieces helps you:

  • Track progress clearly
  • Identify blockers early
  • Maintain context across interruptions
  • Enable parallel work paths

When to Create Work Items

Create a work item whenever you:

  • Receive a user request (capture it immediately!)
  • Identify a problem while working on something else
  • Plan to do something "later"
  • Notice something that needs fixing

Never assume you'll remember it. Context may be compacted. Create the work item first, then decide whether to work on it now.

Work Item Structure

Good work items are:

  • Small: Completable in a focused session
  • Specific: Clear definition of done
  • Independent: Minimal dependencies where possible

Using Dependencies for Ordering

Dependencies ensure items happen in the right order. Use them to:

Test-Driven Development (TDD)

# 1. Create work items
claude-reliability work create -t "Design API for feature X"
claude-reliability work create -t "Write tests for feature X"
claude-reliability work create -t "Implement feature X"

# 2. Add dependencies (use IDs from creation output)
claude-reliability work add-dep <write-tests-id> --depends-on <design-api-id>
claude-reliability work add-dep <implement-id> --depends-on <write-tests-id>

This ensures: Design -> Tests -> Implementation

Multi-Phase Work

claude-reliability work create -t "Research options for auth system"
claude-reliability work create -t "Prototype chosen approach"
claude-reliability work create -t "Full implementation"
claude-reliability work create -t "Write documentation"

# Chain dependencies
claude-reliability work add-dep <prototype-id> --depends-on <research-id>
claude-reliability work add-dep <implement-id> --depends-on <prototype-id>
claude-reliability work add-dep <docs-id> --depends-on <implement-id>

Parallel with Sync Point

claude-reliability work create -t "Implement frontend component"
claude-reliability work create -t "Implement backend API"
claude-reliability work create -t "Integration testing"

# Testing depends on both
claude-reliability work add-dep <testing-id> --depends-on <frontend-id>
claude-reliability work add-dep <testing-id> --depends-on <backend-id>

Both frontend and backend can proceed in parallel, but testing waits for both.

Work Item Granularity Examples

Too Coarse

  • "Fix the authentication system" (too vague, too large)

Just Right

  • "Add password reset endpoint to auth API"
  • "Write tests for password reset flow"
  • "Update user model to track reset tokens"
  • "Add rate limiting to reset endpoint"

Tips

  1. When in doubt, create a work item - You can always delete it later
  2. Add notes as you work - Future context is valuable
  3. Use priorities realistically - P0 is truly critical, most work is P2
  4. Mark blockers with questions - question create and question link
  5. Close items promptly - Don't leave completed work as open

Bulk Work Item Creation

When creating 3+ items at once, use the bulk-tasks binary for better performance:

${CLAUDE_PLUGIN_ROOT}/bin/bulk-tasks create <<'EOF'
{
  "tasks": [
    {"id": "t1", "title": "First item", "description": "...", "priority": 1},
    {"id": "t2", "title": "Second item", "priority": 2, "depends_on": ["t1"]},
    {"id": "t3", "title": "Third item", "priority": 2, "depends_on": ["t1", "t2"]}
  ]
}
EOF

The id fields are temporary for setting up dependencies. Real IDs are returned.

Quick Reference

# Find work to do
claude-reliability work next

# Create work item
claude-reliability work create -t "Task title" -d "Description" -p 2

# Get work item details
claude-reliability work get <id>

# Add dependency (work_item_id depends on depends_on)
claude-reliability work add-dep <id> --depends-on <other-id>

# Start working
claude-reliability work on <id>

# Add context
claude-reliability work add-note <id> -c "Progress note here"

# Mark complete
claude-reliability work update <id> --status complete

# List open items
claude-reliability work list --status open --ready-only
Skills Info
Original Name:task-managementAuthor:drmaciver