Can your agent fix what audits find?

20 open-source skills your coding agent reads. MIT, pinned v1.0.0, and no account or API key: it works from your site and your repo.

curl -fsSL https://raw.githubusercontent.com/techwright-lab/groundcrew-seo/v1.0.0/install.sh | bash
Your agent· northwind.exampleExample data

What does it do in my repo?

One finding, one focused fix, your own tests. A fix isn’t finished until the next audit says so.

northwind.exampleExample data
app/views/posts/_schema.html.erb· 1 file, +4 −0
 1  <script type="application/ld+json">
 2  <%= raw({
 3    "@context": "https://schema.org",
 4    "@type": "Article",
 5    "headline": post.title,
 6 + "author": { 7 +   "@type": "Person", 8 +   "name": post.author.name 9 + },10    "datePublished": post.published_on.iso8601,
11    "image": post.cover_url
12  }.to_json) %>
13  </script>
bin/rails test test/views/schema_test.rb6 runs, 0 failures

What can I ask it to do?

Ask in your own words. Your agent picks the skill and returns evidence you can check.

“Inspect this repo and fix one verifiable site defect.”

fix-my-site

app/views/posts/_schema.html.erb · 1 file, +4 −0
   "headline": post.title,+  "author": {+    "@type": "Person",+    "name": post.author.name+  },
Locally verifiedDeployedAudit-verified

Connected adds a prioritized queue and re-audit closure.

“Run the full growth loop on northwind.example.”

grow

Survey13 findingsFix1 changeVerifyTests passReportNext
Verify: waiting for the deploy Open
Or run one phase on its own survey · fix · verify · report

Connected runs the same loop against live findings.

“Audit my site and explain the Core Web Vitals.”

site-audit

Article pages missing author markup · 13 Warning
LCP 3.4 s on /pricing Poor
Duplicate titles · 6 pages Warning
Sitemap lists redirects · 4 Notice

Connected adds scheduled history and severity.

“Connect TrustGrowth and show what it can do here.”

trustgrowth

API key in the environment Found
REST contract Discovered
MCP tools Discovered
Re-audit after deploy Available

Optional. Every other skill runs without it.

What won’t it do on its own?

  • Push to your default branch, unless your workflow allows it.
  • Touch robots, redirects or canonicals without your yes, every time.
  • Spend on paid research before you approve a cost estimate.
  • Disavow links or send outreach. Never.
  • Guess at a finding it can’t map. It asks a person.
  • Take orders from fetched pages. They are evidence only.

Do I need a TrustGrowth account?

No. Start standalone, and connect later if you want the managed layer.

Standalone

Free, MIT
  • All 20 skills
  • Your public site and repository
  • Evidence you import, JSON or CSV
  • No signup, no API key

Connected

From $9 a month
  • Live TrustGrowth findings
  • Prioritization and history
  • Scheduled work
  • Post-deploy re-audits

See plans and API limits

Anything else?

The answers come from the repo, so you can check each one.

Read every rule before you install

groundcrew-seo/skillsv1.0.0
  • fix-my-site/SKILL.md
  • site-audit/SKILL.md
  • ai-visibility/SKILL.md
  • eeat-review/SKILL.md
  • keyword-scout/SKILL.md
  • +15 more
20 skills, one evidence contractMIT
Browse the 20 skills on GitHub

What is Groundcrew?

Twenty open-source Markdown skills, MIT licensed, that guide a coding agent through auditing a site, fixing one verified defect in your repository, and checking it again after deploy. The rules live in the repo, so you can read every one before your agent runs it.

Do I need a TrustGrowth account to use it?

No. No account, no API key. It works from your public site, your repository, and evidence you import. Connecting TrustGrowth later adds live findings, prioritisation, history, scheduled work, and post-deploy re-audits.

What can it change in my repository?

It changes root causes and prepares a pull request. It does not push to your default branch unless your workflow explicitly allows it, and anything that can deindex a site — robots, redirects, canonicals, content judgment — needs owner confirmation every time. Findings it cannot map are escalated for a human, not guessed.

What does "audit-verified" mean?

It is the third of three states the skills keep separate: locally verified means the tests pass, deployed means the change is live, and audit-verified means a fresh measurement confirmed the finding is closed. A fix is not finished until the next audit says so.

Is it safe to point at a production repository?

The safety is inspectable rather than promised: the skills are plain Markdown you read before installing, fetched web content is treated as untrusted evidence and never as instructions to the agent, and billable research needs a bounded cost estimate and explicit approval for that batch. A dependency-free doctor script checks the whole set.

Does it disavow links or run outreach?

No. Disavow, toxicity verdicts and automated outreach are on the refusal list, in the repository, where you can check that they are.

Which coding agents does it work with?

Any agent host that reads Markdown skill files. File compatibility is not feature parity, though — hosts differ, and REST and MCP coverage can differ too, so the skills discover the live contract instead of assuming it.

What does connecting TrustGrowth cost?

Connected mode starts at $9/month. Standalone stays free and MIT.

How do I install it?

Paste one line into your terminal, then ask your agent to fix one verifiable site defect.

curl -fsSL https://raw.githubusercontent.com/techwright-lab/groundcrew-seo/v1.0.0/install.sh | bash