| name | task |
| description | Execute complex tasks with planning, parallel subagents, atomic commits, and documentation updates. Use when the user asks to implement a feature, refactor code, or complete multi-step work. |
Task Execution Skill
Execute complex tasks using a structured workflow: plan → review → execute → commit → document.
Workflow
Phase 1: Planning
- Analyze the request - Understand what the user is asking for
- Explore the codebase - Use the Explore agent to understand:
- Relevant files and their structure
- Existing patterns and conventions
- Dependencies and integrations
- Break into atomic steps - Create a plan where each step:
- Can be completed independently
- Results in working code (no broken intermediate states)
- Is small enough for a single commit
- Identify parallelization - Mark which steps can run in parallel vs sequentially
Phase 2: Plan Review
- Present the plan to the user with:
- Overview of the approach
- Numbered list of steps
- Which steps will run in parallel
- Any assumptions or decisions made
- Wait for approval - Use AskUserQuestion if needed to clarify:
- Ambiguous requirements
- Implementation choices
- Scope boundaries
- Revise if needed based on user feedback
Phase 3: Execution
- Use TodoWrite to track all steps
- Spawn subagents using the Task tool:
- Run independent steps in parallel (multiple Task calls in one message)
- Run dependent steps sequentially
- Each subagent should complete one atomic unit of work
- Subagent instructions should include:
- Clear scope of what to implement
- Files to modify
- Patterns to follow (reference existing code)
- Instruction to NOT commit (parent will handle commits)
Phase 4: Atomic Commits
After each logical unit of work is complete:
- Stage only related changes - Don't mix unrelated changes
- Write descriptive commit messages:
- First line: imperative mood, max 50 chars (e.g., "Add blur placeholders for images")
- Body: explain WHY, not just WHAT
- Reference the step from the plan
- Commit format:
<type>: <short description> <optional body explaining why> 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com> - Types: feat, fix, refactor, docs, style, test, chore
Phase 5: Documentation Check
After all work is complete:
- Review what changed - List the major changes made
- Check if documentation needs updates:
- README.md - New features, changed setup, updated commands
- Code comments - Complex logic that needs explanation
- Type definitions - New interfaces or types
- Ask the user if documentation updates are desired
- If yes, create a separate commit for docs
Guidelines
- Never leave code in a broken state - Each commit should pass build/lint
- Prefer editing over creating - Modify existing files when possible
- Follow existing patterns - Match the codebase style
- Keep changes minimal - Only change what's necessary for the task
- Test as you go - Run build/tests after significant changes
Example Subagent Prompt
Implement the blur placeholder generation script.
Context:
- This is step 2 of 5 in adding image blur placeholders
- The project uses Next.js with TypeScript
- Sharp is already installed for image processing
Task:
- Create scripts/generate-blur-placeholders.ts
- Read all images from public/images/themes/
- Generate 10x10 blurred base64 data URLs
- Output to src/config/blur-placeholders.ts
Follow the patterns in scripts/generate-images.ts.
Do NOT commit - just implement and verify it works.