This guide gives a new user one short path through the package without requiring knowledge of its internal schemas or agent graph. このガイドは、内部スキーマやエージェント構成を先に理解しなくても、最初の1回を実行できるようにするものです。
| Goal / 目的 | Entry / 入口 | Result / 成果物 |
|---|---|---|
| Find machine-detectable Web barriers / Webの機械検出可能な問題を調べる | accessibility-audit scan-web |
Full internal scan and compact AI context / 内部用scanとAI用context |
| Find likely barriers and improvement ideas / 問題候補と改善案を知る | Natural-language skill or reviewer agent / 自然言語のスキル・reviewer | Review summary and human follow-up / レビュー要約と人による追加確認 |
| Create a standards ledger / 規格台帳を作る | accessibility-audit assessment |
JSON with every profile row initialized as not_tested |
| Preserve a multi-step audit trail / 監査工程を追跡する | init → artifacts → merge → report |
Immutable run chain, merged assessment, report |
Before choosing a longer flow, inspect the installed version, active profiles, requirement catalog, and optional browser capability. The commands are read-only and return machine-readable output where appropriate. 長い監査フローへ進む前に、導入版、active profile、条項カタログ、任意のbrowser capabilityをread-onlyで確認できます。
accessibility-audit --version
accessibility-audit profiles list
accessibility-audit requirements show 1.1.1 --profile web-modern
accessibility-audit requirements search "フォーカス" --profile web-modern
accessibility-audit requirements search "focus" --profile web-modern --level AA
accessibility-audit doctorFor complete, reproducible examples of Natural-language review, a Standalone ledger, and a Run-backed audit, see the runnable examples. 自然言語レビュー、standalone台帳、run-backed監査を最初から最後まで再現する場合は、実行可能なexamplesを参照してください。
The unified CLI manages records and validation and now also exposes a rule-based browser scan. The browser capability requires the exact supported Playwright and axe-core versions. An actual screen-reader session remains an external human or host capability.
Run from the repository root. The example creates a complete WCAG 2.2 A/AA ledger whose rows remain explicitly untested, validates it, and renders a guarded profile-aware report. Internal artifacts are written below the ignored audit-runs/ directory so they are not mixed with source files.
node .\codex\skills\information-accessibility-practice\scripts\accessibility-audit.mjs assessment --profile web-modern --target-name "Example Site" --target-version "2026-08-24" --target-ref "https://example.com/" --evaluator "Accessibility Reviewer" --evaluated-at "2026-08-24" --output .\audit-runs\quickstart\audit.json
node .\codex\skills\information-accessibility-practice\scripts\accessibility-audit.mjs validate-assessment .\audit-runs\quickstart\audit.json
node .\codex\skills\information-accessibility-practice\scripts\accessibility-audit.mjs report --input .\audit-runs\quickstart\audit.json --locale ja --detail full --visibility internal --output .\audit-runs\quickstart\audit-report.mdExpected files / 生成物:
audit-runs/quickstart/audit.json: complete profile ledger; initialization is not completed inspection / 完全な規格台帳。初期化は検査完了ではないaudit-runs/quickstart/audit-report.md: profile-aware report showing every criterion as未確認／Not testeduntil target-specific evidence and judgement are added
The report displays criterion number, localized title, level, profile group, primary source, judgement source, evidence level, and rationale. Use --locale ja or --locale en for human-readable text; requirement IDs and enum values do not change.
レポートには条項番号、名称、レベル、profile group、一次資料、判定の出所、証拠レベル、根拠が表示されます。--locale jaと--locale enは人向け表示だけを変更し、内部IDやenumは変更しません。
The claim section separately shows the requested tier and the validator maximum tier. It uses only fixed wording registered in standards-registry.json; report judgement labels are not a formal conformance declaration.
主張可能な範囲では、要求されたtierと検証上限tierを分離し、registryに登録された固定表現だけを表示します。レポートの判定語だけで正式な適合表明を行うことはできません。
The generator and report writer use exclusive safe outputs. Existing files are not overwritten, and missing parent directories are created through the guarded output path.
Install the optional browser capability:
npm install --no-save --package-lock=false playwright@1.62.1 axe-core@4.13.0
npx playwright@1.62.1 install chromiumThen run:
node .\codex\skills\information-accessibility-practice\scripts\accessibility-audit.mjs scan-web --url "https://example.com/" --profile web-modern --output .\audit-runs\quickstart\automated-scan.json --context-output .\audit-runs\quickstart\automated-scan.context.jsonUse automated-scan.context.json as the first AI input. It contains machine violations, unresolved review candidates, coverage, focus, and reflow summaries without the raw DOM or full accessibility tree. Open automated-scan.json only when the compact item needs its bounded HTML, selector, frame, or raw evidence context.
See web-inspection.md for network, interaction, output, and dependency details.
Use a run when you need immutable target metadata, explicit permissions, registered evidence artifacts, external human review, remediation provenance, or retest history.
- Agree on
quickordetailed, intended report use, scope and environments before inspecting. Initialize the run with--inspection-mode quick|detailed --inspection-purpose <purpose>, a dedicated artifact root and explicit permissions. - Create the baseline assessment for the same target and profile.
- Materialize and register screening, queue, human-review, and remediation artifacts as applicable.
- Merge only registered artifacts into a new assessment.
- Validate and render the guarded report. Check each requested completion criterion against actual evidence and delivered sections; identify any unmet conditions and next steps.
Run-backed reporting uses the same profile title, criterion metadata, group counts, locale contract, and claim section as standalone reporting. Each row distinguishes external human review, AI/automated screening projection, and a check that was not run. Screening projections remain report-only judgements and never become human_verified profile outcomes.
run-backedレポートもstandaloneと同じ表示規則を使い、各行で外部人手レビュー、AI／自動スクリーニング、未実施を区別します。screening projectionはreport-only judgementであり、human_verifiedへ自動昇格しません。
node .\codex\skills\information-accessibility-practice\scripts\accessibility-audit.mjs report --run .\audit-runs\example\audit-run.json --assessment .\audit-runs\example\merged-assessment.json --locale en --detail summary --visibility internal --output .\audit-runs\example\audit-report.en.md --appendix .\audit-runs\example\audit-report.full.en.mdThe complete control-plane contract is documented in:
agent-orchestration.mdstandards-assessment.mddevelopment-web-audit-request.template.mdarchitecture-and-glossary.mdreporting.md
Choose --detail summary for a decision-ready view or --detail full for the complete 55/56-row report. --detail summary --appendix <path> writes both from the same validated presentation model.
--detail summaryは意思決定向けの短い表示、--detail fullは55／56件の完全版です。--appendixを併用すると同じ検証済みmodelから両方を生成します。
--visibility internal preserves raw audit information and explicitly marks the output as not publication-ready. Public output requires a reviewer-disclosure choice and a separate internal redaction manifest.
--visibility internalは監査情報を保持し、公開用ではないことを明示します。公開向け出力では、確認者情報の扱いと内部用redaction manifestを必須指定します。
node .\codex\skills\information-accessibility-practice\scripts\accessibility-audit.mjs report --input .\audit-runs\quickstart\audit.json --locale ja --detail summary --visibility public --reviewer-disclosure redact --redaction-manifest .\audit-runs\quickstart\redactions.json --output .\audit-runs\quickstart\public-summary.md --appendix .\audit-runs\quickstart\public-full.mdAutomated redaction cannot detect every sensitive value. Public output always requires human publication review. The manifest records only the field path, reason, and action; it does not copy the removed value. 自動伏字ですべての機微情報を検出することはできません。公開向け出力には人による公開前確認が必要です。manifestにはfield path、理由、処理だけを記録し、削除した値は複製しません。
Keep each audit's run, evidence, assessment, internal reports, public candidates, and redaction manifests together under audit-runs/<run-id>/. This directory is Git-ignored, but still needs your usual access control and backup. To export a reviewed public report, rerun report --visibility public with a new --output path outside this repository and a new internal --redaction-manifest path; review the exported file before delivery. Outputs are never overwritten by the CLI.
run・証拠・assessment・内部レポート・公開候補・伏字manifestは監査ごとにaudit-runs/<run-id>/へまとめます。Gitの追跡対象外ですが、アクセス制御とバックアップは別途必要です。確認済みの公開用レポートを書き出す際は、report --visibility publicをリポジトリ外の新しい--outputと内部用の新しい--redaction-manifestで再実行し、書き出したファイルを納品前に確認します。CLIは既存ファイルを上書きしません。
Archive each run directory as one unit with its source version and review context. Set retention to the applicable project policy; there is no automatic cleanup. Before manually deleting a working copy, confirm the archive and preserve original, signed, or submitted evidence. runディレクトリはソース版とレビューの文脈を含めて一単位で保管します。保存期間は案件の方針で決めてください。自動削除はありません。作業用コピーを手動削除する前に保管先を確認し、原本・署名済み・提出済みの証拠は保持します。
See reporting.md for the complete contract and examples.
詳細な契約と例はreporting.mdを参照してください。
- AI and automated tools may create E0/E1 screening observations.
- A profile requirement becomes evaluated only through an external human review using the applicable procedure and target-specific evidence.
- Missing, uncertain, and not-applicable states remain visible; they are never converted to pass by omission.
- Raw browser and scanner outputs are internal audit material and should not be copied directly into a public report.