ADK Audit / Site

Standalone task skill under the adk-audit category router. Inspects a deployed website across multiple dimensions and produces one consolidated report with severity-tiered findings, each anchored to a URL or DOM selector.

When to use

• Pre-launch or post-launch audit of a public website or web app.
• Periodic accessibility / performance / SEO check.
• Comparison audit (this URL vs. a target URL).
• Deliverable is a markdown report at .temp/reports/<slug>.md.

When NOT to use

• Code-repo audit -> adk-audit-repo
• Single PR review -> adk-review-pr
• Doc-only review -> adk-docs-review
• Building or fixing UI issues -> adk-frontend-feature / adk-frontend-design

Inputs

Input	Required	Notes
<url>	yes	Public starting URL (or list of URLs)
<dimensions>	optional	Subset of: performance, accessibility, seo, ux, security-headers (default: all)
<depth>	optional	quick (1 page) / standard (default; up to 5 pages) / deep (crawl + Core Web Vitals across templates)
<viewport>	optional	mobile / desktop / both (default both)
<output path>	optional	Defaults to .temp/reports/audit-site-<slug>-<date>.md
--auto	optional	Skip approval gates

Workflow

1. Confirm intent - restate URL(s), dimensions, depth, viewports, output. Approval gate unless --auto.
2. Inventory - capture: site type (marketing / app / docs), framework hints (from headers / source), key user flows, third-party scripts, page tree to depth.
3. Run dimensions in parallel - each dimension produces its own findings list:
   • Performance: Core Web Vitals (LCP, INP, CLS), TTFB, transfer size, render-blocking resources, image strategy.
   • Accessibility: WCAG 2.2 AA - landmarks, headings hierarchy, alt text, color contrast, focus management, ARIA misuse, keyboard traps.
   • SEO: title/meta, canonical, robots, sitemap, OG/Twitter cards, structured data, internal linking, crawlability.
   • UX: layout at 360 / 768 / 1280 viewports, primary CTA visibility, form usability, error states, copy clarity.
   • Security headers: HSTS, CSP, X-Content-Type-Options, X-Frame-Options, Referrer-Policy, Permissions-Policy.
4. Aggregate - merge findings; deduplicate; group by dimension under each severity.
5. Validate - re-fetch the URL or re-run the analyzer to confirm each finding reproduces.
6. Report - findings-first markdown using the template below.

Severity ladder

Label	Site meaning
Blocker	Broken core flow, security header missing entirely, accessibility failure that blocks use, page does not load
Critical	Significant CWV failure, severe a11y / SEO regression, missing meta on the home page
Should Have	Notable improvement (image optimization, missing alt text in non-critical regions)
May Have	Minor polish
Nitpick	Style only
Question	Auditor unsure (e.g. intentional A/B test variant)

Finding template

### [<Severity>] <One-line summary> (<dimension>)
- **URL**: <url>
- **Anchor**: <DOM selector / region / response header>
- **Issue**: <2-3 sentence explanation>
- **Evidence**: <measurement / quoted source / screenshot description>
- **Suggested fix**: <concrete recommendation; route to `adk-frontend-*` / `adk-build-*` if implementation needed>
- **Why this severity**: <one sentence>

Report template

# Site Audit: <site>

## Summary
- URLs audited: <list>
- Viewports: <list>
- Dimensions: <list>
- Findings: <N> Blocker, <N> Critical, <N> Should Have, <N> May Have, <N> Nitpick, <N> Question

## Top Risks
1. <one-line top risk>
2. <one-line top risk>

## Core Web Vitals (per page)
| URL | LCP | INP | CLS | TTFB | Transfer |
| --- | --- | --- | --- | --- | --- |

## Findings

### Blockers
<finding blocks>

### Critical
<finding blocks>

### Should Have
<finding blocks>

### May Have
<finding blocks>

### Nitpicks
<finding blocks>

### Questions
<finding blocks>

## Per-Dimension Notes
<short per-dimension narrative for context the findings cannot carry>

## Out of Scope
- <items not audited and why (e.g. authenticated flows, third-party iframes)>

## Recommended Next Steps
1. <fix Blockers via `adk-frontend-feature` / `adk-build-feature`>
2. <follow-up audit in <area> after fixes>

Tooling notes

This skill calls whatever site-audit tooling the environment provides (browser MCP for DOM / a11y inspection, Lighthouse-like perf checks, header probes via curl -I). When a tool is unavailable, the skill explicitly marks the dimension as not-measured rather than guessing.

Tools the skill may use when present:

• Headless browser MCP (page load, DOM snapshot, axe-core, screenshots).
• HTTP probe (curl -I, httpx) for headers.
• HTML parser for meta, link, structured data.
• Public Lighthouse / PageSpeed API for perf metrics.

Anti-patterns

• Findings without URL + selector / header anchor.
• Reporting Lighthouse scores without per-metric values - scores hide root cause.
• Mixing fixes into the audit; the audit reports.
• Auditing only one viewport when the site is responsive.
• Calling something a Blocker that is intentional (e.g. a maintenance page).
• Relying on cached results from a previous run; always re-fetch.

Examples

adk-audit-site https://example.com --dimensions performance,accessibility --viewport mobile

adk-audit-site https://shop.example.com --depth deep --output .temp/reports/audit-site-shop-2026-04.md

Clarifying questions (default-ask)

When running without --auto, the skill asks these questions in order, one at a time. Under --auto, the skill picks the safest option for each (see references/clarifying-questions.md) and reports the choices.

1. Which URL(s) to audit? Single page, top N pages, or full crawl? — How to pick: Single = focused investigation. Top N = sample (home + 5-10 high-traffic pages). Full crawl = comprehensive but slow; only with explicit approval.
2. Dimensions: performance / accessibility / SEO / UX / security-headers / all? — How to pick: All by default. Narrow when retesting a specific dimension.
3. Devices and connection profiles? — How to pick: Default = desktop + mobile, fast 3G + cable. Match real user distribution if known.

Default report: Per-page Lighthouse score table + severity-grouped findings + screenshot references.

Detailed report (on request or --verbose): Add: full Lighthouse JSON archived, axe full report, network waterfall analysis, page-weight breakdown, contrast-ratio table for any flagged colors.

Artifact: site-audit-report — Markdown report + screenshots/artifacts in .temp/notes/audit-site-<slug>/.

Artifact path: .temp/reports/audit-site-<host>-<date>.md (raw lighthouse JSON + screenshots in .temp/notes/audit-site-<slug>/)

Clarifying questions (default-ask)

When running without --auto, the skill asks these questions in order, one at a time. Under --auto, the skill picks the safest option for each (see references/clarifying-questions.md) and reports the choices.

1. Which URL(s) to audit? Single page, top N pages, or full crawl? — How to pick: Single = focused investigation. Top N = sample (home + 5-10 high-traffic pages). Full crawl = comprehensive but slow; only with explicit approval.
2. Dimensions: performance / accessibility / SEO / UX / security-headers / all? — How to pick: All by default. Narrow when retesting a specific dimension.
3. Devices and connection profiles? — How to pick: Default = desktop + mobile, fast 3G + cable. Match real user distribution if known.

Default vs detailed output

Default report: Per-page Lighthouse score table + severity-grouped findings + screenshot references.

Detailed report (on request or --verbose): Add: full Lighthouse JSON archived, axe full report, network waterfall analysis, page-weight breakdown, contrast-ratio table for any flagged colors.

Artifact: site-audit-report — Markdown report + screenshots/artifacts in .temp/notes/audit-site-<slug>/.

Artifact path: .temp/reports/audit-site-<host>-<date>.md (raw lighthouse JSON + screenshots in .temp/notes/audit-site-<slug>/)

<!-- adk:references:start -->

References shipped with this skill

These files live in references/ next to this SKILL.md. Read them when the skill activates; they are inlined here so the skill is fully self-contained (no cross-skill or shared sources).

File	Purpose
references/anti-patterns.md	Things to avoid when running this skill.
references/artifact-format.md	The deliverable's format and where it lives (.temp/ contract).
references/clarifying-questions.md	The default-ask questions for this skill, with how-to-pick rubrics.
references/constitution.md	Non-negotiable rules and working/communication discipline.
references/interaction-contract.md	Default-ask, explained-options, --auto contract every skill must follow.
references/output-format.md	Default vs detailed report shapes; severity labels; verbosity rules.
references/persona.md	The agent persona that drives this skill.
references/research-protocol.md	Source ordering, stop conditions, evidence buckets, citation discipline.
references/review-comment-format.md	Standard finding format with stable IDs and severities.
references/working-artifacts.md	Legacy: superseded by artifact-format.md; kept for back-compat.

<!-- adk:references:end -->