Why SevenD? AI agents write code fast — SevenD makes sure they build the right thing, in the right order, and know when to stop and ask.
A progressive framework for building software with AI coding agents. Structure your project so AI agents know what to read, what to build, and when to stop and ask.
AI coding agents produce better output when they have structured context. Without it, they invent component names, skip tests, build features nobody asked for, and lose track of what's been decided. SevenD gives your project a skeleton that any AI agent can follow.
- Sprint-based INDEX files -- Every phase folder uses an INDEX.md linking to sprint files. Work is non-linear; the structure matches.
- Build + Bug sections -- Every sprint has both. Sprints close only at zero bugs.
- reference/ directory -- Stable documents (architecture, components, contracts, decisions, agent rules) live separately from active work.
- STATUS.md -- Root-level dashboard. One glance to know where the project stands.
- Sprint archiving -- Closed sprints move to
reference/archive/to keep working directories clean. - Sprint reports -- Built into every sprint file. Metrics and learnings captured at close.
Pick the level that matches your project. You can always upgrade.
| Level | Files | For | What you get |
|---|---|---|---|
| 1 | 2 files | Solo, weekend, prototype | Product.md + Tech.md |
| 2 | 4 files | Small team, real project | + Architecture.md, Resources.md, Project.md |
| 3 | 7 folders + reference/ | Complex project, weekly sprints | Phase INDEX + sprint files, reference docs, STATUS.md |
| 4 | GitHub-native + reference/ | Team on GitHub, CI/CD | Issues, Actions, PR gates + stable reference docs |
Every level includes IDE rules for Claude Code, Cursor, Windsurf, and VS Code Copilot.
# Clone the repo
git clone https://github.com/Ashvinnaik/SevenD.git
# Copy the level you need into your project
cp -r SevenD/level-1/ my-project/ # Solo / prototype
cp -r SevenD/level-2/ my-project/ # Small team
cp -r SevenD/level-3/ my-project/ # Complex project
cp -r SevenD/level-4/ my-project/ # GitHub-native
# Fill in the templates with your project details
# Replace [placeholders] with real contentThe skill walks you through assessment, picks the right level, and generates files filled in with your actual project details instead of placeholders.
Claude Code (CLI):
# Install the skill
cp -r SevenD/skill/ ~/.claude/skills/SevenD/
# Start a conversation
claude
# The skill triggers automatically when you say:
# "set up my project" / "add a feature" / "sprint review"Claude.ai (Projects):
- Create a new Project in Claude.ai
- Paste the contents of
skill/project-instructions.mdinto Custom Instructions - Upload these 6 files to Project Knowledge:
skill/SKILL.mdskill/7Dkb.mdskill/templates.mdskill/levels-overview.mdskill/ide-rules-format.mdskill/upgrade-paths.md
- Start a conversation: "Help me set up my project"
| Phase | Loop | What happens |
|---|---|---|
| Discovery | Product | Identify what to build and why |
| Definition | Product | Specify requirements and acceptance criteria |
| Design | Product | Architect components, interfaces, standards |
| Documentation | Bridge | Track status across all phases |
| Development | Tech | Build it in time-boxed sprints |
| Diagnostics | Tech | Verify correctness and quality |
| Deployment | Tech | Ship, monitor, learn |
Phase gates prevent premature advancement: no Design without approved Definition, no Development without completed Design, no Deployment without Diagnostics passing.
project/
├── STATUS.md # Current state at a glance
├── reference/ # Stable docs (change slowly)
│ ├── ARCHITECTURE.md # System architecture
│ ├── COMPONENTS.md # Component registry
│ ├── CONTRACTS.md # Interface contracts
│ ├── DECISIONS.md # Decision log
│ ├── AGENT.md # AI agent rules
│ └── archive/ # Closed sprints
├── 01-discovery/
│ ├── INDEX.md # Phase index + sprint links
│ └── D1-Sprint1.md # Sprint with Build + Bug sections
├── 02-definition/
│ ├── INDEX.md
│ └── D2-Sprint1.md
├── 03-design/
│ ├── INDEX.md
│ └── D3-Sprint1.md
├── 04-documentation/
│ ├── INDEX.md # Product status board
│ └── D4-Sprint1.md
├── 05-development/
│ ├── INDEX.md
│ └── D5-Sprint1.md
├── 06-diagnostics/
│ ├── INDEX.md
│ └── D6-Sprint1.md
├── 07-deployment/
│ ├── INDEX.md
│ └── D7-Sprint1.md
└── CLAUDE.md / .cursor/ etc. # IDE rules
The SevenD skill covers the full lifecycle:
- Initial setup -- Assess your project, recommend a level, generate customized files
- Feature workflow -- Walk through all 7 phases for your level, pointing to exact files
- Sprint operations -- Start sprints, run reviews, close at zero bugs, archive
- Level upgrades -- Migrate from one level to the next without losing content
- Agent debugging -- Diagnose why your AI agent isn't following the framework
SevenD/
├── level-1/ # Two files (Product.md + Tech.md + IDE rules)
├── level-2/ # Four files (+ Architecture, Resources, Project + IDE rules)
├── level-3/ # Phase folders + reference/ + STATUS.md + IDE rules
├── level-4/ # GitHub-native + reference/ + STATUS.md + IDE rules
├── skill/ # Claude skill for guided setup and operations
│ ├── SKILL.md # Operating instructions
│ ├── 7Dkb.md # Methodology, vocabulary, conventions
│ ├── templates.md # File structures for all levels
│ ├── levels-overview.md # Decision matrix and file inventories
│ ├── ide-rules-format.md # IDE rule file format specs
│ ├── upgrade-paths.md # Level migration guides
│ └── project-instructions.md # Claude.ai project custom instructions
└── README.md
MIT

