ds-env-setup

name	ds-env-setup
description	Use when setting up or repairing the Python env for analysis — detects uv/venv/conda/poetry/pipenv, installs core packages

ds-env-setup : prepare your project's Python environment

When this applies

Use this skill before running any ds-crew analysis for the first time in a project, when an analysis skill fails with ModuleNotFoundError, or when you change env managers and need to reinstall. Also triggered automatically by the optional SessionStart hook (see Stage 5).

The cardinal rule

Never install into a global Python. Every install command targets the active environment detected in Stage 2. If no env exists at all, create one with uv before installing.

Target interpreter: Python 3.11 or 3.12. This is the band with full DS-stack wheel coverage and the only one the feasibility gate's dependence-forecastability supports (requires-python >=3.11,<3.13). Older (3.9/3.10) risks missing wheels; newer (3.13/3.14) silently disables that tool (gate drops to the fallback). On macOS especially, bare python3 is often Apple's stock 3.9, so don't just accept the default — see the Stage 2.5 checkpoint.

How to run it

Stage 0 — Python check

python3 --version   # or: python --version

If neither command works: stop here. Install Python first:

• macOS: brew install python or https://www.python.org/downloads/
• Linux: sudo apt install python3 / sudo dnf install python3
• Any platform: https://github.com/pyenv/pyenv

Note the version — you'll need it in Stage 2.5. Classify it:

• In [3.11, 3.13) (3.11 or 3.12) → ideal, no checkpoint needed.
• Below 3.11 (e.g. macOS stock 3.9) or 3.13+ → off-target; the Stage 2.5 checkpoint applies.

Stage 1 — Scope check (informational)

Check whether ds-crew is installed globally or per-project:

grep -l "ds-env-setup" ~/.claude/settings.json 2>/dev/null && echo "global" || echo "not global"
grep -l "ds-env-setup" .claude/settings.json 2>/dev/null && echo "project" || echo "not project"

Print the result for the user's awareness. This does not change the env logic.

Stage 2 — Detect env manager

Run check_env.py (from this skill's scripts/ dir) to detect the manager:

python3 <skill_scripts_dir>/check_env.py

Reads the returned manager field. Priority order: uv > venv > conda > poetry > pipenv > none.

Stage 2.5 — Interpreter version checkpoint

Skip this stage if the Stage 0 version is already in [3.11, 3.13) and a project env already exists (manager ≠ none). Otherwise:

If uv is the manager (or manager is none and uv is on PATH): uv can fetch and pin any interpreter without touching the system Python — so don't silently inherit an off-target default. Ask the user (a real checkpoint — do not auto-pick):

Your default Python is {version from Stage 0}. ds-crew targets 3.11–3.12 (full DS-stack wheels, and the only band the forecastability gate supports). Since this project uses uv, I can provision any of these without touching your system Python. Which do you want?

• 3.12 — recommended. Newest supported; broadest wheel coverage.
• 3.11. Slightly more conservative.
• Keep default ({version}). Note: <3.11 may miss wheels; ≥3.13 disables dependence-forecastability (gate falls back).

Apply the choice before installing (Stage 3):

uv python install 3.12      # fetches it if not already present (use 3.11 if chosen)
uv venv --python 3.12       # existing-project path; creates .venv on the chosen interpreter
# — or, on the `none` path —
uv init --python 3.12 && uv venv --python 3.12

If the user keeps the default, proceed with no --python flag.

If the manager is venv / conda / poetry / pipenv and the default is off-target: these can't fetch interpreters on their own. Don't block — but surface it:

• Check for a better interpreter already installed: ls /opt/homebrew/bin/python3.1[12] /usr/local/bin/python3.1[12] 2>/dev/null (or pyenv versions).
• If one exists, tell the user they can recreate the env on it (e.g. python3.12 -m venv .venv) or switch to uv.
• If none exists, note the limitation (forecastability tool may be unavailable / wheel gaps) and continue with what's there.

Stage 3 — Install core packages

Run the env-specific command:

Manager	Command
uv	uv add pandas openpyxl matplotlib
venv	python -m pip install pandas openpyxl matplotlib
conda	conda install -y pandas openpyxl matplotlib
poetry	poetry add pandas openpyxl matplotlib
pipenv	pipenv install pandas openpyxl matplotlib
none	See below

If none detected: check for uv on PATH first:

which uv || echo "uv not found"

• If uv present: run the Stage 2.5 checkpoint first, then uv init --python <chosen> && uv add pandas openpyxl matplotlib (omit --python only if the user kept the default).
• If uv absent: stop and print install instructions:

  Install uv first: curl -LsSf https://astral.sh/uv/install.sh | sh
  Then re-run /ds-env-setup
  

  Do NOT fall back to a global pip install.

Stage 4 — Verify

python3 -c "import pandas; import openpyxl; import matplotlib; print('ready')"

If this fails, report which package is missing (the error message names it) and suggest re-running Stage 3. Print the Python path being used:

python3 -c "import sys; print(sys.executable)"

Stage 5 — Offer SessionStart hook (opt-in)

After a successful verify, ask the user:

"Want me to add a session-start check so you get a warning if the env breaks in future? I'll add a one-liner to .claude/settings.json."

If yes, append to .claude/settings.json (create the file + hooks key if absent):

"SessionStart": [{
  "hooks": [{
    "type": "command",
    "command": "python3 -c 'import pandas,openpyxl,matplotlib' 2>/dev/null || echo '⚠️  ds-crew: core packages missing — run /ds-env-setup'"
  }]
}]

Silent on success. One-liner warning on failure. Non-blocking.

Output

• Confirmation of Python version + path (and, if Stage 2.5 ran, the interpreter the user chose)
• Detected env manager
• Install command run + its output
• ready from the verification step
• (Optional) confirmation that the SessionStart hook was added

Quick reference

check_env.py                # print JSON env info for cwd
check_env.py --check        # exit 0 if packages ok, exit 1 + message if not (for hooks)

Packages installed: pandas, openpyxl, matplotlib

For kernel mode (ds-star-plus): also install ipykernel jupyter_client — see ../ds-star-plus/references/execution.md