L1DDER — Spec-Driven Development Framework for AI Coding Agents
🌍 Documentation Site (with visual diagrams): Русская версия | English Version
L1DDER (pronounced "leader" or "l'oneder") is a spec-driven development framework for AI coding agents: Cursor, Claude Code, OpenAI Codex, Windsurf, GitHub Copilot, OpenCode, and Cline. Ideas, specs, roadmap and tasks live in plain Markdown, /l1-* slash commands drive the work, and nothing moves forward without your "ok".
- Transparent Markdown Control: Your Markdown documents in
docs/ (ideas.md, backlog.md, roadmap.md) with end-to-end linked ID Passports are the single source of truth.
- Slash Commands (
/l1-*): Control the entire product lifecycle with /l1-onboard, /l1-idea, /l1-feature, /l1-planning, /l1-task, /l1-hotfix, /l1-chore, and /l1-fast + "ok" approvals.
- Mandatory Sprint Doctor (
TASK-NN.99): Every sprint ends with a 4-block Doctor Checklist (CI, 2-axis code review, Agent E2E verification, mandatory Living Docs Sync, and Human UAT).
- Built-in Engineering Discipline: Ships with DietrichGebert/ponytail (Lazy Senior Dev mode,
shortcut: ledger, 2-line Skipped / Risk footer) and mattpocock/skills (grilling, domain-modeling, codebase-design Deep Modules, vertical tdd at seams, diagnosing-bugs).
⚡ Quick Start (Install via AI Agent or Terminal)
🤖 Option 1: Ask Your AI Agent to Install It for You (Zero Terminal)
Copy and paste this single line into Cursor, Claude Code, OpenAI Codex, Windsurf, or Copilot chat:
Русская версия:
Установи в этот проект фреймворк L1DDER командой `npx l1dder@latest init --lang ru --yes`, а затем сразу начни онбординг по `l1dder/skills/l1-onboard/SKILL.md`
English version:
Install the L1DDER framework in this project by running `npx l1dder@latest init --lang en --yes`, then immediately start onboarding from `l1dder/skills/l1-onboard/SKILL.md`
💻 Option 2: Install via Terminal (1 Second)
# Interactive setup (selects language ru/en and your AI coding agents):
npx l1dder init
# Or non-interactive one-liner (defaults to Cursor, Claude Code & Codex):
npx l1dder init --lang ru --yes
Then open your AI agent chat and run /l1-onboard.
🤖 Supported AI Agents (--agents)
| Agent | Flag | Configured Files |
|---|
| Cursor | cursor | .cursor/rules/l1dder/*.mdc + .cursor/skills/* + AGENTS.md |
| Claude Code | claude | CLAUDE.md + .claude/rules/l1dder/*.md + .claude/skills/* |
| OpenAI Codex | codex | AGENTS.md + .agents/skills/* |
| Windsurf | windsurf | .windsurf/rules/l1dder.md + .windsurf/workflows/* + AGENTS.md |
| GitHub Copilot | copilot | .github/copilot-instructions.md + .github/prompts/*.prompt.md |
| OpenCode | opencode | AGENTS.md + .agents/skills/* |
| Cline / Roo Code | cline | .clinerules/l1dder.md + .roo/rules/l1dder.md + AGENTS.md |
| All Agents | all | Installs for all supported agents at once |
l1dder/skills/ is the single source of truth: npx l1dder sync copies it into the skills folders of the selected agents, so edit skills there rather than in agent folders. Cursor always reads .agents/skills/ and reads .claude/skills/ only with Third-Party Imports enabled, so cursor + claude installs both; .cursor/skills/ is used only when Cursor is the only agent with a skills folder.
🎮 Slash Commands
| Command | Flow | What It Does |
|---|
/l1-onboard | Onboarding | Works out the concept stage by stage, then configures 30-tech-stack.mdc & Makefile and initializes docs/. |
/l1-idea | Idea Bank | Refines an idea (grilling) and records IDEA-NNN in docs/ideas.md. |
/l1-feature | Feature Spec | Creates docs/specs/FEAT-NNN-<slug>/, slices FEAT-NNN.01..99, updates docs/backlog.md. |
/l1-planning | Roadmap | Updates docs/roadmap.md (Delivery Queue & Discovery Queue), creates SPRINT-NN, or rejects the current sprint with rollback (git revert). |
/l1-task | Task & Doctor | Executes the next task from docs/roadmap.md (Two-Approval Rule) and mandatory TASK-NN.99 (Doctor Checklist). |
/l1-hotfix | Hotfix | Reproduces a bug (RED → GREEN), applies minimal fix (ponytail), logs HOTFIX-NNN in docs/backlog.md. |
/l1-chore | Chore | Changes to the project itself, not the product (git rules, CI, libraries, Makefile, agent rules): a single "ok" stop, make ci, logs CHORE-NNN in docs/backlog.md. |
/l1-fast | Fast Mode | Builds a small product feature when no sprint is active (strictly human-invoked, forbidden for the agent on its own): two "ok" stops, full verification before acceptance, records the full document chain marked mode: /l1-fast after the code. |
🛠 CLI Commands & Skills Manifest (l1dder/skills.json)
All skills are managed via l1dder/skills.json (like package.json for AI Agent Skills):
- Bundled Skills Out-of-the-Box:
/l1-* workflow skills, DietrichGebert/ponytail skills, and mattpocock/skills engineering skills (grilling, domain-modeling, codebase-design, tdd, diagnosing-bugs, code-review, retro, prototype, writing-for-agents).
- Project Stack Skills (
stack): configured during /l1-onboard or added via CLI.
# Install in a new or existing project (defaults to Cursor, Claude Code & Codex):
npx l1dder init [--lang ru|en] [--agents cursor,claude,codex] [--yes]
# Update the ENTIRE framework when a new version of l1dder comes out (preserves 30-tech-stack.mdc & stack skills):
npx l1dder@latest update
# Update ONLY skills (leaves rules, templates, and docs untouched):
npx l1dder@latest update --skills
# Manage skills in l1dder/skills.json:
npx l1dder skills list
npx l1dder skills add <owner/repo> --skill <skill1,skill2>
npx l1dder skills install
npx l1dder doctor
License
MIT. See LICENSE.
