Your Claude Engineer

Workflow

Autonomous AI software engineer harness: orchestrator plus Linear, Coding, GitHub, and Slack subagents for long-running tasks.

by Cole Medin (coleam00)coleam00/your-claude-engineer

Stars141

Forks50

Watchers1

Open issues1

LicenseMIT

Last push8 months ago

Install

install

$ git clone https://github.com/coleam00/your-claude-engineer

Update

update

$ git pull

Cloned Python harness — pull latest, then re-run pip install -r requirements.txt to sync dependencies.

When to use it

You want to hand off a feature spec and have an autonomous agent plan, code, test, commit, and report over many iterations.
You already use Linear for issue tracking and want issues created/updated automatically as work progresses.
You want a multi-agent reference architecture (orchestrator + specialized subagents) built on the Claude Agent SDK to study or fork.
You want GitHub commits/PRs and Slack progress updates wired through a single Arcade MCP gateway.

Commands & flags

pip install -r requirements.txt: Install the Python harness dependencies (after npm install -g @anthropic-ai/claude-code).
python authorize_arcade.py: Run the one-time Arcade OAuth flow for Linear/GitHub/Slack tools.
uv run python autonomous_agent_demo.py --project-dir my-app: Run an autonomous session that builds my-app in ./generations/my-app/.
--max-iterations 3: Cap the number of agent iterations (useful for testing).
--model opus: Set the orchestrator model (haiku/sonnet/opus); per-agent models set via env vars.
prompts/app_spec.txt: Edit to change which application the harness builds.

How to use it

This is a Python harness on the Claude Agent SDK, not a Claude Code skill/plugin. Linux/WSL only (Claude Agent SDK subagent limitation — no native Windows).

git clone https://github.com/coleam00/your-claude-engineer
cd your-claude-engineer
python3 -m venv venv && source venv/bin/activate
npm install -g @anthropic-ai/claude-code
pip install -r requirements.txt

Configure credentials and authorize the Arcade gateway (Linear required; GitHub/Slack optional):

cp .env.example .env   # set ARCADE_API_KEY, ARCADE_GATEWAY_SLUG, ARCADE_USER_ID
python authorize_arcade.py

Run an autonomous session:

uv run python autonomous_agent_demo.py --project-dir my-app --model opus

Customize what gets built via prompts/app_spec.txt, the initial issue count via prompts/initializer_task.md, and the bash allowlist via ALLOWED_COMMANDS in security.py.

Why it earns its place

A working, end-to-end reference for long-running autonomous engineering: an orchestrator delegates to Linear, Coding (with Playwright UI testing), GitHub, and Slack subagents, with per-agent model selection (haiku/sonnet/opus) to control cost and context. By Cole Medin (coleam00), a widely-followed AI-engineering educator. The Arcade MCP gateway gives one OAuth flow across Linear/GitHub/Slack, and the defense-in-depth security model (OS sandbox, filesystem restrictions, bash allowlist with rm validation) makes it a credible base to fork rather than a toy demo.

Try it

claude

> uv run python autonomous_agent_demo.py --project-dir todo-app --model opus --max-iterations 5