English | 简体中文
An AI-driven data analysis workflow plugin for data analysts. Covers the full pipeline from requirement clarification to report delivery, while continuously building reusable domain knowledge — the more you use it, the smarter it gets.
Most AI data-analysis tools start from a blank slate every time: table schemas must be re-explained, metrics re-defined, and mistakes re-made.
This plugin is built around a knowledge-driven workflow:
- Teach it a table schema or business rule once, and it remembers forever.
- Save a SQL template once, and reuse it next time.
- Record a pitfall once, and get warned automatically next time.
- After every task, extract reusable knowledge so the system compounds.
Inside Claude Code:
/plugin marketplace add guoqiaoZhou/data-analysis-plugin
/plugin install data-analysis
claude --plugin-dir /path/to/data-analysis-ccThen run /da-help to verify the plugin loads.
Not sure which to use? Run
/da-help— it picks the right command for you.
Core workflow
| Command | Usage | What it does |
|---|---|---|
/da-ideate |
[requirement hint] |
Clarify a fuzzy idea into an executable requirement |
/da-analyze |
[topic] |
Build an analysis plan (framework, metrics, dimensions) |
/da-sql |
[query requirement] |
Generate HiveSQL from the knowledge base |
/da-validate |
[file hint] |
Validate draft SQL by running partial queries |
/da-process |
[file hint] |
Process raw data + generate charts (self-debugging Python scripts) |
/da-report |
[focus hint] |
Write a Markdown report + Excel with traceable data sources |
/da-compound |
[file hint] |
Extract reusable knowledge from task outputs |
Knowledge management
| Command | Usage | What it does |
|---|---|---|
/da-setup |
— | Cold-start guide: configure query tool, seed knowledge base |
/da-ingest-table |
[DDL or description] |
Ingest table metadata |
/da-ingest-sql |
[SQL or file] |
Ingest SQL templates |
/da-ingest-case |
[report or file] |
Ingest analysis cases and frameworks |
/da-ingest-domain |
[rules or description] |
Ingest business domain knowledge |
/da-knowledge-history |
[file path] |
View knowledge-base git history and diffs |
Experiment tooling
| Command | Usage | What it does |
|---|---|---|
/da-split |
[grouping requirement] |
Split samples (random / stratified / PSM) + balance check |
Single query
/da-sql "DAU by city yesterday"Weekly report
/da-analyze "weekly report"
/da-sql
# download data, put CSVs into {task_dir}/data/
/da-process
/da-reportThematic analysis
/da-ideate "analyze user situation"
/da-analyze "new feature launch effect"
/da-sql
/da-validate
# download data, put CSVs into {task_dir}/data/
/da-process
/da-report
/da-compoundExperiment grouping
/da-sql
# download sample data, put CSVs into {task_dir}/data/
/da-split "balance by city and tier, 5:5 split"User data lives inside the current working directory:
{project_root}/
├── .da-knowledge/ # knowledge base (tables, SQL patterns, metrics, cases, domain rules)
├── .da-config.yaml # query-tool configuration
└── outputs/ # task outputs organized by type and date
└── .active-task # pointer to the current active task
Each task is a self-contained directory:
outputs/adhoc/2026-05-24-dau-trend/
├── task.yaml
├── README.md
├── analysis-plan.md
├── sql/draft/
├── sql/final/
├── data/
├── scripts/
├── figures/
├── processed/
└── report/
- Skills orchestrate, agents think: Skills handle flow control, user interaction, file I/O, and knowledge retrieval. Agents handle focused reasoning tasks (SQL writing, validation design, Python generation, reporting).
- Knowledge-first: Before writing SQL, agents automatically retrieve relevant table metadata, SQL patterns, metrics, and join patterns from the knowledge base.
- Traceability: Every data point in a report must cite its source file; no fabricated numbers.
- Standalone scripts: Generated Python scripts follow the D8 principle — fully self-contained, with clear imports and comments, runnable without the plugin.
MIT