| name | chatbi-auto-analysis |
| description | Auto-analysis middleware for uploaded tables. It profiles table rows, proposes metrics for user confirmation, executes accepted metrics with deterministic calculations, and emits frontend-ready analysis proposal/dashboard JSON for charts and dashboards. |
| trigger_conditions | ["chatbi-file-ingestion 已返回 rows，或对话含上传路径且需指标提案/采纳/看板","用户回复采纳、确认指标或要求上传表看板"] |
| when_not_to_use | ["尚无上传文件路径且无上一步 ingestion rows","演示库自然语言问数"] |
| required_context | ["上传路径或上一轮 file-ingestion 的 rows"] |
Workflow
Use this skill after chatbi-file-ingestion has read an uploaded CSV/XLSX and returned rows.
- Profile the uploaded rows: columns, semantic roles, time/numeric/category/id fields.
- Propose analysis metrics as frontend-renderable JSON plus Markdown.
- Ask the user to confirm selected metric IDs before execution when the request is exploratory.
- After confirmation, execute only supported metric plans with deterministic code.
- Return
charts, kpis, and dashboard_middleware so the frontend can render a full uploaded-file dashboard.
Commands
PYTHONPATH=. .venv/bin/python skills/chatbi-auto-analysis/scripts/auto_analysis.py '{"question":"生成上传文件看板","rows":[...]}' --json
PYTHONPATH=. .venv/bin/python skills/chatbi-auto-analysis/scripts/auto_analysis.py '{"question":"采纳全部指标","mode":"execute","rows":[...]}' --json
Safety
- Do not execute arbitrary Python, SQL, or formula text from an LLM.
- LLM planning may suggest metric candidates, but execution must validate field names and use whitelisted formulas.
- If required fields are missing or ambiguous, return
need_confirmation or need_clarification with Markdown guidance.
