| name | create-plan |
| description | Create detailed implementation plans with thorough research and iteration. Use when starting significant features spanning multiple files, planning refactors affecting architecture, working on multi-phase projects with milestones, establishing success criteria before coding, breaking down complex work, or documenting approach for non-trivial technical decisions. |
Create Implementation Plan (Generic)
Workflow
1. Understand the Requirement
- Read all mentioned files completely (no partial reads)
- Present your informed understanding with specific file:line references
- Ask focused questions only about what research couldn't clarify
- Don't ask questions answerable through code exploration
2. Decompose and Plan Research
- Break task into composable research areas
- Identify components, patterns, concepts to investigate
- Use TodoWrite to track subtasks
- Think deeply about underlying patterns and constraints
3. Spawn Parallel Sub-Agents
Codebase Research:
codebase-locator - Find WHERE files/components live
codebase-analyzer - Understand HOW existing code works
codebase-pattern-finder - Find similar implementations to learn from
Agent Tips:
- Start with locator agents to find what exists
- Use analyzer agents on promising findings
- Run multiple agents in parallel
- Tell agents what you're looking for, not how to search
4. Wait and Synthesize
Wait for ALL sub-agents, then:
- Compile results, prioritize live codebase as source of truth
- Include file paths and line numbers
- Highlight patterns, connections, architectural decisions
- Answer questions with concrete evidence
5. Propose Plan Structure
Present outline for approval:
## Overview
[1-2 sentence summary]
## Implementation Phases:
1. [Phase Name] - [What this accomplishes]
2. [Phase Name] - [What this accomplishes]
3. [Phase Name] - [What this accomplishes]
Get explicit user buy-in before investing in detailed planning.
6. Write the Plan
Save to: thoughts/shared/plans/YYYY-MM-DD-brief-description.md (if thoughts/ exists)
Or present inline if no thoughts directory.
Structure:
- Overview - Brief goal description
- Current State Analysis - What exists (file:line refs), what's missing, constraints
- Desired End State - What "done" looks like
- Key Discoveries - Findings with file:line refs
- What We're NOT Doing - Out-of-scope items
- Implementation Approach - High-level strategy
Per phase:
- Overview, changes required (file + specifics), success criteria, confirmation gate
End with:
- Testing Strategy - Unit, integration, manual
- References - Tickets, research, code refs
7. Finalize
- Run
humanlayer thoughts sync (if applicable)
- Present draft location to user
- Iterate on feedback
Success Criteria Format
Automated: Tests, type check, lint, build
Manual: Feature works, performance acceptable, edge cases handled
Key Principles
- File:line references and measurable criteria
- Phases with confirmation gates
- Research before proposing
- Consider backwards compatibility
- No open questions - research or ask first
Common Implementation Patterns
Database Changes:
Schema → Store/Repository methods → Business logic → API endpoints → Client code
New Features:
Research patterns → Data model → Backend implementation → API → UI/Frontend
Refactoring:
Document current state → Incremental changes → Maintain backwards compatibility → Migration strategy
API Changes:
Document current behavior → Deprecation plan → New implementation → Migration guide → Old code removal