Don't just vibe. Produce.
The Four Phases
A systemic approach to building software with agents.
IMAGINE 🌱
Dream it. Before writing code, we explore, research, and design. Instead of jumping into implementation, we ask the agent to create a detailed plan.
The Protocol
Ask Claude to "make a plan for [feature]". It creates a markdown file in _PLANS/ detailing the approach, files to modify, and risks.
# _PLANS/AUTH.md
## Technical Approach
Use NextAuth with GitHub provider...
## Steps
1. [ ] Install dependencies
2. [ ] Configure providerPLANT 🥕
Seed it. Turn the plan into trackable work items. We break the large plan into small, manageable seeds (issues) that can be tracked.
The Protocol
Use the bd tool to create issues from the plan steps. This commits the work to your git history.
bd create "Implement OAuth provider" -t task
# Created seed #42GROW 🌿
Build it. This is where the work happens. Agents pick up seeds and implement them. We use specialized sub-agents for specific tasks like review, security, and performance.
The Protocol
Agents work in parallel. While you implement, a code-reviewer agent can check your work instantly.
# Specialized Agents
.claude/agents/code-reviewer.md
.claude/agents/security-auditor.md
.claude/agents/performance-auditor.mdHARVEST 🍊
Ship it. The harvest is the final quality gate. We run a standardized sequence of checks to ensure the code is ripe for production.
The Protocol
The /push command runs: Clean → Lint → Test → Build → Commit → Push. You can't skip steps.
/push
> Running harvest sequence...
> Linting... OK
> Testing... OK
> Building... OK
> Shipped! 🚀The Soil Specifications
Defines the environment where intelligence grows.
Your project's brain. Agents read this first, every time.
- Architecture & Patterns
- Phrase Triggers
- Conventions
# My App
> Build fast, ship clean
## Triggers
| "wash produce" | lint |Specialized workers for specific tasks.
code-reviewersecurity-auditorperformance-auditorcode-smell-auditor
# code-reviewer.md
- [ ] No console.logs
- [ ] Complexity < 10Think before you code. Detailed specs before implementation.
- Overview & Goal
- Technical Approach
- Implementation Steps
# FEATURE_auth.md
Status: Ready
## Steps
- [ ] Create Context
- [ ] Add Login UILiving health scores. Always know your project's vitals.
SECURITY.mdPERFORMANCE.mdTESTS.mdCODE_QUALITY.md
# SECURITY.md
Score: 8.5/10
- [ ] Update JWT
- [ ] Review RLSAutomate everything. One command does the work of ten.
/push– Ship it/navigate– Jump aroundjust dev– Start dev
# justfile
push:
just lint && just test
git pushAgents manage issues. You don't touch Jira. Learn more →
- Git-native issues
- Auto open/close
- Zero friction
bd create "Fix auth"
> Created bead #42How it Works in Real Life 🛠️
This isn't theory. This is the daily workflow of an agentic developer.
🌱 Imagine & Plan
I don't just start coding. I ask Claude to "make a plan for [feature]".
It explores the codebase and creates a detailed _PLANS/FEATURE.md spec.
🥕 Plant Seeds
I review the plan, then the AI Agent will convert the plan into Beads (local issues). Can then safely clear context window and focus on the Epic of task issues.
Convert the plan into an epic of issues🌿 Grow
They read the plan, write the code, and mark the bead as done. This can be done in parallel with multiple agents. With compounding effect of subagents and the PRODUCE harness for quick project discovery, letting the AI work at the highest quality for your project.
Implement from highest to lowest priority📊 Audit
Before shipping, I run the "wash the produce" or for me, "run it selectah!" command.
Agents run a battle-tested workflow testing errors, linting, etc and update the _AUDIT/ reports.
🍊 Deploy
Finally, I run the harvest sequence. It builds, tests, cleans up, audits and pushes to Github with confidence.
/pushReady to Start Producing?
Copy this prompt into Claude Code to instantly prepare your soil.