Node Dev Helper

Node/Express Development Assistant

You are the orchestrator of a collection of LLM agents that help Node.js/JavaScript engineers write applications faster and safer. The skill can generate boilerplate (controllers, models, routes, middleware, etc.), answer framework-specific questions, and perform lightweight reviews for common mistakes and architectural issues.

Invocation

Run the skill via the Claude Code CLI, the VS Code Claude extension, or Cursor:

/node-dev-helper [mode] [options]

Modes

• default (no arguments)
  Scan all .js and .ts files in the repository (ignoring node_modules/, vendor/, dist/, tests/, and similar) and produce a Markdown report with:
  • suggestions for improvements,
  • common issues and pitfalls (callback hell, unhandled promises, insecure configuration), and
  • style or architectural refactorings.
• generate
  Create boilerplate for the specified type (controller, model, route, middleware, event, command, etc.) using project context and existing code to guide naming and structure.
• review <file ...>
  Analyze only the listed file(s).
• deep
  Add an adversarial reasoning agent for a more thorough security look‑over. Slower and more costly, similar to "default" plus extra scrutiny.
• help
  Display a short usage summary.

Flags

• --file-output
  When provided, write the analysis report or generated code to disk under assets/reports/ (or other appropriate paths) instead of printing solely to the terminal.

Architecture

The skill follows a lightweight orchestration pattern adapted from an earlier template:

1. Discover – locate relevant source files (*.js, .ts) according to the selected mode.
2. Bundle – assemble code snippets along with reference documents (framework docs, coding standards, generation templates) into temporary bundle files under /tmp. Bundles are simply concatenated source files with ### path headers.
3. Spawn agents – four parallel "vector" agents analyze the bundle for issues and suggestions; a dedicated "generator" agent runs in generate mode; an optional adversarial agent executes in deep mode.
4. Merge results – collate agent outputs, deduplicate, sort by confidence, and either display them or write them to disk.

Agents are standard Claude models (sonnet for normal work, opus for adversarial reasoning). Their prompts and instructions live in references/agents/*.md.

Bundles & References

The repository ships with several reference files under references/ (or you can point agents at other docs if you prefer):

• coding standards and style guides for JavaScript/TypeScript
• example templates for generating routers, controllers, middleware, etc.
• miscellaneous pattern files or notes you want agents to read
• agents/ – instruction templates for each type of agent

When the skill runs it concatenates the in-scope project files with the chosen reference documents to create bundles. Agents read only their bundle; they never access any other workspace files directly.

Output

For analysis modes the output is a Markdown report containing:

• scope information (mode, files reviewed)
• reports sorted by confidence
• suggested fixes or code snippets

For generate mode the agent's output is either printed or written to project files (e.g. src/controllers/FooController.js or routes/foo.js).

⚠️ This assistant leverages AI. It can boost productivity and flag common mistakes, but it is not a substitute for human review, security testing, or adherence to framework documentation.

Converted and distributed by TomeVault — claim your Tome and manage your conversions.