Skip to content

About

Hands-on course for learning harness agents and AI repositories

Resources

Stars

1 star

Watchers

0 watching

Forks

Latest commit

 

History

11 Commits

Folders and files

NameName
Last commit message
Last commit date
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Playground AI

Greetings, adventurer.

Playground AI is a guild hall for learning how to build software with coding agents, repository instructions, reusable skills, and quest notices. Each quest lives in Markdown, becomes a GitHub issue with your name on it, and has terms you must meet. Your stronghold is <your-github-login>/main, and each road keeps a chronicle there.

TLDR

  1. Sign in to GitHub CLI if gh auth status fails: gh auth login
  2. Open this repository in your coding agent.
  3. Run /start-journey.
  4. Run /start-task. The agent chooses the task.
  5. Do the work with the agent.
  6. Run /complete.
  7. Repeat steps 4–6 until Quickstart is done.

If you are unsure where you are, run /mentor.

You never name the quest. Your stronghold is <your-github-login>/main. Petitions target that branch.

Start here

Ask your companion to run /start-journey. It greets you, checks that the GitHub CLI is installed and signed in, and asks you to confirm the Git name and @profusion.mobi email that will appear on your commits.

Everyone then completes the Quickstart path before choosing another path:

  1. Decide what to build
  2. Define the specification
  3. Build the application
  4. Improve the application
  5. Demo the application

Companions

From the specification quest onward, rely on your companion. Any companion that can read this repository can take up the work. These are suggestions:

  • Cursor. Profusion provides Cursor for some people.
  • Claude. The most popular companion.
  • Codex. The alternative.
  • OpenCode. The cheapest plan, with many free models.

Prerequisites

  • Git
  • GitHub CLI
  • A GitHub account
  • A companion that reads AGENTS.md and Agent Skills

Confirm that GitHub CLI is authenticated:

gh auth status

If it is not, sign in:

gh auth login

Quest workflow

  1. Ask your companion to run /start-journey.
  2. Ask your companion to run /start-task.
  3. Work with your companion until every term of the quest is satisfied.
  4. Ask your companion to run /complete.

You do not name the quest. /start-journey raises your stronghold, <your-github-login>/main, only after you approve the commands. /start-task reads your chronicle, your road, and your quest notices, then opens the next quest on <your-github-login>/<task-id>. The petition into the stronghold opens once that branch has a real quest commit. An empty commit does not count. A human other than the author glances at the evidence, not the code, and approves it. /complete tests the quest, verifies that approval, merges the petition, marks the quest completed, and closes the notice. Nothing pushes the guild road main.

Repository map

.
├── .agents/
│   ├── references/       # Shared logic the skills read
│   └── skills/           # Reusable agent workflows
├── .claude/
│   └── skills -> ../.agents/skills
├── paths/                # Learning paths
│   ├── README.md
│   ├── _templates/
│   └── quickstart/
├── progress/             # Created per learner and path on their branch
│   └── <github-login>/
│       └── quickstart.md
├── projects/             # Created on the learner's main branch
│   └── <project-name>/   # One folder per project; a route may reuse it
├── AGENTS.md             # Instructions for coding agents
├── CLAUDE.md -> AGENTS.md
└── README.md             # Instructions for learners

Works with any agent

Instructions live in AGENTS.md and skills live in .agents/skills/, the locations Codex, Cursor, OpenCode, and Antigravity already read. Claude Code looks elsewhere, so CLAUDE.md and .claude/skills are symlinks to those same files rather than copies.

Your agent may expose the skills as /start-journey, /start-task, /complete, and /mentor. If it has no slash commands, ask for the skill by name. /mentor reads your progress and names the next step.

Adding a learning path

Copy paths/_templates/task.md for each task. Give tasks ordered filenames, clear deliverables, and acceptance criteria that an agent can verify.

Put new skills in .agents/skills/<name>/SKILL.md with name and description frontmatter, where name matches the directory. Do not copy a skill into a harness-specific directory.

About

Hands-on course for learning harness agents and AI repositories

Resources

Stars

1 star

Watchers

0 watching

Forks

Releases

Packages

Contributors