object-isolation-correction

name	object-isolation-correction
description	Wrap object-isolation correction as a repo-specific workflow. Use when a local image already exists, the current isolation result is imperfect, and the next step is to choose between ImageSorcery-first, imagegen-first, or a hybrid correction path with a bounded correction packet and worker run.

Object Isolation Correction

Overview

Use this skill when an object-isolation result needs repair, not when isolation is still undefined. This skill does not perform the final multimodal edit by itself. It standardizes the correction route first, then allows a bounded worker run that tries alpha split, ImageSorcery fallback, and an imagegen request artifact.

Use This Skill When

• a local source image already exists
• a current isolation result or crop candidate already exists, or the failure mode is visible
• the next step is to choose imagesorcery, imagegen, or hybrid
• a bounded correction packet should be written before retrying the edit

Do Not Use This Skill When

• images still need to be extracted from PPTX
• caption generation is the current task
• OCR-only evidence is the current task
• final metadata write-back or rename commit is the current task

Required Inputs

• one absolute --source-image
• optional --current-result
• one or more --issue
• one --output-md
• optional --output-json

Script

• scripts/prepare_object_isolation_correction_packet.py
• scripts/run_object_isolation_correction_worker.py

References

• knowledge_bases/object-isolation-correction-knowledge-base-at2026-03-27-23-10.md
• references/runtime.md
• references/troubleshooting.md

Workflow

1. Confirm the source image and current failure mode.
2. Build a correction packet before editing anything.
3. Run the worker in a bounded output directory.
4. Try alpha connected-components first when the source is already a transparent PNG with separated objects.
5. Default to imagesorcery-first for boundary, split, or target-selection problems once alpha split is insufficient.
6. Prefer imagegen-first for repair-like problems where the current cutout is visually damaged.
7. Use hybrid when deterministic masking is still useful but a model-assisted cleanup may be needed after it.
8. Keep the original image path stable and write corrected outputs to a bounded run surface.
9. Review the correction result before promoting it into a caption or OCR rerun.

Outputs

• one markdown correction packet
• optional JSON sidecar with the same route and prompt data
• one worker report
• one worker result JSON
• alpha component crops when alpha split is sufficient
• ImageSorcery fallback crops when alpha split is insufficient
• bounded imagegen request artifacts when the route still needs model-assisted cleanup

Not Owned Here

• PPTX extraction
• caption generation
• OCR extraction by itself
• final approval and commit
• direct built-in imagegen execution inside a plain local Python script