Status: Active

Development Philosophy

Vibecoding

Best practices for AI-assisted development.
Extracted from practitioners who are building the future.

Patterns distilled from real-world experience

The Insight

Code is not precious.

For decades, we treated code as a carefully crafted artifact.
With AI assistance, code can be regenerated faster than it can be debugged.
The specification becomes the durable object.

Core Tenets

Six principles from the field

1

The Spec is the Durable Object

Code is regenerable. Specifications persist.

2

Rewrite Over Edit

v1 is just the spec for v2. Embrace regeneration.

3

Plan Before You Vibe

Upfront planning survives context compaction.

4

Fresh Eyes Beat Tired Context

Subagents see what polluted contexts miss.

5

Clear Early and Often

Don't wait until forced to compact. Be proactive.

6

Don't Fight the Model

Work with model strengths, not against them.

The Pattern

Iterative Regeneration

Each version is a full regeneration, not incremental patches

v1

Learn what's wrong

→

v2

Simplify

→

v3

Edge cases

→

v4

Production

Typical pattern: 4 versions before satisfaction. Let go of the code.

Avoid These

Anti-Patterns

Precious Code Syndrome

Spending hours debugging when regeneration takes minutes.

Context Hoarding

Keeping context "just in case" until forced compaction.

Solo Agent Everything

One agent for complex tasks when parallel would be faster.

Fighting the Model

Repeatedly forcing approaches the model resists.

Plan-Free Vibing

Starting implementation without written plan.

Included Behaviors

Composable Pattern Modules

fresh-eyes

Spawn clean subagents for unbiased review

plan-persistence

Maintain PLAN.md that survives compaction

clear-early

Heuristics for proactive context management

understanding-md

Working memory file across sessions

subagent-coordinator

Patterns for parallelizing work via subagents

question-bubbling

Protocol for subagent question escalation

# Pick and choose what you need
includes:
  - bundle: vibecoding:behaviors/fresh-eyes.yaml
  - bundle: vibecoding:behaviors/plan-persistence.yaml
        

Automation

Ready-to-Run Recipes

Recipe	What It Does
overnight-code-review	Multi-agent parallel code review with consolidated reporting
spec-iteration	Two-model spec building through critique cycles
token-burner	Automated code improvement loop for overnight runs
antagonistic-panel	Multi-perspective analysis (VC panel, strategy panel)
wiggum-loop	Iterate on a task until success criteria are met

# Run overnight reviews while you sleep
"run the overnight-code-review recipe on src/"
        

Living Knowledge

Extract patterns from conversations

The bundle includes tools to analyze practitioner chats and extract new patterns

# 1. Export your chat and place in exports/

# 2. Run the analyst
"use the chat-analyst to review new messages"

# 3. Review generated reports in docs/reports/

# 4. New patterns added to state/ideas.yaml

The knowledge base grows from real practitioner experience, not theory.

Getting Started

Three ways to use it

Full Bundle

All patterns, behaviors, and recipes. Opinionated but comprehensive.

bundle: git+https://github.com/ramparte/vibecoding@main

As Include

Add to your existing bundle. Mix with other capabilities.

includes:
  - bundle: vibecoding

Pick Behaviors

Choose only the patterns you want. Maximum flexibility.

includes:
  - vibecoding:behaviors/fresh-eyes.yaml

The Opportunity

Early Adopter Window

Most developers are still at "AI can summarize my emails."
Those who master these patterns have leverage until the world catches up.

Use this time wisely.

Sources & Methodology

How this was researched

Vibecoding repository: ramparte/vibecoding (private, created January 16, 2026). Patterns described are based on practitioner experience documentation. Bundle behaviors and recipes referenced have not been independently verified against the private repository.

Data as of February 20, 2026.

Primary contributor: Sam Schillace (ramparte)

Specs persist.
Code regenerates.

github.com/ramparte/vibecoding

Patterns extracted from practitioner community
Community patterns for the Amplifier ecosystem