hadron-notebook-kernellisted

# Notebook Kernel Hadron launches marimo and Jupyter using a venv recorded in `.hadron/config.json` under `kernels`: ```json { "kernels": { "marimo": "/abs/path/.venv", "jupyter": "/abs/path/.venv" } } ``` If unset, it falls back to the workspace's `.venv/`. marimo and Jupyter can share one env or use different ones. ## 1. Find the workspace + current config ```bash WORKSPACE=$(find ~ -maxdepth 3 -name ".hadron" -type d 2>/dev/null | head -1 | xargs dirname) echo "Workspace: $WORKSPACE" curl -s http://localhost:${HADRON_PORT:-3000}/api/kernels ``` ## 2. Scan for usable venvs ```bash for d in .venv venv env; do p="$WORKSPACE/$d" [ -x "$p/bin/python3" ] && { echo "FOUND: $p"; "$p/bin/python3" -m pip list 2>/dev/null | grep -iE "marimo|jupyter"; } done command -v uv >/dev/null && echo "uv available" ``` ## 3. Create one if needed Prefer `uv`; fall back to stdlib. Always `python3`, never `python`: ```bash cd "$WORKSPACE" uv venv .venv && uv pip install marimo jupyter altair pandas # preferred # or: python3 -m venv .venv && .venv/bin/python3 -m pip install marimo jupyter altair pandas ``` ## 4. Save the choice The kernels route is a mutating endpoint, so it needs the workspace token (read from `.hadron/token`). No Origin header is sent by curl, so the token alone authorizes it: ```bash TOKEN=$(cat "$WORKSPACE/.hadron/token") curl -s -X PUT http://localhost:${HADRON_PORT:-3000}/api/kernels \ -H "Content-Type: application/json" \ -H "x-hadron-token: $TOKEN" \