Restructure skills/ to be operator-focused

[Can-Zhao]

Skills now follow a uniform template oriented at someone running inference: Command to run → How to configure a run → Output → Related scripts → Related skills. Algorithm step-by-step walkthroughs and internal function-args tables removed; workflow figures kept.

Changes per skill:

• skills/infer_mask-generation.md → renamed to infer_mask-only.md to parallel infer_image-only. Body rewritten around the two mask-acquisition paths (Path A diffusion-from-scratch vs Path B training-mask DB lookup) and the anatomy_size 10-d slot table.

• infer_image-only.md: restructured around ## How to configure a run (6 numbered steps: modality, AE knobs, dim/spacing from FOV, cfg_guidance_scale_modality, cfg_guidance_scale_tumor N/A, num_inference_steps). New end-to-end worked example: T1-axial rflow-mr-brain with concrete dim and spacing values. CFG section spells out "classifier-free guidance" at first use and links to the per-plane FOV table in docs/inference.md.

• infer_image-from-mask.md: rewritten as a CLI-operator guide — command to run, input mask format and how to produce one (nv-segment-ct + add_body_envelope, or other segmenter + remap), config knob explanation including cfg_guidance_scale_tumor and the CT-only scope, output file layout. Cross-links to infer_mask-image-paired for the GPU-memory presets.

• infer_mask-image-paired.md: keeps the LDMSampler dispatch flowchart; drops the algorithmic "Two paths in detail" prose, internal LDMSampler.init state table, and Quality-check loop details. Gains a 6-step ## How to configure a run mirroring infer_image-only, with a GPU-memory presets table (output_size + AE sliding-window knobs by GPU) drawn from the shipped config_infer_<XXg>_<dim>.json files. End-to-end Path B chest CT example.

• download-models.md: section name standardization (Code reference → Related scripts) and cross-reference style brought in line with the other skills ([name](name.md)).

[greptile-apps]

Greptile Summary

This PR restructures the skills/ documentation from algorithm-walkthrough style to a uniform operator-focused template (command → configuration steps → output → related scripts → related skills). infer_mask-generation.md is renamed to infer_mask-only.md to parallel infer_image-only.md, and all four inference skills receive new 6-step "How to configure a run" sections that mirror each other for easy cross-comparison.

• docs/inference.md corrects the config-key names to cfg_guidance_scale (fixing the misleading cfg_guidance_scale_modality / cfg_guidance_scale_tumor names flagged in previous review threads), and all four inference skills now use the correct key throughout.
• infer_mask-image-paired.md gains an accurate GPU-memory presets table (verified against the shipped config_infer_<XXg>_<dim>.json files) and a concrete Path-B chest-CT end-to-end example.
• download-models.md standardises section naming and converts plain-text skill cross-links to relative Markdown links.

Confidence Score: 5/5

Documentation-only restructuring; no executable code changed. All factual claims verified against source files.

Every claim introduced by this PR was cross-checked against the codebase: the GPU-memory presets table matches all nine config_infer files exactly, the corrected cfg_guidance_scale key name is confirmed by diff_model_infer.py line 297, the AE hardcoded values (roi_size=[80,80,80], overlap=0.4) are confirmed in diff_model_infer.py, and the find_masks / find_closest_masks attributions match their respective source files.

No files require special attention. The GPU-memory presets table in skills/infer_mask-image-paired.md is the highest-stakes new content, and it verified cleanly against the shipped config files.

Important Files Changed

Filename	Overview
docs/inference.md	Corrects two config-key names in the reference table: cfg_guidance_scale_modality → cfg_guidance_scale (in config_maisi_diff_model_*.json) and cfg_guidance_scale_tumor → cfg_guidance_scale (in config_infer.json), with a disambiguation note that the same key name has different semantics depending on which script reads it.
skills/download-models.md	Section rename Code reference → Related scripts, table reformatted from symbol-oriented to script-oriented rows, and infer_mask-generation cross-link updated to infer_mask-only. Minimal change, no issues.
skills/infer_image-from-mask.md	Rewritten as a CLI operator guide. Internal function-args table and algorithm step-by-step prose replaced with command block, mask format table, Options A/B preprocessing steps, a worked example, and related-scripts table. Content accuracy verified.
skills/infer_image-only.md	Restructured around a 6-step How to configure a run section. Adds T1-axial whole-brain worked example, replaces cfg_guidance_scale_modality with the correct cfg_guidance_scale, and accurately notes AE knobs are hardcoded in diff_model_infer.py.
skills/infer_mask-generation.md	File deleted — content migrated to the new infer_mask-only.md in operator-focused form. All cross-links in other skills have been updated accordingly.
skills/infer_mask-image-paired.md	Gains a 6-step How to configure a run section, a GPU-memory presets table (all 9 rows verified against configs/config_infer__.json), and a Path-B chest-CT end-to-end example. cfg_guidance_scale_tumor corrected to cfg_guidance_scale.
skills/infer_mask-only.md	New file replacing infer_mask-generation.md. Covers Path A/B dispatch, the 10-slot anatomy_size vector, output format, and related scripts. find_masks correctly attributed to scripts/find_masks.py; find_closest_masks correctly attributed to LDMSampler in scripts/sample.py.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A[User: pick pipeline] --> B{Need to generate mask?}
    B -- "No (have mask already)" --> C[infer_image-from-mask\nscripts/infer_image_from_mask.py\nCT-only]
    B -- "No (unconditional image)" --> D[infer_image-only\nscripts/diff_model_infer.py\nCT or MR]
    B -- "Yes (paired output)" --> E[infer_mask-image-paired\nscripts/inference.py\nCT-only]
    E --> F{controllable_anatomy_size\nnon-empty?}
    F -- "Path A: Yes" --> G[Mask DM DDPM\nscripts/sample_mask.py]
    F -- "Path B: No" --> H[DB lookup\nfind_masks\nscripts/find_masks.py]
    G --> I[MAISI 132-class mask]
    H --> I
    I --> J[Image LDM + ControlNet\nscripts/infer_image_from_mask.py]
    C --> J
    J --> K[Quality check CT\nLDMSampler.quality_check_ct]
    K -- Pass --> L[sample_image.nii.gz + sample_label.nii.gz]
    K -- Fail --> J
    D --> M[unet_3d_seed_modality.nii.gz]

Loading

_{Reviews (5): Last reviewed commit: "Trim tumor-CFG bullet to satisfy markdow..." | Re-trigger Greptile}