一键导入
eda-workflow
13-Phase Auditable EDA Pipeline orchestration. Use when user wants to explore data, run analysis, or asks about the pipeline. Triggers: EDA, 資料探索, 分析資料, pipeline, 13-phase, analyze, 探索.
菜单
13-Phase Auditable EDA Pipeline orchestration. Use when user wants to explore data, run analysis, or asks about the pipeline. Triggers: EDA, 資料探索, 分析資料, pipeline, 13-phase, analyze, 探索.
Generate comprehensive test suites including static analysis, unit tests, integration tests, E2E tests, and coverage reports. Triggers: TG, test, 測試, 寫測試, coverage, 覆蓋率, pytest, unittest, 驗證.
Codex drawing harness integration. Triggers: 繪圖, draw, figure, chart, plot, mermaid, SVG, Gemini, graph, 生成圖表.
13-Phase Auditable EDA Pipeline orchestration. Use when user wants to explore data, run analysis, or asks about the pipeline. Triggers: EDA, 資料探索, 分析資料, pipeline, 13-phase, analyze, 探索.
Phase 10 report assembly and export workflow. Use when user wants to generate, review, or export the EDA report. Triggers: 報告, report, assemble, export, 產出報告, 匯出.
13-Phase Auditable EDA Pipeline orchestration. Use when user wants to explore data, run analysis, or asks about the pipeline. Triggers: EDA, 資料探索, 分析資料, pipeline, 13-phase, analyze, 探索.
Phase 8 report assembly and export workflow. Use when user wants to generate, review, or export the EDA report. Triggers: 報告, report, assemble, export, 產出報告, 匯出.
| name | eda-workflow |
| description | 13-Phase Auditable EDA Pipeline orchestration. Use when user wants to explore data, run analysis, or asks about the pipeline. Triggers: EDA, 資料探索, 分析資料, pipeline, 13-phase, analyze, 探索. |
協調 RDE 13-Phase Auditable EDA Pipeline 的完整流程,確保每個 Phase Gate 正確通過。
init_project(name)
→ project.yaml + artifacts/ 目錄樹
scan_data_folder(path)
run_intake()
├ [H-001] 檔案大小 < 500MB
├ [H-002] 格式白名單 (CSV/Excel/Parquet/SAS/SPSS/Stata/TSV)
└ [H-004] PII 初篩
→ intake_report.json
load_dataset(file)
build_schema()
profile_dataset() # ydata-profiling (fallback: basic engine)
→ schema.json + profile report
align_concept(research_question, variable_roles, confirm=true)
→ concept_alignment.md + variable_roles.json
Agent 必須: 向用戶展示對齊結果,等待確認。
propose_analysis_plan(confirm=true)
→ greedy_analysis_candidates.json
→ greedy_analysis_candidates.md
→ greedy_execution_schedule.json
→ greedy_execution_schedule.md
→ greedy_plan_enrichment.json
→ greedy_plan_enrichment.md
→ greedy_statsmodels_base_analysis.py
→ greedy_analysis_candidates.json + greedy_analysis_candidates.md
register_analysis_plan(confirm=true)
→ analysis_plan_review.json
→ analysis_plan_review.md
→ analysis_plan.yaml (LOCKED after confirmation)
Agent 必須: 展示完整計畫(方法、α 值、missing 策略),等待確認後鎖定。
check_readiness()
├ [H-003] 樣本量 ≥ 10
├ [S-001] 常態性檢定
├ [S-005] 缺失模式 (MCAR/MAR/MNAR)
└ [S-007] VIF 共線性
→ readiness_checklist.json
如需調整方法: log_deviation() → 告知用戶。
apply_cleaning()
generate_table_one()
compare_groups() × N
analyze_variable() × N
correlation_matrix()
→ H-009: 每步自動寫入 decision_log.jsonl
→ S-002 多重比較, S-009 Effect size, S-010 Power
偏離計畫時: 必須 log_deviation() 並說明理由。
collect_results()
→ results_summary.json (含 statistically significant candidate markers)
assemble_report()
→ eda_report.md (含 decision_log + deviation_log appendix)
run_audit()
→ audit_report.json (A/B/C/D/F 評分)
auto_improve() # 根據 audit 自動修正
export_handoff() # 產出 handoff package → med-paper-assistant
| 檢查點 | 規則 | 後果 |
|---|---|---|
| Phase 3 | 用戶確認 | 未確認不能進 Phase 4 |
| Phase 4/5/6 | 用戶確認 + 鎖定 | 未鎖定不能進 Phase 8 |
| Phase 8 偏離 | 必須 log_deviation | 審計時扣分 |
| Phase 10 報告 | H-005 完整性 + H-006 路徑清除 | 不完整或有敏感路徑則失敗 |
| Any Phase | H-008 artifact gate | 前一 Phase 未完成不能跳過 |
| 用戶意圖 | 走哪條路 |
|---|---|
| 「只想看概況」 | Phase 0→1→2 → profile → Quick Report |
| 「比較兩組」 | 完整 Phase 0-7 → compare_groups → Phase 9-12 |
| 「做 Table 1」 | 完整 Phase 0-7 → generate_table_one → Phase 9-12 |
| 「完整分析」 | 完整 13-Phase |