AGENT SKILL

ai-data-science-team

Operate the ai-data-science-team package for AI-assisted data loading, EDA, pandas transformations, SQL analysis, H2O/MLflow modeling, multi-agent teams, and Streamlit app workflows.

License 许可
MIT
最近更新
2026-09-03
Stars
★ 351
Forks
27

01 前言 Front Matter

name
ai-data-science-team
license
MIT
metadata
{ "disco-role": "operating" }
description
Operate the ai-data-science-team package for AI-assisted data loading, EDA, pandas transformations, SQL analysis, H2O/MLflow modeling, multi-agent teams, and Streamlit app workflows.
disable-model-invocation
true

02 资源文件 5

03 SKILL.md 正文

AI Data Science Team repo skill

Use this repo skill when a task names ai-data-science-team, ai_data_science_team, AI Data Science Team, AI Pipeline Studio, or asks for AI-assisted data-science agents that load data, clean/wrangle/visualize DataFrames, query SQL databases, run H2O AutoML/MLflow tools, or compose data-science agent teams.

This skill is self-contained: read the bundled references and scripts here instead of reopening the original repository checkout, notebooks, or app source.

Quick package context

  • Distribution name: ai-data-science-team.
  • Import package: ai_data_science_team.
  • Verified version baseline: 0.0.0.9017.
  • Public install floor from package metadata: Python 3.9+; Python 3.10+ is recommended for app workflows.
  • Common LLM backends: OpenAI chat models through langchain_openai.ChatOpenAI, or local Ollama models through langchain_ollama.ChatOllama when an Ollama service/model is available.
  • Optional extras: machine_learning installs h2o and mlflow; data_science installs pytimetk, missingno, and sweetviz; D-Tale report support needs dtale separately.

Minimal import check:

from ai_data_science_team import DataCleaningAgent, SQLDatabaseAgent, PandasDataAnalyst

If that import fails with ModuleNotFoundError: IPython, install ipython; the source imports IPython.display but the package metadata may not declare it.

Route by task

Task signal Read
Discover files, load CSV/Excel/JSON/Parquet, summarize DataFrames, use file tools, run basic EDA summaries, or opt into Sweetviz/D-Tale/missingno/correlation-funnel reports sub-skills/data-access-and-eda/SKILL.md
Use DataCleaningAgent, DataWranglingAgent, DataVisualizationAgent, or FeatureEngineeringAgent; retrieve generated code/functions; handle retries/logs/human review sub-skills/dataframe-code-agents/SKILL.md
Query SQL databases, inspect database metadata, protect read-only SQL, use SQLDatabaseAgent, or troubleshoot schema/token pressure sub-skills/sql-analysis/SKILL.md
Train or inspect H2O AutoML workflows, evaluate models, manage MLflow experiments/runs/artifacts/UI/registry, or check optional ML dependencies sub-skills/modeling-and-mlflow/SKILL.md
Compose agents with PandasDataAnalyst, SQLDataAnalyst, SupervisorDSTeam, WorkflowPlannerAgent, or understand AI Pipeline Studio / Streamlit app patterns sub-skills/multiagent-and-app-workflows/SKILL.md

First actions for future agents

  1. Identify whether the user wants deterministic helper functions, LLM-backed agents, optional report/model services, or Streamlit apps.
  2. Check installed prerequisites before invoking agents. Use scripts/check_env.py for package import/signature and optional dependency visibility.
  3. For deterministic data/SQL helper tasks, prefer bundled smoke scripts from the owning sub-skill before any LLM call.
  4. For LLM-backed agents, confirm the user has configured a LangChain-compatible model object and any provider credentials/service.
  5. For optional H2O/MLflow/EDA reports or app workflows, read the owning troubleshooting reference before installing extras, launching services, or starting long-running training/UI processes.

Shared references and helpers

Safety and verification boundaries

  • Do not run notebook workflows, LLM-backed agents, H2O training, MLflow UI, or Streamlit apps unless the user explicitly wants that side effect and has provided credentials/services/budget.
  • Treat generated code from cleaning/wrangling/visualization/feature-engineering agents as code to review, log, and validate on small data before using on production data.
  • Keep SQLDatabaseAgent(safe_mode=True) unless the user intentionally authorizes non-read-only SQL. The bundled SQL sub-skill describes what the validator does and does not guarantee.
  • Optional extras are not required for base imports. Install only the extra needed by the selected workflow, not all by default.
  • Runtime guidance here must remain independent of the original checkout; if you need an example, use the bundled references/scripts.

04 同仓库更多 Skills

VectorSpaceLab
VectorSpaceLab/AREX-Skill

passnet-triton-opt

Make a single PassNet Triton kernel fast AND numerically faithful: performance model (when a replacement can win at all), block/grid/warp tuning, autotune policy, launch overhead, and per-op numeric recipes to pass the dtype baseline tolerances. Use when a pass matches and is correct but speedup below expected, or when correctness fails by small numeric margins.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

passnet-solve

END-TO-END playbook for solving one PassNet sample: analyze the computation graphs, decide the optimization strategy (what to fuse, what to replace, what to leave alone), drive the iteration loop, and maximize the sample score. This is the ENTRY skill — invoke it first for any PassNet optimization task; it tells you when to use passnet-pattern-fusion, passnet-triton-opt and passnet-feedback.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

passnet-skill

PassNet GPU kernel optimization via compiler passes. Design and implement Triton-based optimization passes, create pass files under ./pass_dir/, self-evaluate with pass_evaluator, and iterate to maximize GPU speedup.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

passnet-pattern-fusion

Author PassNet pass files: write patterns that actually MATCH the FX graph, pick fusion regions (which ops to absorb into one Triton kernel), and structure multi-pass submissions with the mandatory shared-dispatch architecture. Use when creating or fixing pattern()/replacement_args()/replacement_func() files, when a pass "failed to match", or when deciding how to fuse multiple kernels/ops.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

passnet-orchestrate

Round/batch planning and decision support for PassNet graph optimization. Use for multi-sample triage, worker allocation, eval-budget policy, and post-eval keep/revert decisions. For solving one concrete sample end to end, use passnet-solve as the entry skill; consult this skill only when you need broader planning or a specific decision gate.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

passnet-feedback

Fast, deterministic feedback for PassNet work: pre-flight pattern/match verification WITHOUT burning a GPU evaluation, per-node bottleneck analysis of a sample's graphs, and eval-log parsing into per-variant status + estimated score + failure classification. Use BEFORE every GPU evaluation (check_pattern), at the START of a sample (analyze_graph), and AFTER every evaluation (parse_eval_log).

★ 351