roombapy-prime-tools

Diagnostic and field-test tooling for roombapy-prime.

These scripts move a real robot. That is why they are a separate distribution: the library itself is installed into Home Assistant setups via ha_roomba_plus, and robot-moving commands have no business on the PATH of every one of those installations. Installing the library alone gives you no console scripts at all.

What this is for

Almost everything this project knows about the Prime/V4 protocol came from someone running one of these against their own robot and reporting back. If you own a Prime robot, that is by far the most useful thing you can do — and the read-only stages cannot change anything, so it costs you nothing but a few minutes.

Install

One command; it pulls the library in as a dependency:

python3 -m venv ~/roombapy-test-venv
source ~/roombapy-test-venv/bin/activate

pip install roombapy-prime-tools

Requires Python 3.11+. You will need to re-run the source line each time you open a new terminal.

These are not part of Home Assistant. They run on your own computer, outside HA entirely. HACS does not install them and does not need to.

Account details

Set these once per terminal rather than retyping a 32-character BLID on every command — every script reads them:

export ROOMBAPY_PRIME_USERNAME="you@example.com"
export ROOMBAPY_PRIME_BLID="YOUR_ROBOT_BLID"
export ROOMBAPY_PRIME_COUNTRY="US"      # your own: DE, FR, IT, ES ...

The country code matters — it selects which regional endpoint you authenticate against, and the default is US.

The password is deliberately not a command-line argument: it would end up in your shell history and in any terminal output you paste into an issue. Set ROOMBAPY_PRIME_PASSWORD for unattended runs, or let it prompt.

Any of these can be overridden per-run with --username/--blid/ --country-code.

Start here

roombapy-prime-validate

Read-only. Logs in, reads state, shadows, favorites, maps and mission history, then prints a report. Sends nothing to the robot. Its output alone answers several currently open questions.

Add --dump-config diagnose.json to save the raw responses. Review that file before sharing it — redaction catches the known cases, but cannot guarantee every surprise in an unfamiliar response shape.

The staged safety model

Every script that can change something follows the same pattern:

1. Stage 0 — reconnaissance. Reads only, sends nothing, needs no flag.
2. Stage 1+ — each requires an explicit --i-understand-this-... flag and an interactive confirmation showing the exact payload before anything is sent.

Start at stage 0. A stage-0 result is a genuinely useful contribution on its own — several findings in this project came from nothing more.

Everything validates before asking for your password, so an incomplete command tells you what is missing without making you type credentials first.

The scripts

Every script has --help, and it is worth reading before a first run.

The last two are the easiest way to contribute something genuinely new — both start read-only.

Region commands: solved, and worth re-running

Room-specific cleaning works. It took three rounds to establish, and two things were required that are not obvious:

• initiator is mandatory. A stored favorite does not carry one, so resending one unchanged is accepted, acknowledged and silently ignored.
• The wire keys are start and region_id, not clean and id.

roombapy-prime-verify-region-commands-session \
    --i-understand-this-will-move-my-robot \
    --i-understand-this-is-experimental-and-unconfirmed

It lists your eligible favorites and you pick one by number. Watch the robot while it runs. The fastest stop is the real iRobot app or the button on the robot.

Two notes from the sessions that got this working:

• A favorite made from specific rooms tells you far more than a whole-home one. If the robot cleans everything, you have only learned that the command arrived.
• Check the delivery confirmation in the output. Every stage that received one started a mission; every stage that did not got nothing. A "nothing happened" without a confirmation says nothing about the payload.

Virtual walls: reads work, writes do not

Stage 0 (--list-maps, then --list-walls) works and is read-only. It is how the zone types were confirmed against real data.

--update-unchanged works. The HTTP 500 this paragraph described was solved before 0.2.0b1 — virwall starts with a COUNT of the walls — and confirmed on two accounts, including the write / re-read / write round trip that separates "accepted" from "stored".

What is still untested: a write carrying a CHANGED list. Every confirmed write resent its zones unchanged, and set_virtual_wall replaces the whole shared list — anything omitted is deleted.

Developing from an unreleased checkout

The pin above names a git tag, which does not exist yet while a version is still in development. From a checkout, install the core editable first and skip dependency resolution here:

pip install -e .                     # from the repository root
pip install -e ./tools --no-deps

Without --no-deps, pip fetches a second, unrelated copy of the core from GitHub and shadows your editable one — which then fails in ways that look like a code bug rather than an install problem. The CI does exactly this, for exactly this reason.

Versioning

This distribution pins an exact core version. The scripts reach deep into the library, so a mismatched pair fails confusingly rather than cleanly — scripts/check_version_pin.py in the repository root enforces that the two stay in step.