mcpbeat

Nw New

nwave-ai/nw-new

Guided wizard to start a new feature. Asks what you want to build, recommends the right starting wave, and launches it.

1k tokens
context cost
the whole folder, loaded on every use
1
files
instructions only
0
copies elsewhere
how many repositories repackaged it
588
stars on the repo
on the repository, not the skill itself

Install

one command, takes just this skill from the repository
npx skills add https://github.com/nWave-ai/nWave --skill nw-new

What it tells the agent to use

found in the instruction text
Task spawns other agents

The instruction itself

17 sections, as written by the author

NW-NEW: Start a New Feature

Wave: CROSS_WAVE (entry point)

Agent: Main Instance (self — wizard)

Command: /nw-new

Overview

Conversational wizard that asks the user to describe their feature|classifies it|recommends a starting wave|launches it. Eliminates need to understand the 6-wave pipeline before using nWave.

You (the main Claude instance) run this wizard directly. No subagent delegation.

Behavior Flow

Step 1: Feature Description Intake

Ask the user to describe what they want to build. If provided as argument, use that.

If vague (fewer than 3 meaningful words or unclassifiable):

  • Ask: What system is being changed?
  • Ask: What problem are you solving?
  • Ask: Who benefits from this change?
  • Provide example: "Add rate limiting to prevent API abuse"

Do NOT proceed until you have a clear, actionable description.

Step 2: Feature ID Derivation

Derive feature ID per rules in ~/.claude/nWave/skills/common/wizard-shared-rules.md (section: Feature ID Derivation).

Examples: "Add rate limiting to the API gateway" -> rate-limiting-api-gateway | "OAuth2 upgrade" -> oauth2-upgrade | "Implement a real-time notification system with WebSocket support for mobile and desktop clients" -> real-time-notification-system-websocket

Show derived ID via AskUserQuestion. Allow override with custom value.

Step 3: Name Conflict Check

Check if docs/feature/{feature-id}/ exists. If so, offer via AskUserQuestion:

  • Continue that project — switch to /nw-continue
  • Start fresh with different name — ask for distinguishing name
  • Archive and restart — move to docs/feature/{feature-id}-archived-{date}/

Step 4: Clarifying Questions

Use AskUserQuestion:

Q1: New or existing behavior?

  • New functionality that doesn't exist yet
  • Changing/improving existing functionality
  • Fixing a bug or regression

Q2: Requirements readiness?

  • Clear requirements in my head but nothing written
  • Rough idea, need to explore further
  • Haven't validated whether problem is worth solving

Step 5: Feature Classification

Based on description and answers, classify:

  • User-facing — UI/UX visible to end users
  • Backend — APIs, services, data processing
  • Infrastructure — DevOps, CI/CD, tooling
  • Cross-cutting — Spans multiple layers (auth, logging, etc.)

Show classification for user confirmation.

Step 6: Greenfield vs Brownfield Detection

Scan filesystem:

  • Check if src/ or equivalent has code
  • Check if docs/feature/ has prior feature directories
  • Check for existing test directories

No source code and no prior features -> greenfield | Otherwise -> brownfield

Step 7: Wave Recommendation

Decision tree:

IF "fixing a bug":
    -> /nw-root-why ("Investigate the root cause first")
IF "haven't validated the problem":
    -> /nw-discover ("Validate the problem space before building")
IF "rough idea, need to explore":
    -> /nw-discuss ("Define requirements and acceptance criteria")
IF "clear requirements, nothing written":
    -> /nw-discuss ("Formalize requirements into user stories")
IF existing DISCUSS artifacts found:
    -> /nw-design ("Requirements exist, design the architecture")
IF existing DESIGN artifacts found:
    -> /nw-distill ("Architecture exists, create acceptance tests")
IF all prior waves complete:
    -> /nw-deliver ("Ready for implementation")
DEFAULT:
    -> /nw-discuss ("Start by defining what to build")

Show recommendation with rationale via AskUserQuestion: recommended wave command|why this wave (one sentence)|what it produces.

Step 8: Launch

After user confirms, create project directory:

mkdir -p docs/feature/{feature-id}

Invoke recommended wave command by reading its task file and following instructions, passing feature ID as argument.

Error Handling

| Error | Response |

|-------|----------|

| Vague description (< 3 meaningful words) | Ask follow-up questions with example |

| Name conflict with existing project | Offer continue/rename/archive options |

| User cannot classify feature type | Default to "cross-cutting", note uncertainty |

| No clear wave recommendation | Default to DISCUSS with explanation |

Success Criteria

  • [ ] User described feature in plain language
  • [ ] Project ID derived and confirmed by user
  • [ ] No name conflicts (or resolved)
  • [ ] Feature classified by type
  • [ ] Starting wave recommended with rationale
  • [ ] User confirmed recommendation
  • [ ] Wave command launched with correct feature ID

Examples

Example 1: Greenfield backend feature

/nw-new "Add rate limiting to the API gateway"

Derives rate-limiting-api-gateway, detects no prior artifacts (greenfield), asks clarifying questions. User says "new functionality, clear requirements." Recommends DISCUSS, launches /nw-discuss "rate-limiting-api-gateway".

Example 2: Bug fix

/nw-new "Fix authentication timeout errors"

Detects "fix" in description. User confirms bug. Recommends /nw-root-why "authentication timeout errors".

Example 3: Unclear problem

/nw-new "Build a customer feedback system"

User says they haven't validated whether customers want this. Recommends DISCOVER, launches /nw-discover "customer-feedback-system".

How to use it

Copy the folder

Take nwave-ai/nw-new from the repository into ~/.claude/skills for personal use, or into .claude/skills inside a project.

Check the name does not clash

The agent identifies a skill by the name field in its header. Two skills with the same name cannot sit side by side — one of them will be ignored.