transparent-component-triage

name	transparent-component-triage
description	Triage a bounded batch of PPT-extracted images to find the small subset that is mechanically sufficient for alpha-only component splitting. Use when conservative preprocessing is required and only review-gated deterministic candidates should advance beyond the full-image baseline.

Transparent Component Triage

Overview

Use this skill to run a conservative batch pass over PPT-extracted media and answer one narrow question:

• which files are already good enough for deterministic alpha-only splitting
• which files should stay on the full-image baseline

This skill wraps the existing alpha-only batch classifier and keeps the result as a review-gated candidate subset, not an automatic promotion surface.

Use This Skill When

• the source set is a bounded PPT media batch under control/project_domain/resources/pptx_jobs/
• automatic object isolation is still considered unsafe
• a deterministic prefilter is needed before any reviewed component selection
• the goal is to separate alpha_split_sufficient from the rest

Do Not Use This Skill When

• the task is one-image component review
• semantic object selection is the current task
• imagegen correction is the current task
• full-image caption rerun is the current task
• the source set is not a bounded PPT media extraction surface

Required Inputs

• optional --input-root
• optional --output-root
• optional --manifest-jsonl
• optional --summary-json
• optional --report-md
• optional --limit

Script

• scripts/classify_alpha_split_batch.py

References

• knowledge_bases/transparent-component-triage-knowledge-base-at2026-03-28-01-20.md
• references/runtime.md
• references/troubleshooting.md

Workflow

1. Confirm the batch is a conservative preprocessing candidate set.
2. Run the skill-local wrapper, not a handwritten worker loop.
3. Let the shared classifier call the correction worker in alpha-only mode.
4. Read the markdown report and summary JSON before using any subset downstream.
5. Treat alpha_split_sufficient as review-gated only.
6. Keep every other file on the full-image baseline unless a later reviewed gate says otherwise.

Preferred Output Surface

• report: control/project_domain/resources/reports/
• summary: control/project_domain/resources/manifests/
• manifest JSONL: control/project_domain/resources/manifests/
• per-image worker outputs: control/project_domain/archive/object_isolation/alpha_split_batch/

Outputs

• one markdown report
• one summary JSON
• one manifest JSONL
• per-image worker result directories

Known Good Fit

• transparent PNG-heavy PPT extract batches
• phase0 preprocessing where the default baseline must remain conservative
• cases where deterministic alpha-connectivity is useful as a review prefilter only

Not Owned Here

• semantic object selection
• component-level OCR review of a single image
• imagegen correction
• object-level caption reruns
• automatic downstream promotion without review