5 machine-readable skills that make AI coding agents write smaller, more powerful code.
Skills · Research · Quick Start · MIT License
(workflows/skills i use and constantly updating)
This is my first — very symbolic — public release.
Six months. Thousands of hours. 20+ systems. ~50 prototypes. Hundreds of ideas. FewMillion+ of lines of code. Billions of tokens. Zero ships.
Perfectionism+adhd - has some science behind :) strong ideas - momentum loss when execution gets slow or repetitive. I'm not sure where 6 months went. Most of my projects get abandoned at 60–70% . Overwhealming new revolutionary tool every day.
My bot friends sugested to ship at 20%
A grain of sand in the desert storm -
A collection of agent skills — portable instruction sets that AI coding agents load on demand to become better at specific tasks. They follow the SKILL.md open standard and work across:
- Claude Code / Claude.ai
- Windsurf (Cascade)
- Cursor
- Codex CLI
- Any tool that reads SKILL.md files
The skills are opinionated toward lean codebases: fewer files, fewer dependencies, more capability per line. Every pattern in here was chosen because it makes code smaller without making it weaker.
| Skill | What It Does | Lines |
|---|---|---|
| implement | The brain — reads your project structure, detects tech stack, picks the best strategy, implements for minimal code | 131 |
| lean-typescript | TypeScript 2026 anti-bloat rules, modern toolchain, architectural patterns | 144 + refs |
| spec-driven | Contract-first: Spec → Schema → Registry → Module | 134 |
| debug-loop | Autonomous debugging: reproduce → isolate → fix → prove with evidence | 129 |
| roadmap | Planning with machine-readable YAML schemas for sprints, milestones, priorities | 130 + refs |
Level 1: Metadata ~100 tokens Always in context (name + description)
Level 2: SKILL.md body <500 lines Loaded when the skill triggers
Level 3: references/ Unlimited Loaded on demand for deep dives
implement (meta-skill)
├── reads your project first
├── selects the right skill:
│ ├── lean-typescript → for any TypeScript work
│ ├── spec-driven → for new modules or APIs
│ ├── debug-loop → for bugs and failures
│ └── roadmap → for planning tasks
└── verifies: did the codebase get smaller?
git clone https://github.com/CREOAUREAdevlab/agentic-skills.gitThen tell your AI coding agent:
"Read the skills in
skills/implement/SKILL.mdand use them when implementing features."
Each skill is a self-contained folder. Copy any skills/<name>/ folder into your project or your agent's skill directory.
Point your agent config at this repo. Most SKILL.md-compatible tools support loading skills from a path.
The research/ folder contains the curated research, benchmarks, and harvested patterns that informed these skills. Six months of investigating what actually works in 2025–2026 TypeScript development:
| Document | Contents |
|---|---|
| TypeScript 2026 Stack | Library choices with bundle sizes, from a 381-byte event bus to mutation testing |
| TypeScript Patterns | Anti-bloat commandments, architectural patterns, validation strategies |
| Schema-Driven Dev | The 4-pillar methodology for contract-first, agent-safe development |
| Roadmap Planning | YAML roadmap schemas, RICE/WSJF prioritization, sprint workflows |
| SSRIV Method |
Proposed 5-layer architecture pattern for AI-assisted development. Unproven at scale — a thinking tool, not a guarantee |
This is the "show your work" section — the data behind the recipes.
agentic-skills/
├── skills/
│ ├── implement/ # Agentic meta-skill
│ │ └── SKILL.md
│ ├── lean-typescript/ # TypeScript 2026 best practices
│ │ ├── SKILL.md
│ │ └── references/
│ │ ├── toolchain.md
│ │ ├── patterns.md
│ │ └── debloating.md
│ ├── spec-driven/ # Contract-first methodology
│ │ └── SKILL.md
│ ├── debug-loop/ # Autonomous debug workflow
│ │ └── SKILL.md
│ └── roadmap/ # Planning & roadmap management
│ ├── SKILL.md
│ └── references/
│ └── schemas.md
├── research/ # Source research & benchmarks
│ ├── README.md
│ ├── typescript-2026-stack.md
│ ├── typescript-patterns.md
│ ├── schema-driven-dev.md
│ ├── roadmap-planning.md
│ └── ssriv-method.md # ⚠️ experimental proposal
├── .gitignore
├── CONTRIBUTING.md
├── LICENSE # MIT
└── README.md # You are here
- Concise — only include what the agent doesn't already know
- Progressive disclosure — metadata → body → references (loaded on demand)
- Explain the why — reasoning over rigid rules (LLMs are smart)
- Agent/IDE agnostic — standard format, no vendor lock-in
- Ship > perfect — done beats theoretical
See CONTRIBUTING.md. PRs welcome — especially new skills, pattern improvements, or research additions.
MIT — use it, fork it, ship it.
This repo is one piece of a larger ecosystem I've been building — an AI-assisted development platform. Here's the map of what exists behind this first ship:
OPEN-CLAW-BRIDGE Workspace
==========================
┌─────────────────────────────────────────────────┐
│ AI WORKFLOW LAYER │
│ │
│ ┌─────────────────┐ ┌─────────────────────┐ │
│ │ claude-flow- │ │ NODES │ │
│ │ editor │ │ ADHD-focused node │ │
│ │ 18 node types │ │ editor, 12 types │ │
│ │ Multi-AI exec │ │ XYFlow + Zustand │ │
│ └────────┬────────┘ └──────────┬──────────┘ │
│ │ │ │
│ └───────────┬───────────┘ │
│ │ │
├───────────────────────┼─────────────────────────┤
│ BRIDGE LAYER │
│ │ │
│ ┌────────────────────┴──────────────────────┐ │
│ │ BRIDGE — WebSocket relay │ │
│ │ Spawns AI CLI sessions, NDJSON streaming │ │
│ └────────────────────┬──────────────────────┘ │
│ │ │
├───────────────────────┼─────────────────────────┤
│ KNOWLEDGE LAYER │
│ │ │
│ ┌──────────────┐ ┌──┴──────────────────────┐ │
│ │ OpenClaw │ │ 00-indexing │ │
│ │ Driver │ │ Qdrant + Ollama │ │
│ │ 95 handles │ │ Vector code search │ │
│ │ 18 categories │ │ MCP server │ │
│ └──────────────┘ └─────────────────────────┘ │
│ │
├─────────────────────────────────────────────────┤
│ SKILLS & METHODS │
│ │
│ ┌──────────────────────────────────────┐ │
│ │ agentic-skills (THIS REPO) ← YOU ARE HERE│
│ │ 5 skills + research │ │
│ └──────────────────────────────────────┘ │
│ │
│ ┌──────────────┐ ┌────────────────────────┐ │
│ │ context-init │ │ OpenClaw Control UI │ │
│ │ CLI │ │ Lit web components │ │
│ │ .context │ │ Chat, channels, │ │
│ │ scaffolder │ │ gateway │ │
│ └──────────────┘ └────────────────────────┘ │
│ │
│ ┌──────────────────────────────────────┐ │
│ │ Agent Orchestrator (Composio fork) │ │
│ │ Parallel AI agents in git worktrees │ │
│ └──────────────────────────────────────┘ │
└─────────────────────────────────────────────────┘
+ ~30 prototypes, experiments, and backups
-
The vision is coherent. Every project connects to one goal: making AI agents better at writing code. Workflow editors, bridges, knowledge bases, skills, context scaffolding — they form a real pipeline from "human intent" to "agent-executed code."
-
The layering makes sense. Skills (methodology) → Knowledge (search/retrieval) → Bridge (communication) → Editors (UI). Each layer has a clear job and they compose naturally.
-
Too many prototypes, not enough products. 20+ systems with 50+ prototypes means energy is spread thin. The pattern: start strong, explore deep, stall before shipping. This first release proves the cycle can break.
-
The ADHD-aware design is genuine innovation. The NODES editor with dopamine anchors, fear scanners, and energy matchers isn't just novelty — it's applying neuroscience research to dev tooling. That's worth shipping on its own.
-
The skills system is the right first ship. It's self-contained, immediately useful, requires no infrastructure, and showcases the research depth. Other projects need the bridge, the UI, the knowledge base — this one stands alone.
Built by @CREOAUREAdevlab — first ship, not last.