# Goodboy008 Labelize Labelize > ZPL Diff Auto-Fix Skill - Skill: `tomevault-io/goodboy008-labelize-labelize` (Agent Skill, multi-file: 2 files) - Install (CLI): `npx skillmds@latest add tomevault-io/goodboy008-labelize-labelize` - Raw SKILL.md: https://api.skillmd.com/api/skills/tomevault-io/goodboy008-labelize-labelize/raw - Safety review: pending (external: skill-scanner PASS, skillspector PASS) - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: Coding & Dev Tools - Author: tomevault-io (https://skillmd.com/u/tomevault-io) - Updated: 2026-09-17 - Page: https://skillmd.com/skills/tomevault-io/goodboy008-labelize-labelize --- # ZPL Diff Auto-Fix Skill ## Purpose Automatically detect labels with high rendering diff percentages, isolate the problematic ZPL elements into standalone test files for focused debugging, and iteratively fix the rendering logic until the diff drops below a target threshold (default 1%). ## Critical Rule — Fix Rendering Code, Never Tests **DO NOT** modify diff tests, golden test infrastructure, reference images, tolerance thresholds (to hide regressions), or test comparison logic to make diffs pass. The goal is to improve rendering accuracy, not to weaken the test harness. **DO** fix the root cause in rendering code and logic: - Parsers: `src/parsers/` - Elements: `src/elements/` - Barcode encoders: `src/barcodes/` - Renderer / drawing: `src/drawers/` - Font handling: `src/assets/` Threshold updates in `docs/DIFF_THRESHOLDS.md` are only allowed **after** a rendering fix genuinely lowers the diff percentage — set the new tolerance slightly above the new measured diff, never raise it to paper over a regression. ## When to Use - A label has a diff percentage higher than desired - You want to systematically reduce rendering differences across all labels - You need to extract standalone ZPL snippets to isolate a rendering bug - After implementing a rendering change, you want to verify and fine-tune it ## Prerequisites Before starting, ensure: 1. The diff report is up to date: `cargo test --test e2e_diff_report -- --nocapture` 2. All current tests pass: `cargo test` 3. You have reference PNG images in `testdata/` for the target label(s) ## Procedure ### Phase 1: Scan — Identify High-Diff Labels 1. **Read the diff report** from [testdata/diffs/diff_report.txt](../../testdata/diffs/diff_report.txt) 2. **Read the tolerance thresholds** from [docs/DIFF_THRESHOLDS.md](../../docs/DIFF_THRESHOLDS.md) 3. **Identify target labels** — either a specific label from the user's request, or all labels with diff > 1%: - List labels sorted by diff percentage (highest first) - For each, note the current diff%, tolerance, and primary diff source 4. **Pick the highest-impact label** to fix first (highest diff% that is potentially fixable — skip MaxiCode/PDF417 unless specifically requested) ### Phase 2: Analyze — Find Problematic Elements 5. **Read the ZPL source** from `testdata/labels/.zpl` or `testdata/unit/.zpl` 6. **Parse the ZPL** to identify all commands and elements: ``` Use src/skill/snippet_extractor.rs → split_zpl_commands() to list all commands Use src/skill/snippet_extractor.rs → group_commands_into_spans() to group into element spans ``` 7. **Inspect the diff image** at `testdata/diffs/_diff.png` — red pixels show where rendering differs from Labelary 8. **Correlate diff regions with elements** using `src/skill/element_analyzer.rs`: - Parse the ZPL to get `LabelInfo` - Compute bounding boxes for each element via `compute_element_bbox()` - Load the diff image and count red pixels in each bbox via `correlate_diff_regions()` - Report which elements contribute most to the total diff 9. **Rank elements by diff contribution** — focus on the top 3-5 contributors 9b. **Classify each element's diff** as ContentDiff, PositionDiff, or Mixed using `src/skill/diff_classifier.rs → classify_element_diffs()`: ``` // Full pipeline with classification: Use element_analyzer::analyze_label_with_classification(label, diff_path, zpl, use_labelary) ``` The classifier renders each element in isolation, optionally fetches the Labelary reference for the same snippet, and compares the two renders: | Isolated snippet diff | Full-label bbox diff | Classification | |----------------------|----------------------|----------------| | Low (< 2%) | High (> 2%) | **PositionDiff** — element placed wrong | | High (> 2%) | Any | **ContentDiff** — element renders wrong | | High (> 2%) | High (> 2%) | **Mixed** — both | | Low (< 2%) | Low (< 2%) | No classification (no meaningful diff) | For **PositionDiff** elements, the classifier also attempts to detect the offset vector (dx, dy): - Shadow detection: two clusters of red pixels → direct offset measurement - Centroid shift: centroid of diff pixels vs. expected element center ### Phase 3: Extract — Create Standalone Snippets 10. **Extract each high-diff element** into a standalone ZPL file: ``` Use src/skill/snippet_extractor.rs → extract_element() ``` This creates `testdata/unit/