What is the audit website skill?
The audit website skill is in the squirrelscan/skills repository. It needs the squirrel CLI. An AI coding agent runs an audit with the LLM report format, proposes fixes, maps each issue to the template or content behind it, fixes in batches and re-audits with before and after scores. It names 260 plus rules across SEO, performance, security and more.
It is the audit-website skill in squirrelscan/skills. Its description says it runs SEO, performance, security, technical, content and accessibility checks plus 15 other rule categories, 260 plus rules in all, then drives an iterative fix loop.
It needs the squirrel CLI installed and in your path. The skill defers CLI setup, login and publishing to a companion squirrelscan skill in the same repository.
What changes in your agent's output
- The agent runs the audit with the llm format, which the skill calls compact and made for agents, and asks which site to audit if you did not say.
- It presents score, grade and top issues by severity, then proposes fixes and confirms with you before changing anything.
- Each issue is mapped to the template, component or content file behind it, and fixes are applied in batches.
- It re-audits, using the refresh flag after deploys, and shows before and after scores.
- Judgment calls such as whether to remove a broken link are flagged for you instead of guessed.
What the skill covers
Three crawl modes: quick for 25 pages, surface for 100 with one sample per URL pattern, and full for 500. Score targets scale with the starting grade, from reaching 75 plus when below 50 to fine-tuning above 85. Rules carry a level of error, warning or notice and a rank from 1 to 10.
Who the audit website skill is for
Best for
- Sites you control where you want an agent to fix findings in code, then verify with a re-audit.
- Template-level coverage, since the surface mode samples one page per URL pattern.
- Pairing with the Lighthouse-based web quality audit skill or fixing-metadata.
Not for
- Machines without the squirrel CLI. The skill declares it a requirement.
- Sites you do not own. A crawl sends many requests, so audit only sites you have permission to test.
Use it with UiChemy
Audit the published UiChemy page set on your WordPress site, then fix titles, links and content issues in the editor. Re-audit with a full crawl before you sign off. Click Get Started on uichemy.com. For a broader quality pass, see the web quality audit skill and the web SEO skill.
How to install Audit Website Skill: squirrelscan Site Audits for AI Agents
git clone https://github.com/squirrelscan/skills.gitThe squirrelscan README also lists squirrel skills install as a CLI route for the skills, and a plugin route for some agents. The copy route above needs only git.Install the squirrel CLI
The skill requires it. The squirrelscan README gives the download page and says to verify with
squirrel --version.Clone the skills repository
Run
git clone https://github.com/squirrelscan/skills.git. The skill is skills/audit-website.Copy the skill folder
Copy skills/audit-website into your agent's skills folder. Add skills/squirrelscan too if you want the companion skill for CLI setup and login.
Ask for an audit
Name the site and ask the agent to audit and fix it. Start with the quick mode, then move to surface or full.
Check that it loaded
Ask for an audit of one site. The agent should run squirrel with the llm format, show a score and grade, ask before editing, and re-audit after fixes to show the change.
Key features
- CLI-driven auditRuns the squirrel CLI and reads its LLM-optimized report, with a note to use the llm format.
- 260 plus rulesThe skill description names SEO, performance, security, technical, content, accessibility and 15 other rule categories.
- Three scan modesQuick at 25 pages, surface at 100 and full at 500, with flags such as refresh, resume and a page cap.
- Six-step fix loopPresent the report, propose fixes, map issues to source, fix in batches, re-audit and repeat.
- Score targetsTargets by starting grade, from 75 plus for a failing site to 95 plus for a site already above 85.
- Severity and rankRules carry error, warning or notice levels and a 1 to 10 rank. Errors go first, then high-rank warnings.
- Regression checksCompare against a baseline audit to prove improvement or catch regressions.
- Rule docs lookupAny rule is documented under a rule category and rule id path on the squirrelscan docs site.
- Bot-protected sitesWeb Bot Auth headers can be passed with repeated header flags, and their values are redacted in output.
- Clear done criteriaAll errors fixed, warnings fixed or documented, a confirming re-audit and a before and after summary.
Frequently asked questions
What is the audit website skill?
It is a skill in squirrelscan/skills that has an AI coding agent run squirrel CLI audits of a site, read the compact report, map findings to code or content, fix them in batches and re-audit until a score target is met.
How do I install it without npx?
Install the squirrel CLI first, then clone the skills repository with git and copy skills/audit-website into your agent's skills folder. The README also lists a squirrel skills install command, which is the project's own tool.
How many pages does an audit crawl?
Quick mode defaults to 25 pages, surface to 100 with one sample per URL pattern such as a blog post template, and full to 500. A page cap flag overrides the default. The skill advises signing off against a full crawl.
Will the agent change my code without asking?
No. The loop says to propose fixes and confirm with you before changing anything, and to flag judgment calls such as broken links for ya review. After each batch it checks that the project still builds.
Can it audit sites that block crawlers?
The skill says that if a site blocks unknown crawlers, such as behind Shopify or Cloudflare protection, you can pass Web Bot Auth headers. Only do this for sites you own or have permission to test.
How is it different from the web quality audit skill?
The web quality audit skill uses Lighthouse and traces on chosen URLs. This one uses the squirrel crawler for many pages and 260 plus rules, then drives fixes to a score. Together they cover depth and breadth.
What license does it use?
The squirrelscan/skills repository uses the MIT license, as GitHub shows, while the skill frontmatter points to the repository LICENSE. The 72.1K install count on this page comes from a third-party skills listing.



