← All build notes

November 4, 2026 · 3 min read

Try AIWG with Flue

A short how-to: install AIWG, add its SDLC framework to the Flue Node example, and give the app a code review task. The agent gets AIWG skills, three AIWG...

This how-to puts AIWG inside a small agent app built with Flue. When you finish, the app has AIWG's SDLC skills, your project context in AGENTS.md, three AIWG tools and a Code Reviewer helper. You give it a change to review, and it saves the review in your project's .aiwg/ folder, where it stays with the project.

It takes a handful of commands. Everything up to the last step runs without a model key. For the bigger picture of what AIWG brings to Flue, see Bring AIWG into your Flue app.

Or ask your coding assistant

If you work with a coding assistant, you can hand it the whole job:

Add AIWG to my Flue app and give it a code reviewer. Use this Node example, keep conversation state across restarts, and save the review as an artifact.

The steps below are what that involves, so you can follow along or do it yourself.

What you need

  • Node 24 and npm.
  • A key for a model provider, for the last step only. The example uses openrouter/anthropic/claude-haiku-4.5 unless you set FLUE_MODEL.

1. Install AIWG

npm install -g [email protected]
aiwg --version
  aiwg  2026.11.0  [stable]

2. Get the Node example

The example is a small Flue app in the AIWG repository. This fetches the 2026.11.0 release and keeps only the example folder:

curl -fsSL https://github.com/jmagly/aiwg/archive/refs/tags/v2026.11.0.tar.gz \
  | tar -xz --strip-components=2 aiwg-2026.11.0/examples/flue-node
cd flue-node

It ships without any AIWG files on purpose. The next step adds them.

3. Add AIWG to the app

aiwg use sdlc --provider flue
    Agents 0  ·  Commands 0  ·  Skills 116  ·  Rules 0  ·  Behaviors 0

That one command writes everything the agent needs into the app:

What Where
The SDLC skills (116 of them, including sdlc-quickref) .agents/skills/
Your project context and core rules AGENTS.md
Tools to find an AIWG capability, read it with its rules, and save an artifact src/aiwg/tools.ts
Code Reviewer and Test Engineer src/aiwg/specialists.ts
Conversation state in SQLite src/db.ts, storing .flue-state/state.db

It also adds .flue-state/ to .gitignore and keeps AIWG's own records in .aiwg/. If the run lists an advisory about signed-release evidence, the deployment still succeeded, and the advisory names the command that clears it.

4. See what the agent will find

Before you add a model key, you can ask AIWG the same questions the agent will ask:

aiwg discover "sdlc quickref"
aiwg show skill sdlc-quickref
1. SDLC Framework — Quick Reference
   type: skill  score: 1.00
   name: sdlc-quickref

aiwg show lists the rule that comes with the skill, sdlc-orchestration, and an outline of the quick reference. During a run, the agent makes these lookups through its aiwg_discover and aiwg_show tools.

5. Install and build the app

npm ci
npm run build
npm start

Flue prints [flue] Server listening on http://localhost: followed by the port, which is 3000 unless you set PORT. For development, npm run dev starts Vite on http://localhost:5173/ instead.

6. Give it a review

This is the step that uses your model. Set your provider's key in your environment first; for the default OpenRouter model that is OPENROUTER_API_KEY. Then ask the app:

Review this change: export const add = (a, b) => a - b;. Discover the AIWG SDLC quickref, show its governing rules, activate it, ask Code Reviewer to assess the change, and save the review at reviews/example.md.

You can send that to the running app at /agents/assistant, or run one conversation straight from the terminal:

./node_modules/.bin/flue run src/agents/assistant.ts --id aiwg-review --message \
  'Review this change: export const add = (a, b) => a - b;. Discover sdlc-quickref, show its rules, activate it, delegate to code-reviewer, and write reviews/example.md with aiwg_artifact_write.'

The example's assistant is given a fixed path: it finds the SDLC quick reference, reads its rules, activates the skill, hands the change to Code Reviewer, and saves the review. Look for it at .aiwg/reviews/example.md.

Reuse the same --id for a follow-up. The conversation is stored in SQLite, so with the production build or flue run the example is set up to carry on after a restart.

Where to go next

  • Carry more than one framework. Declare each one with --declare, then run aiwg refresh. The launch article shows how.
  • Use the package instead of copied files. Install @aiwg/[email protected] and run aiwg use sdlc --provider flue --package. The generated modules then import skills from the package, and your agent can too, for example import sdlc from '@aiwg/flue/skills/sdlc-quickref/SKILL.md';.
  • Read the full guide. The Flue provider guide covers package mode, remote connections and Cloudflare Workers.

Try it on the example, then on your own Flue app, and tell us what your agent reviews first.