Quickstart

Install the harness, prepare project context, and give your agent its first task.

1. Install

Requires Node.js 20+, npm, Git, and a supported coding agent.

npm install -g wepwawet
wep --version
wep --help

For development builds, see building from source.

2. Init in your project

Go to any repository where you use a coding agent and run:

cd /path/to/your-project
wep init

wep detects existing agent setup (Codex, Claude Code, Copilot, Pi). One detected target is selected directly. With none or several, it asks which target to install in an interactive terminal; pass --target for headless use.

To skip the prompt:

wep init --target codex      # OpenAI Codex
wep init --target claude     # Claude Code
wep init --target copilot    # GitHub Copilot
wep init --target pi         # Pi
wep init --target all        # all at once

Preview what would be created without writing anything:

wep init --target codex --dry-run

3. Open your agent and run doctor

After init, open your coding agent in the project and ask:

Run wep doctor.

Ask it to use the installed wep-doctor workflow. The agent inspects relevant project files and records verified architecture, conventions, testing commands, and decisions in .wep/wiki/. If init wrote a root-instruction suggestion, review and merge it first.

You can also run the CLI doctor to get a readiness score and suggested next prompt:

wep doctor

4. Create your first task

Ask your agent directly — no CLI needed:

Fix the login bug where the session expires too early.

The harness instructs the agent to read AGENTS.md (or its target equivalent), clarify material ambiguity, create a task capsule in .wep/tasks/, select a workflow, and implement. The built-in bug-fix recommendation is TDD; a concrete project default or explicit request can change it.

Or create a skeleton capsule via CLI (useful for CI or scripting):

wep task "Fix login session expiry bug"

5. Check status

wep status

Shows installed targets, task count, and current workflow default.

Independent review

Normal init enables the judge for Codex, Claude Code, and Copilot. After you approve the initial task brief, the installed workflow invokes it automatically after implementation. The brief covers the selected workflow, acceptance criteria, and exact validation commands. If your installation disabled it or omitted its module, enable it with:

wep judge enable

Use wep judge status to inspect the resolved state or wep judge disable to turn it off. Comprehension, the implementer, and tracing are separate opt-ins. See JUDGE.md for snapshot capture, command approval, validation receipts, and verified review records.

Reference

Goal	Command
Install harness	wep init
Audit setup	wep doctor
Check status	wep status
Create task capsule	wep task "<description>"
Compile agent brief	wep compile TASK-001
Show config	wep config show
Change workflow default	wep config set defaultWorkflow SDD+TDD
Enable judge	wep judge enable
Update harness to latest	wep upgrade
Remove harness	wep remove --target codex --force

See the command guide for modules, command approval, task search, templates, tracing, continuation, and removal options.