Skill v3.6.0

name: audit description: "Use when the user wants a complete, whole-site audit that combines search visibility, front-end defects, and design quality in one pass. Runs all 15 specialist sub-agents across SEO, accessibility/interaction/layout/code defects, and UX/visual/motion/content design, then merges them into one scored report with a prioritized action plan. Use for: 'audit my whole site', 'complete site audit', 'full website review', 'audit everything', 'is my site good', 'review my site end to end'. For a search-only audit use /seo audit; for defect-only use /inspect; for design-only use /siteasy audit." version: 3.6.0 user-invocable: true argument-hint: "[url] | full [url] [seo|defects|design|quick] | [verify|checks] [url] | [report|learnings] [file] | compare [A] [B]" allowed-tools:

• Read
• Write
• WebFetch
• Glob
• Grep
• Task
• Bash(node *)
• Bash(python3 *)

Complete-audit toolkit for websites. One pass that orchestrates the plugin's three other skills (/seo, /inspect, /siteasy), dispatches all 15 specialist sub-agents across search visibility, front-end defects, and design quality, then merges their scored sections into a single Site Health Score with a prioritized action plan. The audit skill owns no detection logic of its own. It schedules the existing sub-agents, shares one fetch across them, and consolidates the results.

Commands

Six commands, six references. The agent run scopes (full and its group scopes seo, defects, design, quick) share the orchestration playbook in references/full.md and differ only in which agent group is dispatched; the legacy first-token form (/audit seo [url]) remains accepted and routes to full with that scope. verify additionally re-runs the gating dimensions and reconciles them by majority vote. checks runs the deterministic pre-pass with no sub-agents and is documented in references/checks.md; it is also the ground-truth layer the agent modes consume in their fetch phase. compare diffs two targets (references/compare.md) and report formats an already-produced audit (references/report.md).

How to run a command

When the user invokes a command:

1. Read the matching reference file with the Read tool (full.md for the agent run modes, checks.md for the deterministic pre-pass, compare.md for compare, report.md for formatting).
2. Follow the instructions in that reference exactly. Do not improvise scoring weights or skip a dimension.
3. If no command is specified:

• With a URL, the bare invocation runs full against that URL.
• A bare scope token (seo, defects, design, quick) as the first argument runs full with that scope; legacy names in tools/data/intents.csv route to their canonical command.
• With no URL, ask the user for the site URL before dispatching any agent.

Relationship to the other skills

The three audit groups map one-to-one onto the plugin's three other skills and reuse their sub-agents directly. There is a single source of truth per dimension and no duplicated detection logic. The audit skill changes only the scheduling (parallel, shared fetch) and the consolidation (one merged report).

Because the agents are shared, a fix surfaced here can be re-run or deepened with the owning skill (for example /seo technical for a flagged crawl issue, or /siteasy clarify for flagged copy) without re-auditing the whole site.

Output

Each agent run mode produces two markdown files (see references/full.md for the templates) plus a machine-readable SITE-AUDIT.json:

• SITE-AUDIT-REPORT.md holds the full findings for every group that ran, with each agent's returned section embedded verbatim.
• SITE-ACTION-PLAN.md holds the consolidated, de-duplicated fix list ordered Critical, High, Medium, Low.
• SITE-AUDIT.json holds the machine-readable result (scores plus per-check verdicts plus a cost ledger) that powers compare, CI gating and score-over-time. The checks mode writes only this file. Schema: tools/audit/schema/site-audit.schema.json.

The overall Site Health Score weights Search Visibility at 35 percent, Front-end Defects at 35 percent, and Design Quality at 30 percent. Any critical accessibility or interaction defect caps the Defects group regardless of other passes. The exact weights and the cap rule live in references/full.md.

When NOT to use

A single-dimension request does not need all 15 agents. Routing it to the one owning skill is faster and cheaper. The audit skill is for the whole-site, cross-dimension pass.

Canonical thresholds

L-DATA-2 in tools/data/laws.csv governs when this skill refuses to produce a number: a dimension with fewer than half its inputs available reports insufficient data instead of a score, and two such dimensions suppress the overall score. See references/full.md.