| name | planning |
| description | This skill should be used when in planning/plan mode, or when the user wants to plan features, research implementations, or understand the codebase before generating specs. Use this for exploratory research and clarification before implementation. |
Planning Skill
Overview
This skill provides comprehensive planning workflows for features and system changes. It focuses on the research and clarification phase - gathering context, exploring architecture, and asking good questions before spec generation.
When to Use This Skill
Use this skill when:
- In planning/plan mode (auto-trigger if possible)
- User says: "plan this", "research this feature", "help me understand"
- User asks: "what do I need to know before implementing X?"
- Before generating specs with
/cmd:generate-spec
- Need to explore codebase patterns and architecture
Core Workflow
1. Research Phase
- Explore codebase for similar existing features
- Identify which domain/area this belongs in
- Locate related types, schemas, services, routes
- Check test patterns in relevant areas
- Review database models (if applicable)
- Use
references/research-checklist.md for systematic exploration
2. Context Gathering
- Internal docs: Check
.agent/docs/ for relevant patterns
- CLAUDE.md: Review project conventions and rules
- context7 MCP (if available): Fetch external library docs
- Check for
mcp__context7__* tools
- Use for React, Fastify, Prisma, etc. documentation
- Fall back to WebFetch/WebSearch if not available
- See
references/context-sources.md for detailed guidance
3. Clarification (If Needed)
Ask clarifying questions ONE AT A TIME if implementation approach is unclear:
4. Complexity Assessment
- Estimate file count and scope
- Identify cross-cutting concerns
- Assess if new patterns needed or extending existing
- Determine context requirements for implementation
- Flag if complexity suggests breaking into smaller features
- See
references/complexity-assessment.md for detailed rubric
5. Handoff to Spec Generation
Prepare clear summary for /cmd:generate-spec:
- What was found during research
- Existing patterns to follow
- Recommended architecture approach
- Initial complexity estimate
- Unresolved questions (if any)
- Suggest: "Ready to generate spec with
/cmd:generate-spec?"
Key Principles
Do
- ✅ Research deeply - explore codebase thoroughly
- ✅ Ask questions conversationally (text-based, ONE AT A TIME)
- ✅ Document findings clearly
- ✅ Leverage context7 MCP when available for external docs
- ✅ Check
.agent/docs/ for internal patterns
- ✅ Provide architecture recommendations with rationale
- ✅ Prepare handoff context for spec generation
Don't
- ❌ Use AskUserQuestion tool (breaks exploratory flow)
- ❌ Create specs or files (that's
/cmd:generate-spec's job)
- ❌ Ask multiple questions at once
- ❌ Skip research phase
- ❌ Make assumptions without checking codebase first
Complementary to Slash Commands
This skill complements existing spec generation:
User request → Planning skill (research) → /cmd:generate-spec (spec creation)
- Planning skill: Research, clarification, architecture exploration
- /cmd:generate-spec: Spec writing, file creation, index updates
- /generate-prd: PRD creation for larger features
- /generate-feature: Alternative spec format
Level of Involvement
Moderate - Balance depth with efficiency:
- Research deeply: Explore codebase, find patterns, identify constraints
- Ask strategically: ONE question at a time with context-based suggestions
- Document findings: Summarize architecture, existing patterns, complexity
- Prepare handoff: Create clear context for spec generation
- Don't overstep: Leave spec writing and file creation to slash commands
References
See references/ directory for detailed guidance:
research-checklist.md - Systematic codebase exploration
clarification-questions.md - Question templates and examples
complexity-assessment.md - Complexity rubric and indicators
context-sources.md - How to leverage context7, internal docs, codebase
Example Usage
User: "Plan adding user notifications feature"
Planning skill workflow:
1. Research: Find existing notification patterns (grep for "notification", "alert", etc.)
2. Check: `.agent/docs/` for notification architecture (if exists)
3. Context7: Fetch WebSocket or push notification docs (if needed)
4. Ask: "What type of notifications? In-app, email, push, or all three?"
Suggestions:
1. In-app only (recommended - simpler, use WebSocket pattern)
2. Email integration (adds complexity, need mail service)
3. Other - user specifies
5. Wait for answer, continue clarification as needed
6. Assess: Initial complexity ~5-6/10 (multiple files, WebSocket integration)
7. Output: "Found WebSocket handler pattern in X, session events in Y.
Recommend: Extend existing WebSocket system, add Notification model,
create UI toast component. Estimated 5-6/10 complexity."
8. Suggest: "Ready to generate spec with /cmd:generate-spec?"
Integration Notes
- context7 MCP: Optional but recommended for external docs
- Permission mode: Works best in plan mode (read-only exploration)
- Output: Text summary, not files (spec generation creates files)