onsite-auditlisted

# On-page audit (read-only) 1. Read profile if a brain repo exists (optional - this skill also works bare). 2. Public checks per URL (WebFetch): title (length, keyword presence), meta description (present, and flagged when over ~155 chars - Google truncates around there), H1 count, heading structure, canonical, robots meta, image alts, internal links out, JSON-LD present/valid, answer-capsule presence in the first 200 words, server-rendered content check. 3. Page essentials dimension - runs for each audited page, and site-wide where a check says so. Every check emits the standard falsifiable signal line (observation | why it matters | "we are wrong if..." | leading indicator) and, where a fix is actionable, feeds onsite-propose. Evidence discipline throughout (see plugin/docs/evidence.md): state what was measured versus what was inferred, and never promise a third-party outcome (rankings, citations) as the result of a fix. 1. Author entity (E-E-A-T): the author has a visible on-page bio, sameAs links, and - for AI-attributed content - a named human reviewer note on the page; the author schema node carries `description` + `sameAs`. A thin entity -> P2 signal + a gated proposal. The remediation is site-level: the CMS adapter can write the author profile description where `capabilities()['author_profile_fields']` is true (the WordPress adapter's `update_user` writes it over REST), plus the schema