Use py-harness in VS Code

VS Code talks to py-harness through Tasks. Each task opens the integrated terminal, runs the same command as Start, and stays inside the folder you have open. There is no chat plugin to install and no tunnel.

This page is a live walkthrough on demo/orders. 5 September 2026. brief needs no model. ask calls llama3.1:8b. The NameError repair is a harness demo — no model.

Recorded with asciinema. The GIF is what Tasks: Run Task runs in the integrated terminal, not a screenshot of the editor window. Replay:

asciinema play docs/media/vscode-demo.cast

Re-record it (needs Ollama llama3.1:8b):

PYTHONPATH=src python scripts/measure/record_vscode.py

One-time setup

From a clone (same installer as Start):

python3 -m venv .venv && source .venv/bin/activate
pip install py-harness-cli
ollama pull llama3.1:8b
py-harness editors vscode

From a clone: python3 scripts/run/install.py then source .venv/bin/activate.

macOS often has no pip on PATH. The installer creates .venv and runs python -m pip for you. Activate that venv in every new terminal or py-harness will still be missing.

Without installing, from the checkout:

PYTHONPATH=src python3 -m harness editors vscode

--project defaults to the folder you are in. The command writes .vscode/tasks.json. The project argument is ${workspaceFolder}. The interpreter is the one that ran editors vscode, so the file can hold a home path. Do not commit that path.

Open that folder in VS Code. If the window was already open:

1. Command Palette (Cmd+Shift+P or Ctrl+Shift+P)
2. Developer: Reload Window

Then in VS Code

1. Command Palette → Tasks: Run Task
2. Pick one of:
   • py-harness: brief
   • py-harness: ask
   • py-harness: run
3. For ask and run, type the job in the box. The default in the box is add multiply(a, b) and a unit test. Replace it.

The answer appears in the Terminal panel. That is the whole demo.

Do not type py-harness: ask in the shell. The colon is the task label in the Command Palette. In a terminal the commands are:

source .venv/bin/activate
cd demo/orders
py-harness brief
py-harness ask  "what does compute_total return?"
py-harness run  "find the NameError and fix it"

If the shell says command not found: py-harness, the venv is not active. Activate it in every new terminal.

layout is not a task. In the same terminal:

py-harness layout

Live: this folder, those tasks

Open demo/orders (or this repo and stay in that folder). Run Task. editors, brief, ask, and the NameError run below are the GIF. layout and the already-covered test are the same folder, not on the recording.

py-harness editors vscode

Writes .vscode/tasks.json. No model.

$ py-harness editors vscode
/private/tmp/vscode/.vscode/tasks.json

Reload the window, then Command Palette → Tasks: Run Task → py-harness: ask

py-harness: brief

No prompt. Instant. No model.

$ py-harness brief

10 Python and Markdown files, 2.9 KB in total.
Small enough that py-harness can read all of it, so you can ask about any part.

Files:
  README.md  685 B
  src/__init__.py  68 B
  src/orders.py  467 B
  src/orders_controller.py  568 B
  src/orders_service.py  285 B
  src/render.py  193 B
  src/report.py  155 B
  src/util.py  67 B
  tests/__init__.py  0 B
  tests/test_orders.py  511 B

py-harness has 24 skills it can apply. It picks them from the wording of
your task; you do not choose them.

py-harness layout (terminal)

layout: 1 finding(s), worst first.
  [cycle] src/render.py and src/report.py import each other

Next move (do only this one): Move what they share into a new
module both import.

Also instant, also no model. That cycle is in the tree: render.py imports build_report, report.py imports render_line.

py-harness: ask

Type: what does compute_total return?

Needs the 8B. About half a minute on this run. The first draft was only int. The helper sent that back. The second draft named what the function computes.

$ py-harness ask "what does compute_total return?"
ollama:llama3.1:8b  project /private/tmp/vscode  mode small

--- step 1 ---
Action: done
Summary: “int”

too thin. Action: done Summary: quote int and say in a sentence what it
computes, from the code you read.

--- step 2 ---
Action: done
Summary: "int" - it computes the sum of the line prices of one order.

"int" - it computes the sum of the line prices of one order.

ask never writes a file.

py-harness: run

Type: write tests for apply_discount

The test is already there. Nothing is written. Suite stays green.

already has a test for apply_discount

Type: find the NameError and fix it

The planted typo is subtotl next to subtotal =. One candidate. No model. About a tenth of a second.

$ py-harness run "find the NameError and fix it"

bound unique NameError typo (subtotl → subtotal) in src/orders.py. Tests passed.

run only writes when you pass --allow-writes to a server, or when you use the task (the task is a real py-harness run in the project folder, so it may write). It keeps a .bak of anything it edits.

What to reach for

For jobs the model has to reason about, read Experiments first.

Optional: chat through Continue

py-harness editors continue

That writes a Continue config aimed at local Ollama. Chat then uses the editor’s tools, not the py-harness write limit. For a change that must stay in one folder, use Run Task.

Do not point a hosted OpenAI-compatible chat at 127.0.0.1. That request often leaves the machine. Tasks stay here.

Related: In your editor · Cursor · local editor · IDE plugins · Live demo (shell).