skills/greenfield-one-shot/SKILL.md
One-shot greenfield build orchestration. Takes a project blueprint and executes the full build pipeline using subagents for each phase, including intermediate verification gates. Use when you want to run an entire greenfield project build in one session without manual intervention between phases.
npx skillsauth add barkbarkgoose/ai-agents greenfield-one-shotInstall this skill globally with one command. Works with Claude Code, Cursor, and Windsurf.
3 of 9 scanners reported clean
Some scanners were skipped, did not run, or reported a non-clean status. Review each row below.
Execute a complete greenfield project build from blueprint to working system using phased subagent execution.
You own the full project lifecycle — not just individual task orchestration, but all phases from initial decomposition through final polish and verification.
PROJECT_BLUEPRINT.md from greenfield-initThis is a full-lifecycle skill. One invocation handles decomposition, task creation, task execution across all phases, gate verification, and polish — until the project is complete.
.agent-tasks/tasks/[YYYYMMDD-project-name]/research/PROJECT_BLUEPRINT.mdresearch/API_CONTRACT.mdRead and internalize these skill files before starting:
./greenfield-decomposer/DECOMPOSITION_GUIDE.md./greenfield-decomposer/VALIDATION_GATES.md./task-creator/TASK_STRUCTURE_GUIDE.md./task-creator/TASK_CREATION_GUIDE.md./tailwind-auditor/SKILL.md (read during Phase 2)./vue3-typescript/SKILL.md (read during frontend work)./django-backend-dev/SKILL.md (read during backend work)You ARE the full project orchestrator. You do NOT spawn a separate orchestrator subagent. You own the entire greenfield project from Phase 0 through Phase 5 — including decomposition, task creation, task execution, gate verification, and phase transitions.
agent-transcripts/Your role:
PROJECT_BLUEPRINT.md to understand the full scopeAPI_CONTRACT.md if it exists.agent-tasks/tasks/[YYYYMMDD-project-name]/
├── ORCHESTRATOR.md # Main tracking document (you maintain this)
├── research/ # Blueprint and contracts
├── pending/ # Tasks not yet started
├── in-progress/ # Currently active tasks
├── complete/ # Finished tasks
├── testing/ # Testing strategies
└── agent-transcripts/ # Sub-agent execution transcripts
ORCHESTRATOR.md with project overview, phases, and task tracking tableIf a boilerplate scaffold already exists (e.g., cloned from a template), do not assume it is valid. Verify it works before skipping Phase 1:
Scaffold Validation Checklist:
uv run python manage.py check exits 0uv run python manage.py runserver starts without errorsnpm run dev starts without errorsuv run python manage.py migrate)If any check fails:
If all checks pass:
Decompose using greenfield-decomposer logic:
DECOMPOSITION_GUIDE.md and VALIDATION_GATES.md internallyExecute Phase 1 tasks:
For each task in Phase 1:
agent-transcripts/[phase-#]-[task-id]-transcript.mdIMPORTANT — Avoid subagents for these operations:
npm install — it is too slow and prone to timeout. Run it directly yourself before spawning subagents, or document that dependencies should be pre-installed.Verify Phase 1 gate:
Backend: Models and Migrations
./django-backend-dev/SKILL.md internallyuv run python manage.py makemigrations and migrateFrontend: Initial Components
./vue3-typescript/SKILL.md internally.form-input, .btn--primary, .alert--error) instead of long Tailwind utility stringsRun Tailwind Auditor (Mid-Phase) After frontend scaffolding is complete but before building out more components:
./tailwind-auditor/SKILL.md internally@layer components BEM classes are set up per DEFAULT_STACK.mdtailwind.config.jspending/./tailwind-bem-stylist/SKILL.md)Verify Phase 2 gate:
Backend: API Endpoints
Frontend: Feature Components
Subagent Spawning Pattern: For each work item, spawn ONE subagent. Do NOT spawn subagents for sub-items within a task—the subagent handles all work in that task.
Every subagent prompt MUST include the Project Stack block:
Subagent Prompt Structure:
-----------------------
## Project Stack
This is a greenfield project.
- Language/Runtime: [from PROJECT_BLUEPRINT.md]
- Framework: [from PROJECT_BLUEPRINT.md]
- Package manager: [from PROJECT_BLUEPRINT.md]
- Key libraries: [from PROJECT_BLUEPRINT.md]
- Test runner: [from PROJECT_BLUEPRINT.md]
---
[Task file contents follow]
---
## Final Step
REQUIRED: Save your full transcript to:
`.agent-tasks/tasks/[YYYYMMDD-project-name]/agent-transcripts/[phase-#]-[task-id]-transcript.md`
Wire Frontend to Backend
API_CONTRACT.md to understand endpoint shapesVerify Phase 4 gate:
Tailwind Audit + Consolidation
Error Handling + Edge Cases
Final Verification
Maintain this document throughout execution:
# Project: [Project Name]
**Created:** YYYYMMDD
**Blueprint:** research/PROJECT_BLUEPRINT.md
## Project Stack
- Language/Runtime: [from blueprint]
- Framework: [from blueprint]
- Package manager: [from blueprint]
- Key libraries: [from blueprint]
- Test runner: [from blueprint]
## Phases
| Phase | Name | Status | Validation Gate |
|-------|------|--------|-----------------|
| 1 | Scaffolding | Complete | Dev servers start |
| 2 | Data Layer + Auth | Complete | Admin shows models, auth forms work |
| 3 | Core Features | In Progress | [pending] |
| 4 | Integration | Pending | [pending] |
| 5 | Polish | Pending | [pending] |
## Task Summary
| Task ID | Phase | Name | Status | Agent |
|---------|-------|------|--------|-------|
| 001 | 1 | Create Django project | Complete | subagent |
| 002 | 1 | Create Vue project | Complete | subagent |
| ... | ... | ... | ... | ... |
## Progress Notes
**[Date] - Phase N: [Phase Name]**
- Tasks completed: [list]
- Issues encountered: [list]
- Next actions: [list]
## Current Phase Tasks
### Phase 3 Tasks
| Task ID | Name | Status | Dependencies |
|---------|------|--------|-------------|
| 015 | Create Project CRUD | In Progress | None |
| 016 | Create Task CRUD | Pending | 015 |
Every subagent prompt MUST include the Project Stack block — This is non-negotiable.
When you spawn a subagent, use EXPLICIT language:
GOOD:
Spawn a vue3-typescript subagent to execute Task 015.
The subagent should:
1. Read the task file at pending/phase-3-015-create-project-crud.md
2. Execute all work described in the task
3. Save transcript to agent-transcripts/phase-3-015-create-project-crud-transcript.md
4. Move task file from pending/ to complete/
BAD:
Use a subagent to work on the project CRUD task.
When a work item requires a specialized skill, spawn the appropriate agent type:
For django-backend-dev:
Spawn a django-backend-dev subagent for Task 012.
For vue3-typescript:
Spawn a vue3-typescript subagent for Task 015.
For tailwind-bem-stylist:
Spawn a tailwind-bem-stylist subagent for the button consolidation task.
Each specialized agent reads its corresponding SKILL.md internally. You do NOT need to include the skill file contents in the prompt—the agent knows to read it.
When you need to decompose a phase:
GOOD:
Spawn a subagent to decompose Phase 3 using ./greenfield-decomposer/DECOMPOSITION_GUIDE.md and ./greenfield-decomposer/VALIDATION_GATES.md.
The subagent should:
1. Read PROJECT_BLUEPRINT.md and any existing phase research docs
2. Identify all work items in Phase 3
3. Create research/PHASE-3-[phase-name].md following the format in DECOMPOSITION_GUIDE.md
4. Spawn task-creator subagents to create task files from the research document
Verify each phase gate before proceeding to the next phase. If a gate fails:
Scaffolding Gate:
uv run python manage.py check passesnpm run dev starts without errorsData Layer Gate:
uv run python manage.py makemigrations succeeds with no issuesuv run python manage.py migrate applies cleanlyAuth Gate:
Feature Gate:
Integration Gate:
Polish Gate:
npm run lint passes (frontend)uv run python manage.py check passes (backend)If a subagent fails:
agent-transcripts/[failed-task]-transcript.mdIf a phase gate fails:
MUST verify subagent completion: After each subagent completes:
in-progress/ to complete/After each phase completes:
When all phases are complete and all gates pass:
complete/tools
Use this skill when working on Vue 3 + TypeScript client-side code, including creating new components, refactoring existing UI, implementing store logic with Pinia, or building reusable composition functions.
data-ai
orchestration skill for tasks, takes a task folder as input and runs one sub-agent for each individual task file. Should NOT execute or make any changes on its own, only sub-agents may do that.
tools
create tasks as files in local project directory
data-ai
archives a local agent task directory so it can be recalled for future reference