skilld
All skills
agricidaniel avatar

/seo-agentic

@93b720e
by Agrici.Danielagricidaniel/claude-seo18k stars
2,665

Audit and fix agent readiness: the Lighthouse Agentic Browsing fraction, accessibility tree for agents, robots.txt and Content-Signal for AI agents, WAF treatment of agent traffic, llms.txt, Markdown delivery, ai-catalog.json, /.well-known discovery files, and WebMCP tools. Exclude AI citability and brand signals (seo-geo) and commerce protocol depth (seo-ecommerce).

Ask your Agent

Use this Skill: https://skilld.dev/gh/agricidaniel/claude-seo/seo-agentic

This session only. Nothing lands on disk.

referencesvendor-matrix.md

≈1.3k tokens on demand. Your agent reads this file only when SKILL.md points to it.

Agent vendor matrix (dated, source-graded)

Fast-moving facts live here and nowhere else in this skill. Each row carries a source grade: P = primary (vendor docs, spec, source code), S = secondary only, C = conflicting sources. Last full check: 2026-09-23. Refresh before quoting any row older than 60 days (see "Refresh procedure").

What each agent reads and whether it calls WebMCP

Vendor / product Reads WebMCP Identity Grade
OpenAI: ChatGPT desktop built-in browser ("site tools") ARIA / accessibility tree (publisher FAQ) Yes, imperative only, top-level document, on by default, not Enterprise/Edu ChatGPT-User; signs agent and cloud-browser traffic with Web Bot Auth P
OpenAI: cloud browser (ChatGPT Work), Codex as above not documented Web Bot Auth S
Google: Gemini in Chrome, Chrome auto browse page elements and screen positions no public evidence of consumption; Google co-edits the spec Google-Agent (user-triggered, generally ignores robots.txt); experimenting with Web Bot Auth as https://agent.bot.goog P (identity), S (reading)
Microsoft: Edge, Copilot Mode not documented Edge supports WebMCP for testing; no evidence Copilot consumes tools not researched P (Edge), S (Copilot)
Anthropic: Claude in Chrome (Claude Code, Cowork, Desktop), computer use page text, DOM, console and network, screenshots none (public issue reports discovery missing) ClaudeBot, Claude-SearchBot, Claude-User; all honour robots.txt; IPs at claude.com/crawling/bots.json P (bots), S (reading)
Perplexity: Comet accessibility tree and screenshots no public signal PerplexityBot honours robots.txt; Perplexity-User generally does not P (bots), S (reading)
Brave: Leo not documented experimental, Nightly behind brave://flags/#enable-webmcp-testing not verified S
Opera Neon, Dia, Sigma, Fellou not documented no public signal not verified S
Apple: Safari / WebKit n/a opposes (agents are closer to assistive tech; sites should not detect them) n/a P (label), S (quotes)
Mozilla: Firefox n/a neutral; not implementing n/a P

Two gaps to state plainly in reports: no primary source shows any consumer agent sending Accept: text/markdown, and robots.txt behaviour for Dia, Opera and Comet agent actions is unverified.

Product status notes

Item Status Grade
ChatGPT Atlas browser reported stopped 2026-08-09; ChatGPT agent removed early Aug 2026 S
ChatGPT Instant Checkout reported moved to Apps in 2026-03 S
Project Mariner reported shut down 2026-05-04 S
Chrome WebMCP origin trial M149 to M156; no ship milestone announced P (range), S (end date 2026-11-17)
Lighthouse 13.5.0 (npm latest, 2026-09-18); PSI runs 13.5.0 P
Web Bot Auth draft-ietf-webbotauth-httpsig-protocol-00 (2026-09-01); Signature-Agent dictionary form P
Content-Signal Cloudflare CC0 policy; IETF individual draft expired 2026-04-04 (P); no Google statement found, and Google's robots.txt spec does not list the field (S) P/S
MCP Server Card SEP-2127 open and unmerged (updated 2026-09-12) P

Evidence of impact (flag vendor numbers)

  • Cloudflare's token-reduction figures for Markdown are vendor-sourced.
  • "Token efficiency" percentages for WebMCP come from marketing posts with no traceable benchmark. Do not cite them.
  • Agent benchmarks such as WebArena measure models, not site quality.
  • No controlled public study links accessibility-tree quality or WebMCP to agent task success. Treat such claims as hypotheses.

Contested or unverified (do not quote without rechecking)

  • Exact WebKit and Mozilla position dates and the WebKit quotes (secondary trackers, one written by an automated agent). Check the GitHub timeline.
  • A WebKit suggestion of a new working group or TPAC workshop (vendor tracker paraphrase only).
  • Anthropic and Perplexity Web Bot Auth signing (secondary only).
  • Cloudflare verification of the dictionary-form Signature-Agent (reported mismatch, not tested).
  • Whether Content-Signal is scoped per robots.txt group (no normative text).
  • ChatGPT site-tools model requirements (vendor pages disagreed).

Refresh procedure

  1. PSI: run lighthouse_agentic.py https://claude-seo.md/ --json. If lighthouse_version changed, diff the category auditRefs against references/lighthouse-agentic-category.md.
  2. Lighthouse source: npm view lighthouse version, then read core/config/default-config.js and core/audits/agentic/ for the new tag.
  3. WebMCP: the spec's last commit (webmachinelearning/webmcp), chromestatus feature 5117755740913664, learn.chatgpt.com/docs/webmcp.
  4. Bots: developers.openai.com/api/docs/bots, support.claude.com crawler article, docs.perplexity.ai/guides/bots, Google's user-triggered fetchers page.
  5. Drafts: datatracker.ietf.org for webbotauth and aipref.
  6. Update the dates here and the CHECKED_ON constant in agentic_check.py in the same change.

Source: SKILL.md on GitHub

Third-party checks

1 warning1d3 checks · Risk SAFE
  • Gen Agent Trust Hub1d

    This skill is a utility for auditing and improving website readiness for AI agents. It processes external content which introduces a risk of indirect prompt injection, and it executes local scripts to perform audits. The skill includes security guardrails like SSRF protection and explicit instructions for the AI to treat external data as untrusted.

  • Socket1d

    No alerts

  • Snyk1d

    Risk: MEDIUM · 1 issue

Provenance

Signed by skilld at 93b720e. This ties the file your Agent reads to that commit on GitHub. It does not review the instructions.

Last checked against GitHub 3 days ago.

Activeupdated 3 days ago

Capability

user-invocable
true
argument-hint
[audit|fix|lighthouse|refresh] [url]
metadata
{
  "author": "AgriciDaniel",
  "version": "2.4.1",
  "category": "seo"
}

README badge

README badge for agricidaniel/claude-seo/seo-agentic