accessibility

Router skill. Resolve topic, then Read EXACTLY ONE playbook from routes:. Never load the whole cluster — loading two playbooks is the primary anti-pattern. [DOC]

When to use

Any WCAG / a11y request: contrast and color, keyboard and focus, screen-reader and ARIA, alt text, captions, plain-language copy, or compliance evidence. [DOC]

Topic routing (pick one) [DOC]

• audit → assess an existing artifact against WCAG 2.1 AA; produce a gap list.
• design → bake a11y into components/tokens/layout before build.
• testing → automated (axe-core) + manual keyboard/SR verification in CI.
• writing → alt text, link text, headings, plain-language microcopy. Overlap: pre-build → design; post-build verification → testing; one-time scorecard → audit. If the request spans two, run them in sequence, not merged. [INFERENCIA]

Depth

• quick (default) → essentials, the highest-impact WCAG criteria only.
• deep → apply the playbook exhaustively with verification at each step.

Inputs / Outputs

• In: target artifact (URL, component, copy, repo path) + topic + depth. [DOC]
• Out: findings with WCAG success-criterion refs + remediation, each tagged. [DOC]
• Missing target → ask once; do not audit a hypothetical. [INFERENCIA]

Validation gate (done means) [DOC]

• Exactly one playbook loaded; topic matches user intent.
• Every finding cites a WCAG criterion (e.g. 1.4.3) and a concrete fix.
• Claims carry an Alfa-core tag ([CODE]/[CONFIG]/[DOC]/[INFERENCIA]/[SUPUESTO]); pass/fail is evidenced, never asserted green-as-success.

Spine: Discover → Analyze → Execute → Validate. Gates: constitution v6.0.0 (enforcement), evidence tags, script-first rule. [CONFIG] Gate against assets/checklist.md and score with assets/quality-rubric.json before declaring done; see the DoD bundle in assets/. [DOC]