AI verification engine
Basano, an AI critic whose pass you can trust.
It checks a finished build against a standard and reports what holds and what fails.
What is Basano?
Basano is an open-source AI critic for the Claude Skills catalog, built by RampStack. It reviews a finished build against a standard and reports what holds and what fails, prioritized by severity. The name comes from the Greek basanos, the touchstone, the stone that tests whether gold is genuine.
The idea
The honest verdict.
A critic is only worth having if its judgment is trustworthy in both directions. Basano is built so that its "this holds" is reliable because its "this fails" is unsparing.
A failure is never softened into a pass or a minor note. A verdict reads passes only when every assessed area holds.
A passing verdict that carried a real failure is structurally impossible to construct. The rule is enforced in the synthesis, not promised in the prose.
What Basano does
Five jobs, each in service of the verdict.
Scopes the review.
Turns a build and a standard into the areas worth checking: SEO, accessibility, code quality, performance, content, and others as the build calls for.
Selects verification skills.
For each area, picks the relevant audit and review skills from the catalog, the same selection used elsewhere, so the review maps to skills that actually check that dimension.
Reports a prioritized verdict.
Synthesizes findings into a structured result. Failures lead, ordered by severity, with the reason each one matters. Not a raw issue dump, a verdict.
Surfaces holds, not just fails.
A critic is only worth having if its judgment is trustworthy both ways. Basano shows what holds alongside what fails, so a pass is something a reader can act on.
Reports, does not fix.
Basano returns a verdict and stops. The fix is the human's call, or a follow-on build; the critic only judges.
The stop
It judges, you decide.
Basano reports the verdict and stops. It changes nothing. The critic only judges, which is why the judgment can be trusted.
How Basano works.
Basano is the AI verification engine. Give it a live URL and it runs a fixed set of deterministic checks against the page as it actually renders, grouped into dimensions: accessibility, performance and Core Web Vitals, and SEO including answer-engine and generative-engine readiness. It returns a verdict in three buckets: what holds, what fails, and what could not be assessed.
The third bucket is the point. Basano does not guess. If a page blocks automated access, or a signal is not present in the HTML, it reports that the check could not be run rather than inventing a pass or a fail. Findings come from what the page proves; everything else stays honestly unassessed. That restraint is the product, not a limitation of it.
When a page fails checks, Basano hands the failures forward as a prepared fix plan: the specific catalog skills that address each failure, grouped by area and sequenced into steps, with the inputs each step needs. The plan is prepared, not run. A person reviews and approves it before anything executes.
The pipeline
Select a stage to see what happens there.
URL. One live URL is the input, the page as it actually renders.
A worked example
A live product page.
Input: the URL of a product page on a mid-market retail site.
Basano's verdict
Accessibility
FailsProduct images are missing alt text (WCAG 1.1.1).
SEO, technical
FailsThe canonical link is missing, so duplicate URLs compete for the same page.
SEO, answer-engine and generative-engine readiness
FailsOn three counts: no structured-data depth for AI extraction, AI crawlers blocked in robots.txt, and no llms.txt at the site root.
Performance and Core Web Vitals
Not assessedThese need a rendered measurement the current capture does not take, so Basano reports them as unassessed rather than scoring them.
Everything else
Holds
The prepared fix plan Basano hands forward
Step 1
Accessibility
Skills:accessibility-auditmedia-asset-management
Add descriptive alt text to the product images and fix the asset pipeline that was dropping it.
Step 2
Technical SEO
Skill:seo-technical
Add the self-referencing canonical.
Step 3
AI search readiness
Skill:seo-aeo-geo
Add structured-data depth, allow AI-crawler access in robots.txt, and publish llms.txt at the root.
Each step names the catalog skills it draws on, states one goal, and lists the inputs it needs. Nothing runs until a person approves it.
From failures to skills
The same example as a map. Select a failure to trace its step and skills; select a skill to trace the failures it addresses.
Preparing that plan, selecting the skills and sequencing the steps, is the orchestration engine's job. See how Tholo does it.
What the example shows.
One URL goes in, and a sequenced workflow of named catalog skills comes out, tied to exactly the failures the page proved. A different page surfaces different skills: a thin page pulls in content skills, a slow page pulls in performance-optimization, a page with broken flows pulls in usability-testing. The engine is the judgment that maps a verified problem to the right skills in the right order, and then stops at the point where a person approves the work.
What it does
One task at a time.
Each card is one task Basano handles: what goes in, that Basano runs it, and the artifact that comes out.
Basano workflows
Pick a workflow to see what goes in, and the artifact that comes out.
A page and a standard in, a three-bucket verdict out.
A page and a standard. A built page and the standard it is meant to meet.
Read the code. Read the catalog.
The repo is the working implementation and the worked examples. The catalog is what Basano reviews against. Solutions is how to bring RampStack in to operate it for you.