Start a New Project
Receiving a client brief and turning it into a working skeleton in the same day is the mark of a high-velocity team. With Sun Agent Kit, you feed the brief directly into the toolchain β architecture decisions, folder structure, CI setup, and a development plan all come out the other side before you write a single line of business logic.
Overview
Goal: Go from a client brief to a running, planned project skeleton ready for feature development
Time: 30β60 minutes (vs 4β8 hours manual)
Agents used: planner, reviewer
Commands: /sk:bootstrap, /sk:plan, /sk:scout, /sk:docs
Prerequisites
- Sun Agent Kit installed (installation guide)
- Node.js 18+ or Bun runtime available
- A project brief β plain text, PDF, or a URL to a spec doc is fine
- Git initialized in your target directory (
git init) - Access credentials for any third-party services named in the brief (can be added later)
Step-by-Step Workflow
Step 1: Bootstrap the Project Skeleton
/sk:bootstrap "Node.js REST API for an e-commerce platform β PostgreSQL, Express, JWT auth, product catalog, cart, orders"
What happens: The agent:
- Analyzes your project description to detect tech stack, database, and domain areas
- Generates folder structure with routes, controllers, services, and models
- Installs framework and development dependencies
- Creates configuration files (
.env.example,tsconfig.json,Dockerfile) - Runs initial typecheck to verify the skeleton compiles clean
Step 2: Generate the Implementation Plan
/sk:plan "Three sprints: Sprint 1 auth + catalog, Sprint 2 cart + orders, Sprint 3 payments + admin panel"
What happens: The agent:
- Reads the generated codebase structure and builds a dependency graph
- Writes phase files for each sprint under the
plans/directory - Breaks each phase into tasks and resolves dependencies between them
- Identifies the critical path and saves an overview to
plans/plan.md
Step 3: Review the Architecture
/sk:scout "review the generated structure and flag any architectural concerns before development starts"
What happens: The agent:
- Scans the generated scaffold and reads project files
- Checks for missing middleware, empty directories, and structural gaps
- Flags warnings (e.g., missing rate limiting, empty migrations directory)
- Saves an architecture report to
plans/reports/scout-report.md
Step 4: Generate Initial Documentation
/sk:docs "generate README, API overview, and local dev setup guide"
What happens: The agent:
- Reads configuration files to understand the project setup
- Discovers route stubs and documents expected request/response shapes
- Writes README, API overview, and local dev setup guide
- Verifies that internal cross-references in the docs resolve correctly
Step 5: Commit the Baseline
/sk:git cm
What happens: The agent:
- Stages all project files, respecting
.gitignore(.envexcluded automatically) - Writes a conventional commit message describing the scaffold
- Creates the initial commit as the baseline for Sprint 1
Complete Example: Client Sends Brief for an E-Commerce App
Scenario: A client emails a 2-page brief on a Tuesday afternoon. They need an MVP backend for an online apparel store: user registration, product browsing, cart, checkout with Stripe, and an admin panel to manage inventory. Delivery is 6 weeks.
Monday morning you run:
# 1. Bootstrap from the brief text (paste or pipe it in)
/sk:bootstrap "E-commerce backend: user auth (email + Google OAuth), product catalog with categories and variants, shopping cart with session persistence, Stripe checkout, order history, admin CRUD for products and orders. Stack: Node.js, Express, PostgreSQL, Redis for sessions."
# 2. Plan across the 6-week delivery timeline
/sk:plan "6 weeks, 2-week sprints. Sprint 1: auth + catalog. Sprint 2: cart + checkout. Sprint 3: orders + admin."
# 3. Scout for problems before the team picks up tickets
/sk:scout "review generated scaffold for security gaps, missing middleware, and any structural issues"
# 4. Write docs the client can review
/sk:docs "README with setup instructions, API contract stubs, deployment notes for the client's AWS environment"
# 5. Commit and push
/sk:git cm
Result: By mid-morning, the team has a compiling project skeleton, a multi-phase plan with tasks organized across phases, a scout report flagging warnings, and clean documentation the client can read. Feature development starts immediately.
Time Comparison
| Task | Manual | With Sun Agent Kit |
|---|---|---|
| Folder structure + config files | 1β2 hours | minutes |
| Dependency research + install | 30β60 min | minutes |
| Sprint plan + task breakdown | 1β2 hours | minutes |
| Architecture review | 30β60 min | minutes |
| README + initial docs | 30β60 min | minutes |
| Total | 4β8 hours | 30β60 minutes |
Best Practices
1. Put domain language in the bootstrap prompt β
The more specific your description β data models, third-party services, team constraints β the more accurate the scaffold. βNode.js API with Stripe and PostgreSQLβ produces a better result than βweb backendβ.
2. Run scout before committing to the plan β
The scout agent reads the generated skeleton and catches structural issues before they become technical debt. Run it between bootstrap and plan, not after.
3. Donβt skip /sk:plan even for small projects β
Even a 2-week project benefits from a written plan. It doubles as onboarding material for new teammates and a scope reference for the client.
4. Feeding a vague one-liner to /sk:bootstrap β
/sk:bootstrap "build a website" generates a generic skeleton that will need substantial rework. Spend 10 minutes extracting specifics from the client brief first.
5. Committing .env files β
Sun Agent Kitβs commit command respects .gitignore automatically, but double-check before push. Never commit real credentials, even to a private repo.
Troubleshooting
Problem: Bootstrap hangs at dependency install
Solution: Check your network and npm/bun registry access. If install continues to fail, scaffold the project first and then run npm install manually after bootstrap completes.
Problem: Plan phase files are too granular (100+ tasks per phase)
Solution: Your sprint description is too open-ended. Narrow each sprint to 2β3 major features: /sk:plan "Sprint 1: user registration and JWT auth only".
Problem: Scout report is empty
Solution: The scout agent needs at least some source files to analyze. Run after /sk:bootstrap completes, not before.
Problem: TypeScript errors in the generated skeleton
Solution: Run /sk:fix "fix all TypeScript errors in the generated scaffold" β the debugger agent resolves import and type errors in generated files.
Next Steps
- Onboard to Existing Codebase β use when youβre taking over from another team instead of starting fresh
- Add a New Feature β the next step after your scaffold is in place
- Build a REST API β go deeper on API design and endpoint implementation
Key takeaway: A client brief should never sit in your inbox for more than an hour β /sk:bootstrap turns it into a compiling project skeleton with a written plan while you drink your coffee.