Agentic SEO — LLM-First Audit Toolkit
Overview
Agentic SEO is a toolkit for auditing the search and discoverability health of websites and code repositories. It was built to solve a recurring problem with SEO tooling: traditional crawlers produce raw data without judgement, while purely AI-driven tools produce confident-sounding advice without evidence. This toolkit deliberately separates the two — deterministic scripts gather hard evidence, and a language model reasons over that evidence to produce prioritised, actionable recommendations.
Sub-Skill Architecture
The toolkit is organised into sixteen focused sub-skills, around thirty-three utility scripts, and ten specialist agents, each responsible for one dimension of SEO — technical health, content quality, structured data and schema, sitemaps, images, crawler readiness, and more. A run can perform a full audit or target a single page or aspect. The modular design means each concern is independently testable and the orchestrating model can compose them as needed for a given site.
Evidence Plus Reasoning Model
Every recommendation is grounded in collected evidence rather than guesswork. Scripts fetch pages, parse markup, validate schema, check Core Web Vitals via PageSpeed Insights, and capture visual state with Playwright; the model then interprets the combined signals — for example weighing experience, expertise, authority, and trust factors, or assessing readiness for AI crawlers. This keeps the analysis defensible: each suggestion traces back to something measured.
Multi-IDE Distribution
The toolkit ships as installable skills and agents for multiple AI coding environments, with installer scripts that wire it into each host. It also includes GitHub repository analysis, so a project's discoverability can be assessed directly from its source. The packaging and documentation are a first-class part of the project — the goal was a polished tool that others can install and run, not just a private script collection.