Color accessibility expert

Overview

Measure contrast, compute hue-preserving replacements, apply them to source, and re-audit the actual result. Fixed source files are the primary deliverable. A generated docs/PALETTE_CONTRAST_AUDIT.md documents current evidence rather than replacing the fix.

Ownership boundary

• Edit target-repository color values through scripts/apply_color_fixes.py after a dry-run.
• Write or refresh docs/PALETTE_CONTRAST_AUDIT.md only through scripts/generate_palette_audit.py.
• Treat the shared docs/COLOR_CONTRAST_ACCESSIBILITY.md method guide as read-only authority.
• Ground every documented path and color in files inspected and extraction output produced during the current run.

Project shape and routing

Classify improve-existing versus greenfield work with references/project_workflow.md. Classify one color pair, one source, or one image with references/task_selection.md. Route observed symptoms through references/topic_index.md.

Read references/tooling.md for script flags, input and output formats, dependencies, and document-generation behavior before running the toolchain.

Workflow

1. Run scripts/extract_colors.py over the requested target paths and retain its path, line, and hex evidence.
2. Run scripts/audit_palette.py against the required background and contrast ratio. Capture the measured failures and replacement mapping.
3. Preview the mapping with scripts/apply_color_fixes.py in its default dry-run mode, then apply it with --write when source edits are authorized.
4. Re-run extraction and the same palette audit over the same paths. Continue correcting measured failures until every in-scope color passes. This re-audit is mandatory evidence for a completed fix.
5. When rendered output matters, use scripts/image_contrast.py for explicit pixel points or dominant-color inspection.
6. When the target needs a new palette, use scripts/generate_color_wheel.py with its documented dependencies and audit output.
7. Generate or refresh docs/PALETTE_CONTRAST_AUDIT.md from the current run. Confirm the printed evidence manifest and that colors_documented matches the audit. A colorless repository carries no generated audit file.
8. Record changed files, replacement values, audit evidence, and verification in the target repository's changelog when its rules require that update.

Evidence contract

Every path and color in the generated audit table must come from a target-repository file read in the current run, and every color must occur in the current extract_colors.py output. Follow references/palette_audit_template.md before accepting the generated document.

Use references/testing_and_oracles.md for numeric proof and references/color_contrast_reference.md for WCAG thresholds, luminance, formulas, backward solving, and external calculators.

Completion

Return the files and color values changed, before-and-after ratios, the exact re-audit result, image spot-check results when applicable, and the generated audit evidence manifest. Report a fix only after the post-edit audit passes every in-scope color.