Skip to content
ai-standards

Quickstart

Five minutes from “I cloned the repo” to “I can run /build-plan on a feature”.

Prerequisites

  • Claude Code installed and signed in.
  • Docker + docker compose (every service runs in a container).
  • Git, Node.js ≥ 20, npm.
  • A workspace directory where you can create new repositories.

1. Clone the framework

Open a terminal and create a new workspace folder:

Terminal window
mkdir my-workspace && cd my-workspace
git clone https://github.com/Silfaran/ai-standards.git

2. Install the slash commands and skills

Copy the slash-command stubs and skills into the workspace’s .claude/ directory so Claude Code picks them up:

Terminal window
cp -r ai-standards/.claude/commands .claude/commands
cp -r ai-standards/.claude/skills   .claude/skills

The framework ships 6 slash commands (/init-project, /create-specs, /refine-specs, /build-plan, /update-specs, /check-web) and ~16 on-demand skills.

3. Run /init-project

Open Claude Code in my-workspace/ and run:

/init-project

You will be asked for:

  • The project name (e.g. red-profesionales).
  • The list of services that make up the project (e.g. identity-service, api-gateway, web-front).
  • Whether to enable Playwright MCP (for the /check-web audit command).

The command scaffolds {project-name}-docs/ (your project’s docs repo, where specs / decisions / lessons-learned live), installs the agent model-tier PreToolUse hook into .claude/settings.json, and writes the .workspace-config-path pointer that lets every agent locate the project docs.

4. Your first feature

Once /init-project finishes, create a spec and run the pipeline:

/create-specs

You describe the feature in natural language; the Spec Analyzer writes a business spec under {project-name}-docs/specs/{Aggregate}/{feature-name}-specs.md. Review it, edit if needed, then run:

/refine-specs

This reads your codebase and turns the business spec into a technical spec + execution plan + task checklist. Review again, then run:

/build-plan

This is the orchestrator. It generates per-phase bundles, spawns the agents in sequence (DevOps if needed, then Backend Developer ‖ Frontend Developer in parallel, then DoD-checker, then Reviewer loop, then Tester), reports the per-phase token cost, and asks you to confirm the merge into master.

See the your-first-feature walkthrough for the full narrative with example outputs.

5. (Optional) Install Playwright MCP for /check-web

If you want manual browser audits of your running UI, set up Playwright MCP per the USAGE guide. The /check-web command then walks your deployed UI, captures findings, and produces paste-ready /create-specs prompts.

What’s next