一键导入
flow-planner
Plan phases, tasks, and iterations. Use when structuring new work, adding features, or organizing development plans.
用 Codex 或 Claude 帮你安装 复制这段 Prompt,粘贴到 Codex、Claude 或其他助手里,让它检查 Skill 页面并帮你完成安装。
菜单
Plan phases, tasks, and iterations. Use when structuring new work, adding features, or organizing development plans.
用 Codex 或 Claude 帮你安装 复制这段 Prompt,粘贴到 Codex、Claude 或其他助手里,让它检查 Skill 页面并帮你完成安装。
Execute implementation work with gate checks and verification. Use when user wants to code, build features, implement iterations, or execute action items. Enforces gates, guides workflow, tracks progress.
Mark tasks and phases complete after verification. Use when finishing tasks, closing phases, or marking work done. Ensures all work verified before completion.
Manage backlog and archive completed work. Use for backlog operations, archiving phases, or keeping active plan focused. Curates plan organization.
Design features through brainstorming and make architectural decisions. Use when user wants to brainstorm, design approach, make architecture choices, or update PLAN.md.
Initialize Flow projects from scratch, migrate existing docs, or update old structures. Use when user says "start flow", "initialize", "migrate to flow", "set up flow project".
Navigate Flow projects with dashboard-first pattern. Use when user asks about status, current work, what's next, or project progress. Read-only skill.
| name | flow-planner |
| description | Plan phases, tasks, and iterations. Use when structuring new work, adding features, or organizing development plans. |
Help users plan and structure new work using Flow framework. Guide task structure decisions (standalone vs iterations) and suggest brainstorming for complex features.
Activate when the user wants to add new work:
Flow's Core Principle: Plan before code. Structure work hierarchically (phases → tasks → iterations) with clear boundaries.
Key Decision: Every task is EITHER:
NEVER mix both - This is the Golden Rule.
User wants to add work
↓
Is it complex/multi-step?
↓
YES → Task with Iterations
Break into 3-5 iterations
Each iteration = milestone
↓
NO → Standalone Task
Direct action items
Complete in one go
Always suggest brainstorming for:
Skip brainstorming for:
When users DO need brainstorming, help them understand how subjects get resolved:
When: Small blocking code change needed BEFORE iteration starts
Criteria: Required for iteration (blocking), small scope (< 30 min), can be done independently
Examples: Fix interface, rename file, update enum, fix bug
What happens: Action items go into "Pre-Implementation Tasks" section, must complete BEFORE main implementation
When: Architectural decision that affects system design
Criteria: No code changes yet, updates PLAN.md Architecture section NOW
Examples: Design pattern choice, API contract, data model
What happens: AI updates PLAN.md immediately during brainstorming
When: Subject answered by another subject's decision
Criteria: No independent decision needed, cascade from another subject
What happens: No action items, just note which subject resolved this
When: Substantial feature work that IS the iteration
Criteria: Main implementation work, takes significant time (> 30 min)
Examples: Build API endpoint, implement validator, create service
What happens: These action items become the iteration's implementation action items
Indicators: Single file change, clear requirements, no integration points, < 1 hour, similar to previous work
Examples: "Add validation function", "Fix typo", "Export function", "Add logging"
Guidance: "This looks straightforward - standalone task with direct action items. No brainstorming needed."
Indicators: Multiple approaches possible, affects architecture, integration needed, > 4 hours, user unsure
Examples: "Add authentication", "Integrate Stripe", "Implement caching", "Design database"
Guidance: "This is complex - I recommend brainstorming first. Let's discuss: [list 3-5 subjects]."
Indicators: Moderate complexity (2-4 hours), some design decisions, user hasn't expressed preference
Examples: "Add error handling to API", "Refactor data layer", "Implement search"
Guidance: "This could go either way. We could brainstorm first, or jump into iterations if you have a clear vision. Which do you prefer?"
Ask clarifying questions:
Simple → Standalone task Complex → Task with iterations Uncertain → Suggest brainstorming first
Present options to user:
I suggest structuring this as:
**Option A: Standalone Task** - "Add Feature X"
- Direct action items
- Single completion
- Estimated: 1-2 hours
**Option B: Task with 3 Iterations**
- Iteration 1: Basic implementation
- Iteration 2: Add advanced features
- Iteration 3: Polish and optimize
- Estimated: 4-6 hours
Which approach fits better?
Read DASHBOARD.md to find current phase and task count, then create appropriate files:
For new phase: Create .flow/phase-N/ directory, update DASHBOARD.md with new phase entry
For new task: Create .flow/phase-N/task-M.md, add to DASHBOARD.md progress overview
For new iteration: Add iteration section to existing task file
Use templates from TEMPLATES.md for proper structure.
Help user fill in:
When user wants to begin work on a phase:
When user wants to begin work on a task:
**Status**: ⏳ PENDING to **Status**: 🚧 IN PROGRESSWhen user asks "what's next" or "what should I work on":
Read DASHBOARD.md to find next ⏳ PENDING item:
Read current work context:
See TEMPLATES.md for complete templates:
DO:
DON'T:
Good Iteration Size:
Too Large (split into multiple iterations):
Too Small (combine or make standalone):
User: "I want to add real-time notifications"
AI Process:
1. Assess complexity: Complex feature (WebSocket, multiple components)
2. Suggest structure: Task with 3 iterations
3. Propose brainstorming: Technology choice, message format, persistence
4. Create task file with iteration structure
Structure Created:
**Task: Real-Time Notifications** (3 iterations)
**Iteration 1: Skeleton - WebSocket Connection**
- Goal: Establish basic WebSocket connection
- Set up WebSocket server, implement client connection, basic keepalive
**Iteration 2: Veins - Notification System**
- Goal: Core notification sending/receiving
- Design message format, server-side broadcast, client-side handlers
**Iteration 3: Flesh - UI & Polish**
- Goal: User-facing notification display
- Create UI component, add history, implement dismiss/mark-read
Before Starting:
Suggest brainstorming: WebSocket library choice (ws, socket.io, native)? Notification types? Persistence strategy?
User: "The search filter isn't working correctly"
AI Process:
1. Assess complexity: Simple bug fix (focused scope)
2. Suggest structure: Standalone task
3. Skip brainstorming: Clear requirements
4. Create task file with action items
Task Created:
**Task: Fix Search Filter Bug**
**Action Items**:
- [ ] Reproduce bug with test case
- [ ] Debug filter logic
- [ ] Identify root cause (likely query building)
- [ ] Implement fix
- [ ] Add unit test for this scenario
- [ ] Verify fix resolves issue
Estimated: 1-2 hours
User: "Let's brainstorm the retry logic for API calls"
AI Response:
Great! Let me suggest subjects to discuss:
Subjects:
1. Retry Strategy - Exponential backoff vs fixed delay?
2. Error Classification - Which errors trigger retry?
3. Max Retry Attempts - How many before giving up?
4. Existing Error Handler - Need to update ErrorHandler.ts for async?
5. Timeout Configuration - Per-request vs total operation?
I'll help categorize each decision by resolution type (A/B/C/D).
---
[After discussing Subject 4...]
User: "We need to update ErrorHandler.ts to support async operations"
AI Response:
This is Type A (Pre-Implementation Task):
✅ Blocking? YES - Can't implement retry without async support
✅ Small? YES - Estimated 20-30 minutes
✅ Independent? YES - Self-contained change
Resolution Type: A
Action Items (Pre-Implementation):
- [ ] Update ErrorHandler.ts to support async operations
- [ ] Add retryAsync() method
- [ ] Update 3 existing call sites
- [ ] Add tests for async error handling
Must complete BEFORE running implementation.
---
[After discussing Subject 1...]
User: "Let's use exponential backoff with base delay of 1 second"
AI Response:
This is Type D (Iteration Action Items):
✅ Substantial work? YES - Core feature implementation
✅ Takes > 30 min? YES - Estimated 2-3 hours
✅ Main iteration work? YES - This IS what we're building
Resolution Type: D
Action Items (Implementation):
- [ ] Create RetryPolicy class
- [ ] Implement exponential backoff algorithm (base: 1s, max: 32s)
- [ ] Add configuration for max retries (default: 3)
- [ ] Integrate with API client
- [ ] Add tests for retry scenarios
These become the iteration's implementation action items.
READ:
.flow/DASHBOARD.md - Find current phase, count tasks, check structure.flow/phase-N/task-M.md - Existing tasks (when adding iterations)WRITE:
.flow/phase-N/ - Create new phase directories.flow/phase-N/task-M.md - Create new task files, add iterations.flow/DASHBOARD.md - Add tasks/phases to progress overview, update status