Thalovant Skills for VS Code
Write a Thalovant skill and try it as you write it. The simulator runs your
skill next to the code in a real OVOS core, with the skill's own Python. Type
what someone would say, and see every way the skill answers:
- 🔊 what the speaker says, and the SSML twin, if the dialog has one;
- ✎ what an app or a screen writes (
speak_to(..., written=...));
- ▭ the card a screen shows, on a simulated Mark II screen, with buttons
that tap back into the skill;
- which intent and handler took the sentence, or who took it instead, or
that nobody did;
- the skill's own log lines, warnings, exceptions (with a link to the line
that raised) and every bus message.
Save a file and the skill restarts with your change. The conversation stays
on screen.

Install
- VS Code: search for Thalovant Skills, published by Thalovant,
in the Extensions view, open its
Microsoft Marketplace listing,
or run
code --install-extension thalovant.thalovant-skills.
- VSCodium, Cursor, Windsurf and other editors that install from
Open VSX:
search for Thalovant Skills in the Extensions view, or run
codium --install-extension thalovant.thalovant-skills.
- Install from a file: download
thalovant-skills-<version>.vsix from the
releases, then
Extensions → … → Install from VSIX…, or run
code --install-extension thalovant-skills-<version>.vsix.
The guide is at
docs.thalovant.com/developers/vscode-extension.
Getting started
Open a folder that holds a skill (a thalovant_skill_* package, as
thalovant-skillkit new writes it). Then click ▷ in the editor title, or press
Ctrl+Alt+V.
No skill yet? Run Thalovant: New Skill…. It asks for a name and where the
skill sits in the fallback ladder. It then creates the skill with its own
.venv, with SkillKit and OVOScope installed and its tests passing, and opens
it in the simulator.
The skill runs with its tests' Python: thalovant.python if set, then the
skill's .venv, then the Python extension's interpreter. If that Python lacks
thalovant-skillkit[testing], the simulator offers Set up the environment.
It installs the kit and the skill into the skill's .venv and changes nothing
outside the skill folder.
What else it does
Golden utterances in the Testing view. Rows in
test/end2end/golden_utterances*.jsonl are tests, in the fleet's format:
{"skill_id": "thalovant-skill-plant-care.thalovant", "utterance": "water the plants", "intent_label": "water.plants.intent", "lang": "en-US"}
A row passes when the sentence reaches this skill, and the intent when the row
names one. A failing row says who took the sentence instead. "negative": true
marks a sentence the skill must leave alone. In the simulator, Save as
golden and Should not match add the turn as a row.
Code that knows the locale tree. Inside speak_dialog("…"),
mentions(utterance, "…"), @intent_handler("…") and
display.layout_file(self, "…") you get:
- completion of the skill's names;
- a hover with each locale's lines and the locales that lack the file;
- go to definition;
- a warning for a name no locale has.
Over each handler and each intent line, a lens says one of the intent's
sentences.
SkillKit checks on save. thalovant-skillkit check runs when you save, and
its problems appear in the Problems panel on the file and line they name: a
missing translation, placeholders that differ, a pipe outside a group.
Thalovant: Check the Skill Against the Fleet compares this skill's
sentences with every other skill's.
The Skill view. In the activity bar: the skill's id and class, its locales
with what each lacks, its intents, dialogs, vocabulary and layouts.
Good to know
- The simulator routes a sentence the way a paired Thalovant hub does: stop,
converse, Padatious, common query, and the low fallback tier
(
thalovant.simulator.pipeline: "hub"). Paired hubs don't run Adapt, so
the simulator warns when the skill registers Adapt intents: they would answer
nowhere. "hub-m2v" adds the hub's model2vec prototype stages on
Thalovant's own encoder (downloaded on first use); "unpaired" adds Adapt,
as unpaired runtimes keep it. A sentence that another skill already claims is
caught by the fleet check, not by the simulator, because the simulator loads
only your skill.
- A layout card (
.rfwtxt) is drawn by the satellite's own widget library.
The simulator shows what a screen that cannot draw it shows: the card's
body, and the data bound to it.
- The skill runs your code, so the simulator, checks and tests need a trusted
workspace. Completion, hover and highlighting work in restricted mode too.
- The location sent with each sentence comes from
thalovant.simulator.location
or the 📍 menu. Settings the skill starts with come from
thalovant.simulator.skillSettings.
- Replies can be read aloud with
thalovant.simulator.speech: "system", where
the editor offers speech synthesis. The microphone is not available in an
editor panel, so you talk by typing.
Settings
| Setting |
Default |
|
thalovant.python |
|
The Python that runs the skill. ${skillFolder} and ${workspaceFolder} are expanded. |
thalovant.simulator.pipeline |
hub |
hub (a paired hub), hub-m2v (plus its model2vec stages) or unpaired (plus Adapt). |
thalovant.simulator.reloadOnSave |
true |
Restart the skill when one of its files is saved. |
thalovant.simulator.languages |
[] |
Languages the hub loads; empty means locale/supported.json. |
thalovant.simulator.location |
null |
The location in each sentence's context. |
thalovant.simulator.skillSettings |
{} |
The skill's settings at start. |
thalovant.simulator.speech |
off |
system reads replies aloud. |
thalovant.check.onSave |
true |
Run thalovant-skillkit check on save. |
How it works
The extension starts python/thalovant_dev_runner.py with the skill's Python.
The runner:
- enters SkillKit's isolated XDG tree;
- imports the skill class named by the
opm.skill entry point (no install
needed);
- boots an OVOScope MiniCroft with it.
The runner and the extension exchange one JSON object per line over stdio.
Each turn is collected from the bus firehose. A turn ends at
ovos.utterance.handled, or as soon as the skill asks a question, because a
get_response handler waits for the next utterance. A save restarts the
process; a MiniCroft boots in under a second.
npm install
npm run build # dist/extension.js
npm run test:unit # THALOVANT_TEST_PYTHON=<python with skillkit[testing]> for the runner tests
npm run test:integration # VSCODE_BINARY=<code> to use an installed VS Code
npm run package # thalovant-skills-<version>.vsix
The idea comes from Emkay Dev's VS Code extension for OpenVoiceOS skills.
Apache 2.0.