accessibility

The **first accessibility specialization** in the library (verified: none existed anywhere under library/). It carries a real product through a full agentic WCAG audit-to-conformance lifecycle with routed human approvals on every action that commits engineering effort or leaves the org boundary. It **consolidates the previously scattered near-misses** by referencing them as seeds rather than duplicating their content, and **supersedes** the pre-bar web-development/accessibility-audit-remediation.js (now @deprecated).

Flagship process: wcag-audit-remediation

wcag-audit-remediation.js (@process accessibility/wcag-audit-remediation) walks one target and its surfaces from scope to a published conformance statement:

**Inputs:** { target: { name, surfaces: [{ id, kind:'page'|'component', url?, path?, description? }] (required, non-empty), assistiveTech?: string[] } (required), wcagLevel?='AA', repoRoot?='.', artifactsDir?, maxFixAttempts?=2, kipEnabled?=true, kipDir?='.a5c/kip', kipModel?='sonnet' }

**Outputs:** { success, scope, priorKnowledge, audits, findingsRegister, remediationPlan, remediation, conformanceGate, regressionGuard, conformanceStatement, gatedActions, kipFactsAsserted, artifacts, metadata } — success = conformanceGate.passed && gatedActions.remediationPlanSignoff.approved && (statement not required OR gatedActions.conformanceStatementPublish.approved === gatedActions.conformanceStatementPublish.executed) && regressionGuard.captured.

Policy-gated actions

All approvals go through routedBreakpoint; for both policy-gated actions the breakpointId **equals** the actionId and tags are ['policy-gated','accessibility']. Fail-closed: the executor task runs **only** on approved === true — a rejection is honored, recorded, and never worked around.

Additional (non-gated) breakpoint on the surface:

adversarialGate combinator on fix-budget exhaustion (expert owner, combinator-fixed); the process does not re-declare it, but operators should know it is part of the surface.

• accessibility.conformance-reaudit.gate-escalation — raised **internally** by the

outputs.gatedActions records **every** decision — { actionId, required, approved, autoApproved, response, executed } per action, including non-interactive auto-approvals (recorded raw from the BreakpointResult) and skipped / not-required gates ({ required:false, approved:false, autoApproved:false, response:null, executed:false } — never omitted). metadata.breakpointsHit logs every raised breakpointId in order.

Quality bar

critics must RE-RUN axe on each remediated surface and re-perform the manual keyboard/AT checks; file-read / diff citations alone do not satisfy the gate, and passed:true with an empty evidence array is a protocol failure enforced by the combinator (IRON LAW: at least one executed axe re-scan per remediated surface).

assistive-tech-critic) are distinct from the implementer (accessibility-remediation-engineer); the combinator fans critics out in parallel — no self-review.

maxFixAttempts rounds between critic rounds, then the combinator escalates to the owner.

unscannable surface records executed:false with a reason (never a fabricated clean pass); a rejected remediation-plan-signoff ends the run with success:false before any fix work; a failed conformance re-audit gate ends the run before any statement is drafted or published; policy-gated executors run strictly on approved===true.

• **adversarialGate with EXECUTED axe re-scan evidence** — the conformance-reaudit gate's
• **Reviewer independence** — critic agent names (wcag-conformance-critic,
• **Bounded fix loop** — the built-in gateFixerTask edits the remediation for up to
• **No fallbacks** — a target with an empty/absent surfaces[] throws before any work; an

kip integration

kipRecall at scope (topic: known WCAG issues for the target across its surfaces, kind accessibility) and kipAssert at close (per-surface conformance outcome, register summary, remediation outcome, gate outcome, conformance-statement decision), per shared/skills/kip-librarian. An empty store is a fresh brain, never an error; assert failures are reported by the librarian task, never swallowed.

Near-miss consolidation map

This specialization consolidates the previously scattered accessibility work. The seeds stay in place and are **referenced, not duplicated**; the pre-bar web-development process is **superseded**.

Usage

babysitter run:create \
  --process library/specializations/accessibility/wcag-audit-remediation.js \
  --inputs '{
    "target": {
      "name": "checkout-flow",
      "surfaces": [
        { "id": "cart-page", "kind": "page", "url": "https://app.example.com/cart" },
        { "id": "payment-form", "kind": "component", "path": "src/components/PaymentForm.tsx" }
      ],
      "assistiveTech": ["VoiceOver", "NVDA", "keyboard-only"]
    },
    "wcagLevel": "AA",
    "maxFixAttempts": 2
  }'

Files

acc.* Style-A agent tasks + orchestration).

— routedBreakpoint, adversarialGate, kipRecall, kipAssert, gateFixerTask.

`../ux-ui-design/accessibility-audit.js`, `../qa-testing-automation/accessibility-testing.js`.

• `wcag-audit-remediation.js` — the flagship process (9
• Combinators: `../common-utilities/routed-gate-combinators.js`
• Seeds (referenced, not duplicated): `../ux-ui-design/wcag-compliance.js`,