From vibe coding to full stack AI development.
The AI agent management system that brings structure, memory, and precision to every project you build with AI.
Quick Start · Commands · Community Skills · Marketplace · cosama.co
Part of the CodeStacker Marketplace — AI prompts, skills & agentic workflows for building with AI
Built by CoSama — AI Development Company
⚠️ Early Development — v1.0.0 This is an early-stage release. The system is functional and ready to use, but expect active development, evolving documentation, and new features as the project grows. Contributions and feedback are welcome.
Vibe coding is how most people use AI to build software today — and it's costing them.
You open a chat, describe what you want, and ask the AI to "just build it." The AI generates code. You paste it in. It kind of works. You ask for a fix. It breaks something else. You paste that fix in. Now something else is wrong. You're not building a product — you're playing whack-a-mole with an AI that has no idea what your actual codebase looks like, no memory of what it did last session, and no plan beyond the last prompt.
Vibe coding works for demos, prototypes, and throwaway scripts. For anything real — it creates technical debt faster than you can ship.
The symptoms are familiar:
- AI contradicts itself between sessions
- Generates code that doesn't match your existing patterns
- Adds features you didn't ask for
- Skips documentation and test coverage
- Drifts from the original spec halfway through
- Leaves you asking "what did it actually change?"
Full Stack AI Coding is the professional standard for building production software with AI. It treats your AI assistant as a disciplined engineering partner — not a magic text generator.
It means the AI knows your codebase before it touches it. It plans before it builds. It follows rules you've set. It documents everything it does. It stays on scope. It hands off cleanly between sessions. And when it makes a mistake, it records it so it doesn't make it again.
Full Stack AI Coding is:
- Specification-driven — every task starts with a written plan
- Structure-aware — AI reads your actual project before acting
- Traceable — every decision, change, and session is documented
- Rule-governed — behavioral constraints enforced on every request
- Context-managed — memory maintained across sessions and handoffs
- Skill-informed — domain expertise loaded for every technology in the stack
This is how you build real products with AI, at pace, without chaos.
We built CodeStacker because the tools weren't keeping up with how developers actually work with AI.
As AI coding assistants became genuinely powerful, a new problem emerged: the AI itself became the bottleneck. Not because it wasn't capable — but because there was no system to manage it. Every session started from zero. Every decision was undocumented. Every mistake was repeated. Teams were spending more time corralling their AI tools than building their products.
We needed a layer between the developer and the AI. Something that enforced discipline, preserved context, and turned an unpredictable assistant into a reliable engineering partner. Something that lived in the project, worked with any AI tool, and required no external services to get started.
That's .codestacker.
.codestacker is a folder-based AI agent management system that sits inside your project and governs how your AI assistant behaves. It enforces a structured specify → plan → implement workflow. It keeps the AI on scope, on spec, and fully documented — from the first line of code to handoff.
One folder. Any AI. Total control.
-
📋 Plan-First Workflow — Every task starts with a written plan and checklist. The AI cannot implement until you review and approve. No more "what did it just do?"
-
🚫 Zero Drift — Strict scope enforcement. The AI works only what's in the approved plan. No added features, no scope creep, no surprise refactors.
-
🔍 Structure-Aware Planning — Before every plan, the AI reads your actual project structure, schema, and architecture docs. Plans are based on what's real, not what it assumes.
-
🛠️ Skill System with Auto-Builder — AI capabilities are loaded from skill files for every technology in your stack. Missing a skill? The built-in CodeStacker Skill Builder auto-generates one on the spot, then flags it for upgrade to an engineering-grade version from the marketplace.
-
🏷️ Tag-Based Skill Registry — Skills are indexed by tag (
#react,#supabase,#docker) for instant lookup. Query withMode Skill #[tag]— the AI finds, loads, or builds the right skill automatically. -
📝 Full Session Traceability — Every session is recorded. Every decision documented. Every error logged. You always know what the AI did, when, and why.
-
🧠 Context Window Management — The AI monitors its own context, writes a full summary at 80%, and signals when a clean handoff is needed. No lost context between sessions.
-
📊 Error Reporting & Learning — When the AI makes a mistake, it documents it in
reports/. Patterns are visible. Rules can be tightened. The same mistake doesn't happen twice. -
🎮 Mode Command System — Simple, memorable commands (
Mode Plan,Mode Approved,Mode Fix,Mode Finish) control exactly what the AI is allowed to do at any moment. -
🎭 Props — Specialized AI Personas (Marketplace) — Load a Prop to instantly transform your AI into a role-specific expert: senior engineer, API architect, security auditor, and more. Each Prop bundles a Persona, system instructions, and a matched skills set.
-
⚙️ Agentic Workflows (Marketplace) — Deploy coordinated agent teams for complex, assignment-oriented tasks. Spec Agent → Planner → Implementer → Tester → Documenter — fully managed, fully traced.
-
🔌 Tool Agnostic — Works with Cursor, GitHub Copilot, GPT, Gemini, or any AI assistant that reads files. No lock-in.
-
🌐 Community Skills Library — 2,800+ community-built skills available via CodeStacker-Skills, curated from the open-source AI development ecosystem.
The CodeStacker Marketplace is the home of engineering-grade AI development resources — built for developers who build with AI professionally.
Browse and install:
- Skill Packs — Expert-curated skill files for every major language, framework, database, and tool. Production-ready patterns, edge cases, and best practices — far beyond what the auto-builder generates.
- Props — Specialized AI system prompts that turn your assistant into a domain expert with the right persona, instructions, and skills pre-loaded.
- Workflows — Agentic workflow pipelines that deploy coordinated agent teams to handle complete development assignments from spec to delivery.
The
.codestackerbase system is free and open source. The marketplace is where you go to level up.
CoSama is the AI development company behind CodeStacker. We build tools, systems, and infrastructure for developers who use AI as a core part of their engineering workflow — not as a shortcut, but as a first-class development partner.
CodeStacker is our answer to the question every serious AI developer eventually asks: "How do I actually manage this thing?"
We're building the tooling layer that makes Full Stack AI Coding the default — not the exception.
Option A — One-liner (Mac/Linux)
curl -sSL https://raw.githubusercontent.com/CoSama-Ai/CodeStacker/main/install.sh | bashOption B — One-liner (Windows PowerShell)
iwr -useb https://raw.githubusercontent.com/CoSama-Ai/CodeStacker/main/install.ps1 | iexOption C — Git (Manual)
git clone --depth 1 https://github.com/CoSama-Ai/CodeStacker.git /tmp/cs
cp /tmp/cs/stack.md .
cp -r /tmp/cs/.codestacker .
rm -rf /tmp/csOption D — ZIP Download
Download the ZIP from GitHub → Code → Download ZIP, extract it, then copy stack.md and the .codestacker/ folder into your project root.
⚠️ Windows File Explorer won't show.codestackerby default (hidden folder) — use PowerShell or enable "Show hidden items".
The core instruction file is stack.md — it tells any AI how to operate under .codestacker. You have three integration options:
Option A — Automatic (Recommended)
Run Mode Start or Mode New Project. The system detects your existing AI config files (CLAUDE.md, .cursorrules, AGENTS.md, .github/copilot-instructions.md, etc.) and asks to add a reference to stack.md. Your existing instructions stay intact.
Option B — Create New
If you have no AI config files yet, the system offers to create the appropriate one for your tool, pointing to stack.md.
Option C — Manual
Tell your AI: "Read stack.md and follow it."
⚠️ Manual mode may conflict with existing tool instructions.
Open a chat with your AI assistant. The system auto-detects first run:
You: (start chat)
AI: 🔍 Detected first run — initializing .codestacker...
I found existing AI instruction files:
- CLAUDE.md
→ Add .codestacker reference? (A) Add / (B) Manual / (C) Skip
Please answer a few questions:
1. What is your project name?
2. What language/framework?
3. Do you have existing code?
4. What's the main goal?
Once initialized, use mode commands to control the AI:
You: Mode Plan - Add user authentication
AI: 📋 Creating plan with checklist...
[creates plan in .codestacker/project/plan/]
You: Mode Approved
AI: ✅ Executing plan...
[implements with full documentation]
| Command | Description | When To Use |
|---|---|---|
Mode New Project |
Initialize entire system with project details | First time setting up |
Mode Start |
Verify system, check docs, report status | Beginning of each session |
Mode Plan |
Create detailed plan + checklist | Before any new feature/task |
Mode Approved |
Execute current plan | After reviewing and approving a plan |
Mode Free |
Quick tasks — skip planning, still document | Minor tweaks, small fixes |
Mode Fix |
Create fix plan and implement | When bugs/errors are identified |
Mode Proceed |
Continue last approved plan | After interruption or context switch |
Mode Finish |
Full summary + structure update + final check | End of session or milestone |
| Command | Description | When To Use |
|---|---|---|
Mode Skill Library |
List all installed skills | See what's available in your skills folder |
Mode Skill #[tag] |
Filter skills by tag (#ui, #supabase, #react) |
Find skills for a specific technology |
Mode Skill Build |
Build a new skill with the skill-builder | Create a custom skill file |
Mode Skill Import |
Import from CodeStacker-Skills community repo | Add pre-built community skills |
Mode Skill Index |
Rebuild the skill registry | After adding new skill files manually |
💡 Skills are automatically assessed during
Mode Plan— the AI scansskills/SKILL_REGISTRY.md, builds any missing skills on the spot with the Skill Builder, and references them inline in the plan.
| Command | Description | When To Use |
|---|---|---|
Mode Props |
Load a specialized AI Prop (requires API key) | When you need expert-level AI persona + skills |
Mode Workflow |
Run a specialized agentic Workflow (requires API key) | When you need a full agent team for an assignment |
🔑 Props and Workflows require a CodeStacker API key. Get access at codestacker.co →
| Command | Description | When To Use |
|---|---|---|
Mode Reset |
AI rereads entire system | If AI seems confused or off-track |
Mode Report |
Document why AI failed | After any mistake |
Mode Rule |
Add new rule to rules doc | Establishing new project standards |
Mode Help |
Display all commands | Any time you need a reminder |
your-project/ # ← your project root
│
├── stack.md # 🚀 AI ENTRY POINT — the only CodeStacker file in your root
│
└── .codestacker/ # 📦 Entire system lives here — clean, contained, hidden
├── brief.md # 🎯 Core memory + golden rules (read on EVERY request)
├── .codestacker_init # Marker file — detects first run
│
├── setup/ # 🔧 One-time setup (internal — auto-invoked on first run)
│ └── project_start.md
│
├── rules/ # 🤖 What AI can/cannot do
│ ├── 00_golden_rules.md
│ ├── 01_behavioral_rules.md
│ ├── 02_code_standards.md
│ ├── 03_communication_rules.md
│ └── 04_security_rules.md
│
├── docs/ # 📚 Your architecture & design notes
│ ├── 01_architecture.md
│ ├── 02_ui_design.md
│ ├── 03_api_design.md
│ └── 04_data_models.md
│
├── build/ # 🏗️ What to build (specs)
│ ├── 01_project_overview.md
│ ├── 02_phase_1_spec.md
│ └── backlog/
│
├── props/ # 🎭 AI Props (Marketplace — requires API key)
│ └── README.md
│
├── workflows/ # ⚙️ Agentic Workflows (Marketplace — requires API key)
│ └── README.md
│
├── skills/ # 🛠️ AI capabilities registry
│ ├── SKILL_REGISTRY.md # 🏷️ Tag-based index (AI reads during Mode Plan)
│ ├── INDEX.md # Full skills catalog by category
│ ├── community.md # Install guide → CodeStacker-Skills repo
│ ├── skill-builder/ # 🔨 Auto-builds missing skills
│ │ ├── SKILL.md
│ │ └── templates/skill-template.md
│ ├── languages/ # JS, TS, Python, Go, Rust, PHP, Ruby
│ ├── frameworks/ # Next.js, Express, Django, FastAPI, Laravel
│ ├── databases/ # PostgreSQL, MongoDB, Supabase, Prisma, Redis
│ ├── testing/ # Jest, Playwright, Pytest, Vitest, Cypress
│ └── devops/ # Docker, GitHub Actions, Vercel, Nginx
│
├── reports/ # 📊 Failure analysis & corrections
│ ├── 01_error_log.md
│ ├── 02_drift_reports.md
│ ├── 03_misinterpretations.md
│ └── 04_system_violations.md
│
└── project/ # 📁 All project tracking lives here
├── plan/ # 📋 Plans + checklists
│ ├── 01_master_plan.md
│ ├── 02_current_sprint.md
│ ├── 03_checklists/
│ └── 04_completed/
├── sessions/ # 📝 Every conversation recorded
│ ├── session_log.md
│ └── archive/
├── summaries/ # 📄 Work summaries + handoffs
│ ├── latest_summary.md
│ ├── handoff_notes.md
│ └── archive/
└── structure/ # 🔍 Complete project analysis
├── 01_project_overview.md
├── 02_database_schema.sql
├── 03_environment.md
├── 04_file_structure.md
├── 05_dependencies.md
├── 06_credentials.md
└── 07_evolving_state.md
Every folder includes a
README.mdexplaining what goes there and how to use it.
Pre-built skill files for common languages, frameworks, databases, and testing tools — curated from many contributors across the AI development ecosystem.
Skills are gathered, reviewed, and adapted from open-source communities including Awesome OpenClaw Skills (2,800+ community-built AI skills), direct contributor PRs, and the CoSama team. All skills are reviewed for quality and compatibility before inclusion.
| Skill Category | Examples |
|---|---|
| Language Skills | JavaScript, TypeScript, Python, Go, Rust |
| Framework Skills | Next.js, Express, Django, FastAPI |
| Database Skills | PostgreSQL, MongoDB, Supabase, Prisma |
| Testing Skills | Jest, Pytest, Playwright, Vitest |
| Method | Use Case | Command |
|---|---|---|
| Full Install | Want everything | Clone repo → copy all folders into skills/ |
| By Category | Only need frameworks + databases | Clone repo → copy specific folders |
| Individual | Just need one skill | curl a single file from GitHub raw |
| AI-Assisted | Let the AI handle it | Mode Free - Install the Next.js skill from CodeStacker-Skills |
# Full install example
git clone https://github.com/CoSama-Ai/CodeStacker-Skills.git /tmp/cs-skills
cp -r /tmp/cs-skills/language_skills/* .codestacker/skills/languages/
cp -r /tmp/cs-skills/framework_skills/* .codestacker/skills/frameworks/
cp -r /tmp/cs-skills/database_skills/* .codestacker/skills/databases/
cp -r /tmp/cs-skills/testing_skills/* .codestacker/skills/testing/
cp -r /tmp/cs-skills/devops_skills/* .codestacker/skills/devops/
rm -rf /tmp/cs-skills# Single skill example
curl -o .codestacker/skills/frameworks/nextjs.md \
https://raw.githubusercontent.com/CoSama-Ai/CodeStacker-Skills/main/framework_skills/nextjs.mdSee skills/community.md for full installation guide.
💡 Can't find a skill for your stack? The CodeStacker Complete AI Coding Toolkit includes curated skill packs for every major language, framework, and tool.
Take .codestacker further with Props and Workflows — premium features available through the CodeStacker Marketplace.
Props instantly transform your AI assistant into a highly capable, role-specific expert by bundling a Persona + System Instructions + Skills into a single ready-to-use file.
Mode Props - senior-fullstack-engineer
AI: 🎭 Loading Prop: senior-fullstack-engineer
✅ Persona active — Expert full-stack engineer
✅ Instructions loaded — Architecture-first approach
✅ Skills bundled — React, Node, PostgreSQL
Ready to build.
Workflows deploy coordinated agent teams that handle complete, assignment-oriented development tasks end-to-end — from spec to implementation to testing to docs.
Mode Workflow - feature-development - Add user authentication
Workflow: ⚙️ Deploying Feature Development Workflow...
Stage 1: Spec Agent → Stage 2: Planner → Stage 3: Implementer
→ Stage 4: Tester → Stage 5: Documenter
✅ Complete. Summary in project/summaries/.
Both features require a CodeStacker API key.
See props/README.md and workflows/README.md for full details.
You: Mode Plan - Add user profile page with avatar upload
AI: 📋 Reading project structure...
🔍 Checking existing patterns...
Plan ready:
- Create Profile component
- Add avatar upload handler
- Update database schema
- Add tests
Approve with "Mode Approved" to proceed.
You: Mode Approved
AI: ✅ Executing plan...
[implements each checklist item]
[updates structure doc]
[records session]
You: Mode Free - Change button color from blue to green
AI: ⚡ Free mode active — skipping planning, still documenting
[Makes change]
[Updates session log]
Done. Return to planned work with "Mode Proceed".
You: Mode Fix - Login form throwing 500 error
AI: 🔧 Fix mode activated
[Creates fix plan]
Found: Missing DATABASE_URL env var
Fix plan: Add var → Update connection settings
Approve fix?
You: Mode Approved
AI: [Implements fix]
[Documents in reports/ why error occurred]
You: Mode Finish
AI: 📊 Generating final summary...
- Completed: User profile page
- Files modified: 5
- Structure updated: Yes
Handoff document created. Use "Mode Start" next session.
DO:
- ✅ Start every session with
Mode Start - ✅ Always review plans before approving
- ✅ Use
Mode Freesparingly — only tiny tweaks - ✅ Run
Mode Finishat natural stopping points - ✅ Keep
docs/updated with architecture decisions
DON'T:
- ❌ Never approve a plan without reading it
- ❌ Don't let AI implement without a plan (except
Mode Free) - ❌ Never ignore structure docs
- ❌ Don't let context grow too large — start fresh sessions
The AI monitors context and alerts you at 80%:
⚠️ Context at 82% — recommend new chat
Summary written to summaries/
Start fresh with "Mode Start" to continue
AI seems confused or off-track:
You: Mode Reset
AI: 🔄 Resetting... rereading entire .codestacker system
✅ System reloaded
Missing documentation:
AI: ⚠️ Missing: project/structure/02_database_schema.sql
Please provide your database type and existing tables.
AI made a mistake:
You: Mode Report
AI: 📝 Documenting failure...
Report saved to reports/
Ready to proceed with fix.
Q: Do I need to use all the folders? A: Yes — each folder serves a critical purpose. The system works as a whole.
Q: Can I use this with any AI assistant?
A: Yes. .codestacker works with any AI that can read/write files — GPT, Gemini, Cursor, Copilot, and more.
Q: What if I'm starting a brand new project?
A: Use Mode New Project first. The AI will guide setup.
Q: How do I hand off to another developer?
A: Run Mode Finish — it generates a complete handoff doc in summaries/handoff_notes.md.
Q: Can I customize the rules?
A: Yes — use Mode Rule to add project-specific rules, saved in rules/.
Q: What if the AI isn't following the system?
A: Try Mode Reset. If that doesn't work, check reports/ for drift patterns and tighten rules with Mode Rule.
Q: What if I need a skill that isn't installed? A: The AI auto-builds a starter skill on the spot using the Skill Builder. For engineering-grade skills, browse codestacker.co or the CodeStacker-Skills community repo.
- Start every day with
Mode Start— refreshes AI context - Review
reports/weekly — spot recurring AI mistake patterns - Keep
structure/accurate — better structure = better AI performance - Finish clean — always end with
Mode Finishfor clean handoffs - Browse CodeStacker-Skills — don't reinvent skill files others have built
Contributions are welcome!
- 🐛 Report bugs — open an issue
- 💡 Suggest features — start a discussion
- 📝 Improve docs — fix typos, add examples
- 🔧 Submit PRs — follow existing conventions
- 🛠️ Share skills — contribute to CodeStacker-Skills
- 🛒 Publish on the marketplace — share prompts, skills & workflows at codestacker.co
MIT License — see LICENSE for details.
CodeStacker provides structure and context to AI models — it does not control them. You are responsible for reviewing all skills, prompts, Props, and Workflows before use. CoSama accepts no liability for AI model behavior or outcomes resulting from use of this software or any associated content, free or paid.
Stop vibe coding. Start CodeStacking. 🏗️
CodeStacker · CoSama · GitHub · Community Skills · Report Bug · Request Feature
🚦 Status: 🟡 v1.0.0 — Early development, actively evolving 🔮 Roadmap: IDE plugins · Skill auto-install · Analytics dashboard · More Props & Workflows
Part of the CodeStacker marketplace — AI prompts, skills & agentic workflows
Created by CoSama — AI Development Company