Software Factory

My Attempt to build a software factory using fundamental components as much as possiible so that the setup is harness agnostic.

Factory Lifecycle

graph TD;
    A[PRD]-->B[Notion MCP]
    C[Issue] -->D[Github CLI]
    B --> E[Coding Agent]
    D --> E
    E --> F[Brainstorming]
    F --> G[Spec File TODOs]
    G --> H[Implementation]
    H --> J[Unit Tests]
    J --> I[Evidencing using agent-browser]
    I --> K[Add PR to stack]
    K --> G
    K --> L[Push Stacked PRs to GitHub]
    L --> M[Code Review]

Loading

Tooling

• Runtimes - NodeJS, JDK, Miniconda
• PRD source - Notion via MCP
• Issue Tracker - GitHub Issues
• Version Control - Github via gh CLI + stack PR extension
• Evidencing using agent-browser
• Harness - Tested with
  • Codex CLI + DeepSeek
  • CommandCode
  • Cursor

Installation

Quick Install All Agent Skills

Install every skill in this repo in one line — no clone required, same as the OpenCode / Claude Code installers:

curl -fsSL https://raw.githubusercontent.com/engineeringmadness/agent-skills/master/scripts/install-skills-global.sh | bash

Windows (CMD — download the batch script and run it; note that piping a script into cmd does NOT work, because %~1-style batch arguments and if (...) blocks only work in a real .bat file):

curl -fsSL -o "%TEMP%\install-skills-global.bat" https://raw.githubusercontent.com/engineeringmadness/agent-skills/master/scripts/install-skills-global.bat && call "%TEMP%\install-skills-global.bat"

Install Specific Agent Skills

Install the whole plugin with any Agent Plugins-compatible client, or install individual skills:

# List all skills in the plugin
npx skills add https://github.com/engineeringmadness/agent-skills --list

# Install a specific skill
npx skills add https://github.com/engineeringmadness/agent-skills --skill name-of-skill

Cloud Agent

Containerized the above setup with Codex + DeepSeek as the harness

Build the image

docker build -t coding-agent .

Create and run a container

Pass your API keys as environment variables and mount your project into /workspace:

docker run -d --name coding-agent -e DEEPSEEK_API_KEY=<key> -e GH_TOKEN=<token> -v "$(pwd):/workspace" coding-agent -c "tail -f /dev/null"

Attach to the running container when you want an interactive shell:

docker exec -it coding-agent bash