The /develop command supports intelligent parallel development with automatic dependency detection.

$3

`
/develop 2, 3
│
▼
┌─────────────────────────────────────┐
│ Phase 0: Haiku Dependency Check │
│ Model: haiku (latest) + think hard │
│ Cost: ~$0.002 (100x cheaper) │
│ │
│ Analyzes: │
│ - Type/schema dependencies │
│ - Import dependencies │
│ - Data flow dependencies │
│ │
│ Output: parallel OR sequential │
└─────────────────────────────────────┘
│
├── No dependencies found?
│ └── Spawn ALL section agents in parallel
│ /develop 2, 3 → 10 Opus agents simultaneously
│
└── Dependencies found?
└── Execute sections sequentially
Phase 1: Section 2 (5 agents)
Phase 2: Section 3 (5 agents, uses Section 2 output)
`

$3

| Command | Sections | Dependency Check | Agents Spawned |
|---------|----------|------------------|----------------|
| /develop 2 | 1 | Skipped | 5 Opus |
| /develop 2, 3 | 2 (independent) | Yes | 10 Opus (parallel) |
| /develop 2, 3 | 2 (dependent) | Yes | 5 + 5 Opus (sequential) |
| /develop 2, 3, 4 | 3 (independent) | Yes | 15 Opus (parallel) |

$3

Each section spawns 5 parallel agents:

| Workstream | Focus | Output |
|------------|-------|--------|
| Schemas | Zod schemas, TypeScript types | src/schemas/ |
| Storage | Database models, persistence | src/storage/ |
| Core Logic | Business logic, algorithms | src/core/ |
| Implementation | API endpoints, UI components | src/features/ |
| Tests | Unit tests, integration tests | tests/ |

Extended Thinking

All agents have extended thinking enabled by default for deeper reasoning.

$3

| Agent | Model | Thinking | Purpose |
|-------|-------|----------|---------|
| @dev | Opus (latest) | ultrathink | Feature implementation |
| @qa-reviewer | Opus (latest) | ultrathink | QA review with 5-dimension framework |
| @debugger | Opus (latest) | ultrathink | Bug investigation |
| @code-reviewer | Haiku (latest) | think hard | Fast read-only review |
| @dependency-check | Haiku (latest) | think hard | Pre-flight dependency analysis |

$3

Use these keywords in prompts to control thinking depth:

| Keyword | Budget | Use Case |
|---------|--------|----------|
| think | Low | Simple lookups, formatting |
| think hard | Medium | Analysis, planning, Haiku tasks |
| think harder | High | Complex reasoning |
| ultrathink | Very High | DEFAULT - Implementation, architecture |
| megathink | Maximum | Critical decisions, error recovery |

$3

`markdown
Task: Implement user authentication

ultrathink

Requirements:
- JWT-based authentication
- Refresh token rotation
- Secure cookie storage
`

Directory Structure

`
.
├── .claude/ # Claude Code configuration
│ ├── CLAUDE.md # Main project memory
│ ├── agents/ # Agent definitions
│ │ ├── dev.md # Developer agent (opus + ultrathink)
│ │ ├── qa-reviewer.md # QA reviewer (opus + ultrathink)
│ │ ├── code-reviewer.md # Read-only reviewer (haiku + think hard)
│ │ └── debugger.md # Bug investigator (opus + ultrathink)
│ ├── commands/ # Custom slash commands
│ │ ├── develop.md # /develop command with dependency check
│ │ ├── qa.md # /qa command
│ │ ├── setup-project.md # /setup-project command
│ │ └── audit-config.md # /audit-config command
│ ├── skills/ # Context-forking skills
│ │ ├── develop-section/ # Parallel development (5 agents)
│ │ ├── dependency-check/ # Haiku pre-flight analysis
│ │ └── qa-section/ # QA cycle with fixes
│ ├── rules/ # Modular coding rules
│ │ ├── code-style.md
│ │ ├── testing.md
│ │ ├── security.md
│ │ └── api-design.md
│ ├── settings.json # Model/thinking defaults
│ └── hooks/ # Automation hooks
│ └── hooks.json
├── src/context/ # VectorDB infrastructure
│ ├── types.ts
│ ├── retrieval.ts
│ ├── storage.ts
│ └── index.ts
├── scripts/ # CLI tools
│ ├── init-project.sh
│ ├── seed-context.ts
│ ├── retrieve-context.ts
│ ├── store-journal-entry.ts
│ └── get-todo-section.ts
├── docs/ # Project documentation
├── todo/ # Task tracking
├── journal.md # Session history
├── .mcp.json # MCP servers (Context7)
└── package.json
`

Session Management Skills

Built-in skills for session workflow (automatically available):

| Skill | Purpose |
|-------|---------|
| /startagain | Bootstrap new session with VectorDB context |
| /journal | Record structured session entry |
| /sync | Git commit, pull --rebase, push |

Skills Reference

| Skill | Model | Thinking | Purpose |
|-------|-------|----------|---------|
| /develop-section | Opus | ultrathink | 5 parallel dev agents per section |
| /dependency-check | Haiku | think hard | Pre-flight dependency analysis |
| /qa-section | Opus | ultrathink | 5-dimension QA with auto-fixes |

VectorDB Context System

The template includes a semantic retrieval system using LanceDB:

`bash


Seed documentation into VectorDB

| Collection | Content |
|------------|---------|
| prd_sections | Documentation chunks |
| todo_snapshots | Task state history |
| journal_entries | Session summaries |
| session_summaries | Quick context |

Hooks

| Hook | Event | Action |
|------|-------|--------|
| SessionStart | Session begins | Check VectorDB availability |
| PostToolUse | After Write/Edit | Auto-format with Prettier |
| PreToolUse | Before Write/Edit | Validate protected files |

Context7 Integration

Include use context7 in prompts for up-to-date library documentation:

`
Implement user authentication using NextAuth. use context7
`

Customization

$3

Create .claude/agents/my-agent.md:
`yaml
---
name: my-agent
description: Description for auto-detection
model: opus
thinking: ultrathink
---

Agent instructions...

Create .claude/commands/my-command.md:
`yaml
---
description: Command description
argument-hint:
---

Command instructions...

Create .claude/skills/my-skill/SKILL.md:
`yaml
---
name: my-skill
description: Skill description
context: fork
model: opus
thinking: ultrathink
allowed-tools: Read, Write, Edit, Bash, Glob, Grep, Task
---

Skill workflow...

Create .claude/rules/my-rules.md and import in CLAUDE.md:
`markdown
@.claude/rules/my-rules.md
``

Cost Optimization

The dependency check pattern saves money:

| Scenario | Without Pre-Check | With Pre-Check |
|----------|-------------------|----------------|
| 2 independent sections | 10 Opus agents (~$0.50) | 10 Opus agents (~$0.50) |
| 2 dependent sections (fails) | 10 Opus agents wasted (~$0.50) | Haiku check (~$0.002) + 10 Opus sequential (~$0.50) |
| Savings on failure | $0 | ~$0.50 saved |

The ~$0.002 Haiku pre-flight check prevents wasting ~$0.50+ on parallel Opus runs that would fail due to missing dependencies.

Requirements

- Node.js >= 18
- npm
- OpenAI API key (for embeddings)
- Claude Code CLI

License

MIT