nv-implement

Implements planned work via parallel subagents on isolated git worktrees with TDD and staged merges.

39.7k|4.4k|Updated Aug 26, 2021
One-click install
npx skills add https://github.com/novuhq/novu --skill nv-implement
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: nv-implement
Source: https://github.com/novuhq/novu/tree/main/.cursor/skills/nv-implement
Command: npx skills add https://github.com/novuhq/novu --skill nv-implement

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

Large implementation tasks in the Novu monorepo are slow and error-prone when done serially in a single checkout, and parallel edits often collide on the same files. This Skill coordinates the whole implementation: it cuts a plan into disjoint vertical slices, runs one subagent per slice on its own git worktree, and merges everything back with a single integration pass.

Core Features & Use Cases

  • Parallel slice execution: Splits a plan into seams touching disjoint files, then fans out subagents in parallel, each on an isolated worktree branched from the feature branch.
  • TDD at pre-agreed seams: Decides upfront which slices are TDD-able (use cases, resolvers) versus verify-by-lint (dashboard UI), and instructs each subagent accordingly.
  • Staged review and merge: Each slice commits and runs a park-and-review pass before merging back with --no-ff, followed by a cross-slice integration pass, a whole-diff code quality review, and one full test suite run.
  • Use Case: Given a feature plan touching the Novu API and dashboard, invoke this Skill to implement backend policy, migrations, and UI copy in parallel worktrees, then merge and validate the combined diff.

Quick Start

Ask the agent to implement the current plan or ticket using nv-implement, splitting the work into parallel slices on separate worktrees.

Frequently Asked Questions about nv-implement

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I implement a large feature plan with parallel agents?

Cut the plan into vertical slices that touch disjoint files, then launch one subagent per slice on its own git worktree in a single parallel batch. Each subagent implements, tests, commits, and reviews its slice before the slices are merged back into the feature branch in dependency order.

How do I use git worktrees to run parallel coding agents?

Create one worktree per slice with git worktree add -b <feature-branch>-<seam> ../wt-<seam> HEAD, and symlink node_modules into each worktree to avoid reinstalling. Each agent works only inside its own worktree so sibling slices never corrupt each other.

When should a slice use TDD versus lint-only verification?

Slices with public functions, use cases, or resolvers are TDD-able and should follow a red-green loop one behavior at a time. Dashboard UI slices without a unit test harness are verified via lint checks instead, since inventing a test harness is out of scope.

What happens if two slices need to edit the same file?

The two slices are merged into a single slice rather than parallelized, because concurrent edits to the same file guarantee a conflict. The seam-cutting step is done when every plan item is owned by exactly one seam with no shared files.

Why run the full test suite only once at the end?

Slices run typechecks and individual spec files regularly during development, which catches most issues cheaply. The full suite runs once after merging and the integration pass, since cross-slice fallout is only visible on the combined diff and repeated full runs are slow.

Can slices merge directly into the next or main branch?

No. Slices always merge into the feature branch, either the existing one detected at startup or a new one created from origin/next. Merging slices directly into next is explicitly prohibited by the guardrails.