by nWave-ai
AI agents that guide you from idea to working code, with you in control at every step.
# Add to your Claude Code skills
git clone https://github.com/nWave-ai/nWaveLast scanned: 5/16/2026
{
"issues": [],
"status": "PASSED",
"scannedAt": "2026-05-16T06:21:46.156Z",
"semgrepRan": false,
"npmAuditRan": true,
"pipAuditRan": true
}nWave is an open-source ai agents skill for AI coding assistants such as Claude Code, Codex CLI, and ChatGPT, built by nWave-ai. AI agents that guide you from idea to working code, with you in control at every step. It has 573 GitHub stars.
Yes. nWave passed SkillsLLM's automated security scan — a dependency vulnerability audit plus prompt-injection heuristics — with no high-severity issues. You can read the full report in the Security Report section on this page.
Clone the repository with "git clone https://github.com/nWave-ai/nWave" and add it to your Claude Code skills directory (see the Installation section above).
nWave is primarily written in Python. It is open-source under nWave-ai on GitHub, so you can review or fork the full source.
Yes. SkillsLLM lists many other AI Agents skills you can browse and compare side by side. Open the AI Agents category from the badge at the top of this page, or use the Related Skills and comparison links further down to weigh nWave against similar tools.
No comments yet. Be the first to share your thoughts!
AI agents that guide you from idea to working code, with human judgment at every gate.
nWave runs inside Claude Code. It breaks feature delivery into seven waves (discover, diverge, discuss, design, devops, distill, deliver). Specialized agents produce artifacts at each wave. You review and approve before proceeding.
Documentation: docs.nwave.ai — guides, reference, and explanations, versioned per release.
Requirements: Python 3.10+ and Claude Code.
sh -c "$(curl -fsSL https://raw.githubusercontent.com/nWave-ai/nWave/main/scripts/install/install.sh)"
This installs the nwave-ai CLI and wires nWave into Claude Code in one step. It uses uv when available (recommended); pipx is supported but not recommended. Restart Claude Code when it finishes.
Need CLI flags, environment variables, CI/non-interactive use, or manual and offline steps? See the Installation Guide.
Inside Claude Code, type:
/nw-buddy What should I do next?
The buddy reads your project and tells you which wave to start, where your artifacts are, and how to use nWave for your specific context. It works on day one with no configuration.
Before nWave: "Where do I start? Requirements doc or code first? Which agent?" After nWave: The buddy reads your project and gives you a concrete next step.
Using pipx, OpenCode, or Codex instead? See the Installation Guide.
| Resource | What it covers |
|---|---|
| Your First Feature | End-to-end walkthrough, zero to working code |
| Team Rollout Guide | Onboard a second developer onto an nWave project |
| Offline / Air-Gapped Install | Install nWave on a machine without PyPI access |
| Jobs To Be Done | Which wave fits your task |
| Wave Directory Structure | How artifacts are organized per feature |
| Feature Delta Format (L7) | Author features in the lean single-file model |
| Outcomes Registry | Catch duplicate rules and operations at design time |
| Configuring Doc Density | Control lean vs full wave output |
| Agents and Commands Reference | All agents and commands |
| Troubleshooting | Common issues and fixes |
.nwave/local-config.json marker plus a global activation.mode (opt-in default, or all) decide where nWave runs. Manage it with nwave-ai project enable|disable, nwave-ai mode, and nwave-ai status. Existing projects auto-adopt on first /nw- use, so nothing breaks. See Activating nWave in a Project.nwave-ai CLI reference — the user-facing command surface (install, uninstall, doctor, status, project, mode, attribution, completion, version) is documented in one place, including the install flag pass-through to the underlying installer. See CLI Reference.See the full What's New in v3.19 for details.
feature-delta.md with schema-typed section headings (## Wave: <WAVE> / [REF|WHY|HOW] <name>). Tier-1 [REF] is auto-produced; Tier-2 [WHY] and [HOW] are opt-in via --expand. Downstream agents grep section headings instead of reading whole subdirectories. See Feature Delta Format (L7).nwave-ai validate-feature-delta <path> checks structural rules (E1–E5) and emits JSON for CI integration. Vendor-neutral: no hooks auto-installed; pick a recipe from Enforcement Recipes (12 platforms covered).nwave-ai outcomes register|check|check-delta flags spec-level collisions before code is written, via type-shape + keyword Jaccard. See Why an outcomes registry? and Your first outcome.lean vs full density controls how much each wave emits. Tune token cost per wave. See Configuring Doc Density.nwave-ai uninstall --force now removes all installed artifacts (skills/nw-*, lib/python/des/, all 5 DES hook event types in settings.json) while preserving user-created skills. Previous versions left ~197 skill dirs and 3 hook entries behind. See Troubleshooting → Uninstall left files behind.For upgrading from v3.3 or earlier, see Breaking Changes below.
machine human machine human machine
│ │ │ │ │
▼ ▼ ▼ ▼ ▼
Agent ──→ Documentation ──→ Review ──→ Decision ──→ Agent ──→ ...
generates artifacts validates approves continues
Each wave produces artifacts that you review before the next wave begins. The machine never runs unsupervised end-to-end.
The workflow has seven waves. Entry point depends on your context:
| Wave | Command | Agent | Entry? |
|---|---|---|---|
| DISCOVER | /nw-discover |
product-discoverer | Greenfield: explore market and problem space |
| DIVERGE | /nw-diverge |
nw-diverger | Greenfield: structured brainstorming before converging |
| DISCUSS | /nw-discuss |
product-owner | All: write requirements and user journeys |
| DESIGN | /nw-design |
system-designer, ddd-architect, solution-architect | All: architecture and domain model |
| DEVOPS | /nw-devops |
platform-architect | All: infrastructure and deployment |
| DISTILL | /nw-distill |
acceptance-designer | All: acceptance tests (Given-When-Then) |
| DELIVER | /nw-deliver |
software-crafter | All: TDD implementation |
Wave routing: Entry points vary by context:
DISTILL then DELIVER is always the terminal pair. See the Wave Routing Guide for the full decision matrix.
40 agents total: 10 wave agents (including 3 DESIGN specialists), 1 concierge, 8 cross-wave specialists, 14 peer reviewers, 7 business agents. Full list: Commands Reference
match/case statements and X | Y union type syntax introduced in Python 3.10. Verify with python3 --version.Run the one-liner under Install in 5 Minutes at the top of this page. Agents and commands go to ~/.claude/.
Don't have uv? Install with:
curl -LsSf https://astral.sh/uv/install.sh | shor see uv installation docs. Alternatively, use pipx (requires Python 3.10+):pip install pipx && pipx ensurepath. **Windows