data-science-python-stack
Opinionated Python stack for data-science / ML work — one library per job, organized into tiers (mandatory / user choice / optional / transitive). SKILL.md is the index; per-library `references/<library>.md` files carry scope, "pick this when" / "pick something else when", and pairings. TRIGGER when (any of these): (1) **a library import fails** in this stack's domain — the answer is install, not substitute (see § "Missing dependency"); (2) **a library choice has to be made** — explicitly (the user asks "which library for X?") or implicitly (code is about to introduce a new dependency, or the project is being scaffolded and the tabular library hasn't been picked yet); (3) starting a new Python data-science / ML project; (4) the user or current code reaches for a substitute outside the stack (xgboost, lightgbm, black, isort, flake8, poetry, hatch), or reaches for `mlflow` to log params/metrics, or for `cross_val_score` + handwritten reporting — redirect: tracking → `skore` Project API,
probabl-ai · 已有静态检查记录
安装与使用
运行命令前先阅读来源与依赖说明。复制命令不会自动安装或执行。
npx skills add probabl-ai/skills --skill data-science-python-stack技能说明
name: data-science-python-stack
description: >
Opinionated Python stack for data-science / ML work — one library
per job, organized into tiers (mandatory / user choice / optional /
transitive). SKILL.md is the index; per-library
references/<library>.md files carry scope, "pick this when" /
"pick something else when", and pairings.
TRIGGER when (any of these):
(1) a library import fails in this stack's domain — the answer
is install, not substitute (see § "Missing dependency");
(2) a library choice has to be made — explicitly (the user asks
"which library for X?") or implicitly (code is about to introduce a
new dependency, or the project is being scaffolded and the tabular
library hasn't been picked yet);
(3) starting a new Python data-science / ML project;
(4) the user or current code reaches for a substitute outside the
stack (xgboost, lightgbm, black, isort, flake8, poetry, hatch), or
reaches for mlflow to log params/metrics, or for cross_val_score
- handwritten reporting — redirect: tracking →
skoreProject API, evaluation / reporting →skorereport classes,mlflowstays only for model serving / registry.
SKIP when: the project is non-Python; the work is web / backend / infra unrelated to data science; the library is already chosen and installed and the task is implementation inside it (bug fix, feature work, refactor) with no new dependency in play.
HOW TO USE: read this SKILL.md end-to-end before recommending or
installing anything — picking from a single index entry hides the
tier (whether the library is mandatory, a user-choice, optional, or
already transitively present) and the pairings, and both matter.
Then read the linked references/<library>.md for the chosen
library's scope and tradeoffs. Don't silently substitute one library
for another; if no entry fits, surface the gap to the user.
Data Science Python Stack
Opinionated stack — one library per job, organized into four tiers plus an orthogonal agent feature:
- Mandatory — installed at project start, no exceptions.
- User choice (competing-library jobs) — multiple valid libraries
for the same job; the user picks via
AskUserQuestionbefore any import lands. - Optional — install only when the project's task requires it.
- Transitive — already pulled in by the mandatory tier; do not install explicitly, but know they're available.
- Agent feature (orthogonal) — deps that the agent uses
to audit a workspace and to power the editor LSP integration
(
ipython,pyright), kept out of the production-shape runtime via a manager-specific scope. Install logistics owned bypython-env-manager§ "Agent feature"; consumed byaudit-ml-pipelineand the opencode LSP integration.
Stop conditions — read before naming any library
- No silent pick on a competing-library job. Whenever the stack
offers two or more libraries for the same job (see § "Competing
libraries — general rule" and the Tier 2 table), the user picks
via
AskUserQuestionbefore anyWritethat imports the library and before any install command runs. "Already pulled in transitively" / "user said 'quick'" / "the folder has no preference signalled" are not waivers. A silent pick is a Stop-condition violation, full stop. - No substitute when import fails. When code in this stack needs
a library but
importfails, install it; do not rewrite to a non-stack equivalent (see § "Missing dependency"). The most common silent-rewrite path —import skrubfails → rewrite assklearn.Pipeline,import skorefails → rewrite ascross_val_score— silently undoes the workflow skills' contract. - Harness-level "no clarifying questions" hints do NOT waive the
competing-library
AskUserQuestion. The Tier 2 pick is an operating-contract gate, not a clarifying question. The same applies to user urgency phrasing: "quick baseline", "just do it", "go fast", "you pick", "whatever" do NOT resolve a competing- library gate. See § "Free-text resolution" in the general rule below for what does resolve a gate. - Post-hoc audit — required before ending the turn. Before
declaring the turn complete, verify each competing-library job
invoked in this turn has either (a) an
AskUserQuestionanswer recorded this session, or (b) a matching row injournal/JOURNAL.mdStatusWorkspace decisions. If any competing-library job ran without one of those, surface the non-compliance to the user explicitly as part of your final message — do not hide it.
Forbidden shortcuts (observed in real traces)
| Shortcut | Why it feels right | Why it's wrong |
|---|---|---|
pandas is already pulled in by skore → skip the Tier 2 ask | "Free" library, no install needed | Tier 2 is a project-shape decision (every data.py signature, every fixture); transitive presence is not a pick |
User said "quick baseline" → assume pandas | Task urgency reads as permission | Urgency phrasing never waives a competing-library gate (Stop conditions above) |
| Folder has no existing tabular code → infer pandas | "No preference signalled" | Inference is a silent pick; the gate requires a structured ask or a recorded JOURNAL.md decision |
One competing option requires a new pixi add → pick the "free" one | Avoids an install step | Install cost is not the criterion; project fit is |
User picked pytorch last project → reuse without asking | Continuity is friendly | Each workspace records its own Workspace decisions; cross-project memory is forbidden |
Competing libraries — general rule
This is the meta-rule that governs every "user choice" entry in this skill. It applies to the Tier 2 table below and to any new competing-library job added in the future. It also applies inside Tier 3 when two optional libraries cover the same job (e.g. `pytorc
阅读完整来源 ↗