11 KiB
description
| description |
|---|
| Analyze requirements and create detailed implementation plan |
Titanium Plan Command
You are creating a structured implementation plan from requirements. Follow this systematic process to break down work into actionable tasks with agent assignments.
MCP Tools Used: This command uses the tt MCP server (Titanium Toolkit) which provides:
mcp__plugin_titanium-toolkit_tt__plan_parser- Generates structured implementation plans from requirements
The tt server wraps Python utilities that use Claude AI to analyze requirements and create detailed project plans with task-to-agent assignments.
Agent Assignment: The plan_parser automatically assigns tasks to appropriate specialized agents based on task type (API work → @api-developer, UI work → @frontend-developer, etc.). These assignments are used by /titanium:work to delegate implementation.
Process Overview
This command will:
- Gather and validate requirements
- Use Claude (via
plan_parserMCP tool) to generate structured plan - Validate plan with vibe-check
- Create human-readable documentation
- Store plan in Pieces for future reference
Step 1: Gather Requirements
If user provides a file path:
# User might say: /titanium:plan ~/bmad/output/user-auth-prd.md
- Use Read tool to read the file
- Extract requirements text
If user provides inline description:
# User might say: /titanium:plan
# Then describe: "I need to add JWT authentication with login, register, password reset"
- Write description to
.titanium/requirements.mdusing Write tool - Ask clarifying questions if needed:
- What tech stack? (Node.js, Python, Ruby, etc.)
- What database? (PostgreSQL, MongoDB, etc.)
- Any specific libraries or frameworks?
- Security requirements?
- Performance requirements?
Step 2: Generate Structured Plan
Use the plan_parser MCP tool to generate the plan:
mcp__plugin_titanium-toolkit_tt__plan_parser(
requirements_file: ".titanium/requirements.md",
project_path: "$(pwd)"
)
This will:
- Call Claude with the requirements
- Generate structured JSON plan with:
- Epics (major features)
- Stories (user-facing functionality)
- Tasks (implementation steps)
- Agent assignments
- Time estimates
- Task dependencies
- Save to
.titanium/plan.json - Return the JSON plan directly to Claude
Important: The plan_parser tool needs ANTHROPIC_API_KEY environment variable. If it fails with an API key error, inform the user they need to add it to ~/.env
Step 3: Review the Generated Plan
Read and analyze .titanium/plan.json:
# Read the plan
Read .titanium/plan.json
Check that the plan:
- Has reasonable epics (1-5 major features)
- Each epic has logical stories (1-5 per epic)
- Each story has actionable tasks (2-10 per story)
- Agent assignments are appropriate
- Time estimates seem realistic
- Dependencies make sense
Common issues to watch for:
- Tasks assigned to wrong agents (e.g., frontend work to @api-developer)
- Missing testing tasks
- Missing documentation tasks
- Unrealistic time estimates
- Circular dependencies
If the plan needs adjustments:
- Edit
.titanium/requirements.mdto add clarifications - Re-run the
plan_parsertool - Review again
Step 4: Validate Plan with vibe-check
Use vibe-check to validate the plan quality:
mcp__vibe-check__vibe_check(
goal: "User's stated goal from requirements",
plan: "Summary of the generated plan - list epics, key stories, agents involved, total time",
uncertainties: [
"List any concerns about complexity",
"Note any ambiguous requirements",
"Mention any technical risks"
]
)
Example:
mcp__vibe-check__vibe_check(
goal: "Implement JWT authentication system with login, register, and password reset",
plan: "2 epics: Backend API (JWT middleware, 3 endpoints, database) and Frontend UI (login/register forms, password reset flow). Agents: @product-manager, @api-developer, @frontend-developer, @test-runner, @security-scanner. Total: 4 hours",
uncertainties: [
"Should we use refresh tokens or just access tokens?",
"Password hashing algorithm not specified - suggest argon2",
"Rate limiting strategy needs clarification"
]
)
Handle vibe-check response:
- If vibe-check raises concerns:
- Review the concerns carefully
- Update requirements or plan approach
- Re-run the
plan_parsertool with adjustments - Validate again with vibe-check
- If vibe-check approves:
- Continue to next step
Step 5: Create Human-Readable Plan
Write a markdown version of the plan to .titanium/plan.md:
# Implementation Plan: [Project Goal]
**Created**: [Date]
**Estimated Time**: [Total time from plan.json]
## Goal
[User's goal statement]
## Tech Stack
[List technologies mentioned in requirements]
## Epics
### Epic 1: [Epic Name]
**Description**: [Epic description]
**Estimated Time**: [Sum of all story times]
#### Story 1.1: [Story Name]
**Description**: [Story description]
**Tasks**:
1. [Task 1 name] - [@agent-name] - [time estimate]
2. [Task 2 name] - [@agent-name] - [time estimate]
#### Story 1.2: [Story Name]
**Description**: [Story description]
**Tasks**:
1. [Task 1 name] - [@agent-name] - [time estimate]
2. [Task 2 name] - [@agent-name] - [time estimate]
### Epic 2: [Epic Name]
[... repeat structure ...]
## Agents Involved
- **@product-manager**: Requirements validation
- **@api-developer**: Backend implementation
- **@frontend-developer**: UI development
- **@test-runner**: Testing
- **@doc-writer**: Documentation
## Dependencies
[List any major dependencies between epics/stories]
## Next Steps
Ready to execute? Run: `/titanium:work`
Step 6: Store Plan in Pieces
Store the plan in Pieces LTM for future reference:
mcp__Pieces__create_pieces_memory(
summary_description: "Implementation plan for [project name/goal]",
summary: "Plan created with [X] epics, [Y] stories, [Z] tasks. Agents: [list agents]. Estimated time: [total time]. Key features: [brief list of main epics]. vibe-check validation: [summary of validation results]",
files: [
".titanium/plan.json",
".titanium/plan.md",
".titanium/requirements.md"
],
project: "$(pwd)"
)
Example:
mcp__Pieces__create_pieces_memory(
summary_description: "Implementation plan for JWT authentication system",
summary: "Plan created with 2 epics, 5 stories, 12 tasks. Agents: @product-manager, @api-developer, @frontend-developer, @test-runner, @security-scanner. Estimated time: 4 hours. Key features: JWT middleware with refresh tokens, login/register/reset endpoints, frontend auth forms, comprehensive testing. vibe-check validation: Plan structure is sound, recommended argon2 for password hashing, suggested rate limiting on auth endpoints.",
files: [
".titanium/plan.json",
".titanium/plan.md",
".titanium/requirements.md"
],
project: "/Users/username/projects/my-app"
)
Step 7: Present Plan to User
Format the output in a clear, organized way:
📋 Implementation Plan Created
🎯 Goal: [User's goal]
📦 Structure:
- [X] epics
- [Y] stories
- [Z] implementation tasks
⏱️ Estimated Time: [total time]
🤖 Agents Involved:
- @agent-name (role description)
- @agent-name (role description)
- [... list all agents ...]
📁 Plan saved to:
- .titanium/plan.json (structured data)
- .titanium/plan.md (readable format)
✅ vibe-check validated: [Brief summary of validation results]
📝 Key Epics:
1. [Epic 1 name] - [time estimate]
2. [Epic 2 name] - [time estimate]
[... list all epics ...]
---
Ready to execute this plan?
Run: /titanium:work
This will orchestrate the implementation using the plan,
with voice announcements and quality gates throughout.
Important Guidelines
Always:
- ✅ Use the
plan_parserMCP tool (don't try to generate plans manually) - ✅ Validate with vibe-check before finalizing
- ✅ Store the plan in Pieces
- ✅ Create both JSON (for machines) and Markdown (for humans)
- ✅ Get user approval before they proceed to /titanium:work
- ✅ Be specific about agent roles in the summary
Never:
- ❌ Skip vibe-check validation
- ❌ Generate plans without using the
plan_parsertool - ❌ Proceed to implementation without user approval
- ❌ Ignore vibe-check concerns
- ❌ Create plans without clear task assignments
Error Handling
If ANTHROPIC_API_KEY is missing:
Error: The plan_parser tool needs an Anthropic API key to generate plans.
Please add your API key to ~/.env:
echo 'ANTHROPIC_API_KEY=sk-ant-your-key-here' >> ~/.env
chmod 600 ~/.env
Then restart Claude Code and try again.
If vibe-check is not available:
Warning: vibe-check MCP is not available. Proceeding without quality validation.
Consider setting up vibe-check for AI-powered quality gates:
1. Create ~/.vibe-check/.env
2. Add at least one API key (GEMINI_API_KEY, OPENAI_API_KEY, or OPENROUTER_API_KEY)
3. Restart Claude Code
If requirements are unclear:
The requirements need more detail. Let me ask some questions:
1. [Specific question about tech stack]
2. [Specific question about features]
3. [Specific question about constraints]
Once I have this information, I'll create a detailed plan.
If plan generation fails:
The plan generation failed. This could be due to:
- Invalid requirements format
- API rate limits
- Network issues
Let me try again with simplified requirements.
[Retry with more basic prompt]
Example Usage
Example 1: With BMAD PRD file
User: /titanium:plan ~/bmad/output/user-auth-prd.md
Claude:
- Reads PRD file
- Calls plan_parser tool
- Reviews generated plan
- Validates with vibe-check
- Creates plan.md
- Stores in Pieces
- Presents formatted summary to user
Example 2: With inline description
User: /titanium:plan
Claude: "What would you like to implement?"
User: "Add a search feature to filter products by name and category"
Claude:
- Writes description to .titanium/requirements.md
- Asks: "What's your tech stack?"
User: "React frontend, Node.js backend with PostgreSQL"
- Calls plan_parser tool with complete requirements
- Generates plan with appropriate tech stack
- Validates with vibe-check
- Presents plan to user
Example 3: Complex feature with multiple clarifications
User: /titanium:plan "Build a payment processing system"
Claude: "I need more details:
1. Which payment provider? (Stripe, PayPal, etc.)
2. What types of payments? (One-time, subscriptions, both?)
3. Currency support? (Single, multiple?)
4. Security requirements? (PCI compliance needed?)
5. Refund handling?"
User: "Stripe, one-time payments only, USD, yes PCI compliance, yes refunds"
Claude:
- Writes complete requirements
- Generates comprehensive plan
- vibe-check raises concern about PCI compliance complexity
- Adjusts plan to include @security-scanner more heavily
- Re-validates with vibe-check
- Presents approved plan
Voice Feedback
The voice hooks will automatically announce:
- "Starting plan generation" (when script is called)
- "Plan created with [X] epics" (when complete)
- "vibe-check validation complete" (after validation)
No additional voice calls needed - the hooks handle this automatically.
Next Command
After creating the plan, the user should run:
/titanium:work
This will execute the plan with orchestrated agent coordination.