asdt

README

AI Software Delivery Team

Ship features the way a real team would — architect, developer, QA, security, UX — even when you're building alone.

ASDT installs a team of AI specialists into Claude Code or OpenCode. Each specialist owns a distinct discipline and hands its work to the next through structured artifacts — the architect's decision informs the developer's plan, the developer's code informs the QA's tests. Without ASDT, AI assistants have no enforced discipline around how a feature gets designed, built, or reviewed.

How It Works

flowchart TD
    req([Feature request]) --> asdt
    asdt["/asdt\nroutes — never executes"]
    asdt -->|user confirms| specialists
    specialists["Specialists run in suggested order\n/asdt-ux-ui · /asdt-architect · /asdt-developer · …"]
    specialists -->|each launches isolated steps| steps
    steps["Step sub-agents\none artifact per step"]
    steps -->|saved to| engram[(Engram\npersistent memory)]
    engram -->|next specialist reads automatically| specialists

Each specialist orchestrates its own isolated steps. Steps that produce artifacts run as separate sub-agents so they don't pollute each other's context. Artifacts are saved to Engram — not to files on disk — so specialists can run minutes or days apart and still pick up where the last one left off.

Requirements

• Claude Code (claude) or OpenCode (opencode) installed
• Engram MCP server installed and running
• Go 1.22+ to install from source

Installation

curl -fsSL https://raw.githubusercontent.com/vitualizz/asdt/main/install.sh | bash

Downloads the pre-built binary for your platform (Linux/macOS, x86_64/arm64) and installs it to ~/.local/bin/. No Go required.

Getting Started

1. Install skills into your AI assistant

asdt-tui

Interactive TUI that checks Engram is installed, lets you choose which AI assistant(s) to target, and copies the ASDT skills into them. Each specialist is installed as its own top-level directory — e.g. ~/.claude/skills/asdt-architect/ — so each is independently invocable.

2. Initialize your project

Open your AI assistant in the project directory and run:

/asdt-init

The assistant will detect your project stack, ask a few configuration questions, and write .asdt/config.yaml and .asdt/knowledge/platform.yaml.

3. Start Engram, then use the specialists

ASDT uses Engram as its memory layer. Each specialist saves its artifacts there, and the next specialist retrieves them by change name. Engram must be running before you invoke any specialist — artifacts from one run must survive until the next.

Follow the Engram setup guide to install and start the MCP server, then configure it in your assistant's MCP settings.

Using the Specialists

Invoke from inside your AI assistant:

Command	What it does	Produces
/asdt	Meta-orchestrator — analyzes your request, recommends which specialists to run and in what order	Routing suggestion
/asdt-init	Initialize ASDT for the project — detects stack, writes config	.asdt/config.yaml, platform.yaml
/asdt-architect	Architecture decisions, system design, risk analysis	Architecture Decision Record + system design
/asdt-developer	Implementation plan with code and test snippets	Step-by-step implementation plan
/asdt-qa	Test plan and acceptance criteria	Test cases, quality report
/asdt-security	Threat model and hardening checklist	Security findings, hardening checklist
/asdt-ux-ui	User flows, component specs, responsive strategy	UX brief, component spec

Each specialist reads prior artifacts from Engram automatically — the developer finds the architect's decisions, the QA finds the developer's plan.

Start with /asdt if you are unsure which specialist to invoke.