Kai Master Builder

Build anything efficiently with structured plans and automated execution.

When to Use

• User wants to build a new app/feature/fix
• Need structured approach to complex project
• Want automated building via cron
• Need to track progress across sessions

The Process

Phase 1: Create Project Plan

Input: GOAL (what to build/improve/fix)

Output: PROJECT/PROJECT_PLAN.md

# Project Plan: [PROJECT_NAME]

## Overview
What this project does and why.

## Goals
- Primary goal
- Secondary goals

## Architecture/Approach
How it will be built.

## Security Checklist
- [ ] No API keys hardcoded (use env vars)
- [ ] No .env loading in scripts
- [ ] All secrets in config, not code
- [ ] Input validation
- [ ] Error handling

## Risks & Mitigations
| Risk | Impact | Mitigation |
|------|--------|------------|
| ... | ... | ... |

## Dependencies
- External: ...
- Internal: ...

## Milestones
1. [ ] Phase 1: ...
2. [ ] Phase 2: ...
3. [ ] Phase 3: ...

## Validation
How to test/validate each milestone.

Phase 2: Create Task List

Output: PROJECT/PROJECT_TASKS.md

# Tasks for: [PROJECT_NAME]

## Task Format
- [ ] TASK_NAME | status: TODO | depends: none | validated_by: test command

## Tasks

### Phase 1: Setup
- [ ] Create project structure | status: TODO | depends: none | validated_by: ls project/
- [ ] Setup dependencies | status: TODO | depends: 1 | validated_by: pip list / npm list

### Phase 2: Core Feature
- [ ] Implement X | status: TODO | depends: 1 | validated_by: python3 test_x.py

### Phase 3: Testing & Polish
- [ ] Write tests | status: TODO | depends: 2 | validated_by: pytest
- [ ] Update docs | status: TODO | depends: 2 | validated_by: doc builds

## Blocked
- Task X: waiting on external API access

## Done
(Completed tasks move here with timestamp)

Phase 3: Build!

Prompt provided to Agent/User:

BUILDER PROMPT

You are building: [PROJECT_NAME]

Context:

• Read PROJECT/PROJECT_PLAN.md for architecture and goals
• Read PROJECT/PROJECT_TASKS.md for task list
• Follow the iteration loop below

Iteration Loop:

1. Pick a task - Choose TODO task whose dependencies are all DONE
2. Implement it - Write the code
3. Test it - Run validation command
4. Document - Update CHANGELOG.md with what you built
5. Mark DONE - Update TASKS.md, move to Done section with timestamp
6. Report - Report back to main session

If blocked:

• Mark task as BLOCKED with reason
• Report what you need to proceed

When all tasks done:

• Report: "PROJECT is DONE! Built with kai-master-builder."
• Update PROJECT_README.md with final state

Automation

Setup cron for autonomous building:

/call_skill kai-master-builder --build PROJECT_NAME

Or let Agent run it via sessions_spawn with this prompt.

Project Folder Structure

PROJECT/
├── PROJECT_PLAN.md      # Architecture and goals
├── PROJECT_TASKS.md     # Task list with status
├── PROJECT_CHANGELOG.md # What was built
├── PROJECT_README.md    # Documentation
└── src/                 # Source code

Tips

• Break big tasks into < 1 hour chunks
• Every task should have clear validation
• Mark tasks BLOCKED if waiting on something
• Update CHANGELOG after each task
• Backup TASKS.md before modifying

kai-master-builder: Turn goals into working code