habit-recipes

name

habit-recipes

description

End-to-end HABIT workflow recipes for common research scenarios (multi-modal MRI habitat, DCE-MRI kinetic, CT-only radiomics, test-retest reproducibility, demo dataset walkthrough). Use when the user wants to run the entire pipeline (preprocess -> habitat -> features -> ML -> comparison) instead of one isolated step. Triggers on phrases like "全流程", "端到端", "整个流程跑一遍", "demo 走一遍", "full pipeline", "end-to-end", "complete workflow".

HABIT End-to-End Recipes

Pre-baked workflows for the most common research scenarios. Each recipe lists:

1. The user phrase that should trigger it
2. Required inputs the agent must collect
3. The exact CLI sequence to run
4. Where each output lands
5. A validation checkpoint after every step

Available recipes

Recipe file	When to use	Modality
references/recipe_mri_habitat_full.md	Multi-modal MRI habitat study	T1+T2(+DWI/ADC)
references/recipe_dce_kinetic.md	DCE dynamic habitat study	DCE-MRI / dynamic CT
references/recipe_ct_radiomics_only.md	Classical radiomics on CT	CT
references/recipe_test_retest.md	Reproducibility / ICC study	any modality, two scans
references/recipe_demo_walkthrough.md	First-time user demo	bundled demo data

How the agent should use this skill

1. Identify which recipe matches the user's request (use the table above).
2. Open the corresponding recipe file in references/.
3. Walk the user through the Required Inputs checklist BEFORE writing any config — do not invent paths or modality names.
4. Generate each YAML config in turn, starting from the templates listed in skills/CONFIG_SOURCES.md under config/<module>/.
5. Run each command, then immediately call the scripts/validate_*.py from the corresponding skill to verify outputs before moving on.
6. If a step fails, hand off to habit-troubleshoot.

Recipe selection cheat sheet

User says...                                                Pick recipe...
---------------------------------------------------------- -------------------------------
"我有 T1+T2+DWI+ADC, 想做生境分析建模"                       recipe_mri_habitat_full
"DCE 动态对比增强分期生境 / kinetic"                         recipe_dce_kinetic
"我只有 CT, 想做传统影像组学"                                recipe_ct_radiomics_only
"两次扫描看重复性 / test-retest / ICC"                       recipe_test_retest
"我刚装好, 想跑一下 demo 看看"                               recipe_demo_walkthrough
"我想跑全流程"                                               ASK which scenario, then pick

Universal best practices for any recipe

1. Always run check_environment.py first — verify HABIT is installed.
2. Always run check_data_layout.py before preprocess — catches missing modalities early.
3. Always validate output after each major step — validate_preprocess_output.py then validate_habitat_output.py then inspect_feature_csv.py.
4. Save every config the agent generates in a ./configs/ subfolder of the user's working directory; reproducibility matters.
5. Use absolute paths in configs when the user has files scattered across drives; use relative when everything is under one project root.