GETTING_STARTED.md

Getting Started

This guide shows how to run two CLI agents in separate PTYs and have them iterate until they agree.

1) Install

npm install

2) Setup wizard (no manual edits)

node ./bin/criticloop.js setup

This writes criticloop.config.json if you choose to save, or runs from in-memory settings.

3) Run a task

node ./bin/criticloop.js --task "Design a minimal API and test plan"

4) Interactive roles (no config edits)

node ./bin/criticloop.js --interactive --task "Create a UI spec and review it"

5) Using the same model twice

Copy the template and edit the command to your CLI:

copy examples\criticloop.config.same-model.json criticloop.config.json

node ./bin/criticloop.js --task "Create a UI spec and review it"

6) Using two different CLIs

Copy the template and edit both commands:

copy examples\criticloop.config.two-clis.json criticloop.config.json

node ./bin/criticloop.js --task "Draft a plan and critique it"

7) Logs

Each run creates sessions/<timestamp>/ with:

• agentA.log
• agentB.log
• transcript.jsonl
• summary.json

On Windows, tail logs like this:

Get-Content -Wait sessions\<timestamp>\agentA.log

Tips

• If responses cut off, increase timeoutMs or idleMs in config.
• If a CLI prints extra banners, the sentinel helps detect message boundaries.
• For common issues, see docs/TROUBLESHOOTING.md.