agentleFS
XPOZpublicXPOZpublic/geo-seo-agent/.claude/skills/site-readiness/SKILL.md

site-readiness

Skill0 starsChanged 3 months agoMITView on GitHub

Measure whether the company's site is technically readable and extractable by AI engines: crawler access, metadata, structured data, heading structure, extractability. Deterministic checks via scripts/site-checks.py; judgment and a dated report here. Run at setup, after major site changes, and when gap analysis suspects extraction problems.

1 other project uses a version of this.

Adopted by

Most recently taken up by

The file

Files
SKILL.md

Site Readiness

The site-side counterpart of /visibility-measure: the sensor measures how the company appears in answers; this skill measures whether the site itself gives engines something to retrieve, parse, and quote. A page that engines can't crawl or can't extract from loses prompts no matter how good its content is, so this reading is a prerequisite the rest of the program leans on. It's also one of the program's quick wins: a first run takes minutes and its site-level findings (a blocked AI crawler, a missing llms.txt, pages without structured data) are usually cheap, high-leverage fixes.

When to run

At setup (part of the day-one quick-win pass), after any major site change (redesign, framework migration, new rendering pipeline), on a slow cadence otherwise (every month or two; the site changes slower than the answers do), and on demand when /content-gap-analysis hits its "asset exists but isn't cited" branch and suspects an extraction problem rather than an authority problem.

Process

1. Run the deterministic checks

.venv/bin/python scripts/site-checks.py crawls up to 25 same-domain pages (--url and --max-pages to override; the default start URL comes from config.json) and writes a run folder recordings/site-checks/<timestamp>/ with site-checks.json and site-checks-summary.md. It records facts only:

  • Site level: robots.txt with a per-AI-crawler access verdict (GPTBot, ClaudeBot, PerplexityBot and peers), sitemap presence and URL count, llms.txt presence.
  • Per page: title and meta-description lengths, canonical, OpenGraph tags, JSON-LD blocks and their types (plus blocks that fail to parse), heading structure, word and paragraph counts, the share of paragraphs in the 40-75 word extractability band, image alt coverage, link counts.

One caveat that travels with the numbers: the script fetches raw HTML. On client-rendered sites the checks describe what a non-JS crawler sees, which is itself a finding (several AI crawlers don't execute JavaScript), but confirm with a rendered fetch before concluding a tag is truly absent.

2. Judge what matters

The script can't rank severity; that depends on what the program is trying to win. Weigh the findings against the panel and the writing principles:

  • Crawler access failures dominate everything. A blocked or partially blocked AI crawler means engines can't read the site at all; nothing else in the report matters until it's fixed.
  • Pages that target panel prompts get page-level scrutiny. A missing Article/FAQ JSON-LD block or a 0% extractability share matters far more on a page a brief is counting on than on the careers page.
  • Extractability needs a human-quality read, not just the share. Spot-read the top pages (the ones targeting high-value prompts, plus any the latest snapshot shows retrieved-but-not-cited): does the opening answer the question in the first two sentences? Are paragraphs self-contained? This is the same judgment strategy/geo-writing-principles.md encodes for new content, applied to what's live.
  • Consistency across pages beats perfection on one. Engines see the site as a corpus; one exemplary page among twenty bare ones reads as an exception, not a pattern.

3. Write the report and route the findings

One dated report: memory/site-readiness/<date>.md (conventions in that folder's index): the run folder it interprets, the site-level verdicts, the page-level findings that matter and why, what changed since the previous report, and a prioritized fix list. Findings route to their owners:

  • Site-level and template-level fixes (robots rules, llms.txt, structured-data templates, rendering) → the awaiting-human list, each with the specific change and the evidence.
  • Page-level content findings on pages targeting panel prompts → inputs for /content-gap-analysis, which owns turning them into rework briefs; this skill never proposes briefs itself.

End the session per ground rules: journal entry, state.md refresh if state changed.

Never

  • Edits the site, templates, or content: it measures and recommends; humans (or approved rework briefs) change things.
  • Proposes content briefs: that is gap analysis's job, with this report as evidence.
  • Substitutes for the visibility sensor: a perfectly readable site can still lose every prompt; only /visibility-measure says what engines actually do.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.