Quick start: the command line

ruby exe/harness init                                   # write config.json, asking a few questions
ruby exe/harness run --repo /path/to/repo --request request.md
ruby exe/harness status                                 # the newest run: tasks, current step, rejections
ruby exe/harness status RUN_ID --watch                  # refresh every second until it stops
ruby exe/harness answer RUN_ID                          # answer the planner's blocking questions
ruby exe/harness resume RUN_ID                          # continue after a stop, a crash or new answers
ruby exe/harness report RUN_ID                          # (re)write runs/RUN_ID/report.md and print its path

request.md is plain text, as you would write it to a colleague:

Add a Badge component at src/components/Badge.astro. It takes a required label and an optional tone,
"info" or "warn" (default "info"), and renders <span class="badge badge--<tone>">. Use it on the pricing page.

Useful options:

Option	Where	What it does
--config FILE	every command	Use another config; runs and caches live next to it.
--json	every command	Machine-readable output.
--plan-only	run	Stop after planning; edit runs/<id>/plan.json, then resume.
--from-plan FILE	run	Skip the planner and run a plan you wrote.
--answers FILE	answer	A JSON array of answers instead of typing them.
--port N, --[no-]open	ui	The app’s port, and whether to open the browser.

Progress lines go to stderr; the result goes to stdout. Exit codes: 0 done, 1 a task was blocked or something regressed, 2 a configuration or environment problem, 3 waiting for your answers.