Tools provided
Tools the agent exposes — your AI client calls them automatically when it needs them.
- audit_page — Run the full audit (SEO, accessibility, security, performance, UX, GEO, AIO and hidden issues) on a page you supply: score 0-100 with category sub-scores and confidence, a top-3 'fix these first' list, every finding with its evidence and a concrete fix, and a 'not checked' list. Use this when the user wants to audit a site or page. Pass the page HTML; add response headers and the URL for the checks that need them.
- check_seo — Run only the SEO rules (title, meta description, canonical, robots, hreflang, structured data, Open Graph, headings, links) on the supplied HTML.
- check_accessibility — Run only the accessibility rules (WCAG 2.2 static checks: alt text, labels, landmarks, ARIA validity, names, tables, media) on the supplied HTML.
- check_security_headers — Run only the security rules: CSP parsing, HSTS, cookie flags, framing, CORS and disclosure headers (judged only when headers are supplied), plus HTML-level checks (mixed content, SRI, target=_blank, inline handlers, forms).
- check_performance_hints — Run only the performance hints derivable from HTML and headers: render-blocking resources, preload/preconnect, third parties, images, fonts, DOM size, compression and caching. No timings are measured.
- check_ux — Run only the UX rules: form input types, submit controls, button labels, password-manager friendliness, open dialogs, wall-of-text paragraphs, fake lists, navigation size, nested controls, breadcrumbs and tappable phone numbers.
- check_ai_visibility — Run only the AI-visibility rules: GEO (authorship, dates, publisher entity and sameAs, citations, trust pages, clean canonical, content present without JavaScript) and AIO (question headings answered at once, FAQ markup matching visible text, ordered how-to steps, chunk-sized sections, summary and sources).
- check_hidden_issues — Run only the hidden-issue rules: defects that look fine on screen. Leaked localhost or staging URLs, template debris, mojibake, controls with no name, broken in-page anchors, secrets or personal data in URLs, leaking HTML comments, missing doctype, tracker inventory and charset mismatches.
- compare_audits — Compare two audit_page results rule by rule: what regressed, what improved, what still fails, with a plain-language verdict. Pass the full result objects of two audit_page calls.
- explain_finding — Explain one rule id (for example SEC-002 or A11Y-013): why it matters, the standard it comes from, how it is detected and the precise fix, plus the exact nodes found when you pass the audit result or the page.
- plan_inputs — Plan the inputs of a tool: returns the questions, JSON schema and an example for audit_page (or another tool). Use it when you are unsure what to supply.
- open_form — Use this the moment the user wants to run this agent on their own input. It returns the form of run arguments to collect (the page HTML, its URL and response headers). Do not run any analysis before calling this.
- run_form — Internal: invoked by the open_form form when the user submits. Runs the full audit with the submitted fields. Not for direct use.
- run_full — Run the agent end to end on the supplied page: the same full audit as audit_page. Returns a needs_input form if no HTML was supplied; it never invents data.
- list_capabilities — List the agent's capabilities: tools, the rule catalogue by category and severity, input limits and the honesty rules. Useful for discovery.
- discover_intent — Understand your goal and co-design the exact input via clarifying questions before running an audit (for example which category matters, and how to fetch the page and headers).
