Accessibility Reviewer

The Accessibility Reviewer is an orchestrator agent that audits a codebase against accessibility skills, verifies the findings, and produces a consolidated conformance report. It delegates to specialized subagents and supports audit, diff, and plan modes.

A reviewer is only as useful as the findings it can defend. Every FAIL and PARTIAL finding passes through deep verification before it reaches the report.

Purpose

• Profile the codebase to determine which accessibility skills apply.
• Assess applicable skills (WCAG 2.2, ARIA APG, COGA, Section 508, EN 301 549) and collect findings.
• Verify FAIL and PARTIAL findings through a deep verifier before reporting.
• Generate a consolidated report with summary counts and a severity breakdown.

When to Use

Scenario	Recommended mode
Full review of an existing codebase	audit (default)
Review only the surfaces touched by a set of changes	diff
Evaluate an accessibility plan document before build	plan

Modes and Inputs

Mode	Behavior	Key inputs
audit	Profiles the full codebase, assesses applicable skills, and verifies findings	Optional path focus, skills override
diff	Assesses only the changed files, then verifies findings	Changed files
plan	Assesses a plan document and passes findings through without verification	Plan document

Additional inputs accepted across modes include an optional path focus, a skills-list override, a target skill (which fast-paths past the profiler), and a prior scan report.

The Pipeline

flowchart LR
  A["Setup"] --> B["Profile Codebase"]
  B --> C["Assess Applicable Skills"]
  C --> D["Verify Findings"]
  D --> E["Generate Report"]
  E --> F["Compute Summary"]

Step	Stage	What happens
1	Setup	Resolve mode, focus paths, and the skill set to assess
2	Profile Codebase	The Codebase Profiler determines which skills apply; a supplied target skill bypasses this step
3	Assess Applicable Skills	Each applicable skill produces findings; diff mode assesses only changed files, plan mode assesses the plan content
4	Verify Findings	Audit and diff modes run the Finding Deep Verifier per skill for FAIL and PARTIAL findings; plan mode skips verification
5	Generate Report	The Report Generator produces the consolidated report from verified findings
6	Compute Summary	A completion summary displays counts, assessed skills, the report path, and any excluded skills with reasons

Subagents

Subagent	Responsibility
Codebase Profiler	Determines which accessibility skills apply to the codebase
Accessibility Skill Assessor	Assesses a single skill and produces findings
Finding Deep Verifier	Re-verifies FAIL and PARTIAL findings in audit and diff modes
Report Generator	Produces the consolidated conformance report

If a subagent response is incomplete or malformed, the orchestrator retries once and then excludes that skill, recording the reason.

Outputs

The reviewer writes a consolidated report under .copilot-tracking/accessibility-reviews/{{YYYY-MM-DD}}/. Report filenames encode the review mode and a three-digit sequence:

• accessibility-report-{{NNN}}.md for audit mode
• accessibility-report-diff-{{NNN}}.md for diff mode
• accessibility-plan-risk-assessment-{{NNN}}.md for plan mode

The report includes summary counts, the list of assessed skills, a severity breakdown, and any excluded skills with reasons.

Cross-Planner Integration

The reviewer's deterministic signals can be shared with the Security Planner's profiler and vice versa when both run in the same session. See Cross-Planner Integration for the full matrix.

Related Files

File type	Location
Agent	.github/agents/accessibility/accessibility-reviewer.agent.md
Skills	.github/skills/accessibility/
Reports	.copilot-tracking/accessibility-reviews/

Next Steps

• Accessibility Planner for upstream framework selection and backlog handoff.
• Accessibility Planner Quickstart for a five-minute walkthrough.

🤖 Crafted with precision by ✨Copilot following brilliant human instruction, then carefully refined by our team of discerning human reviewers.