AI agentic verification harness

AI Test Harness

Download CARBON Plugin

QA for your coding agent.

Works with

Claude
Codex
Cursor
Antigravity
MCP agents

Test frameworks and suites

Vibium
Playwright
Selenium
Cypress
pytest
JUnit
Gherkin
Robot Framework
TestRail
Xray
Manual cases

Browser-scale experience

Built by the team that automated the Chrome browser at scale.

That work turned browser behavior into repeatable evidence. CARBON brings the same operational discipline inside AI coding agents—coordinating focused testers to increase confidence and make software claims more credible.

One command

Tell CARBON what to test.

Use /carbon by itself for the complete testing cycle—or add any testing request in plain language. CARBON turns it into a robust, repeatable system with evidence and confidence analysis.

More specialized commandsOptional

Current testerJason Arbon · Overall CARBON

/carbon

Complete testing cycle

Scope Test Evidence

Evidence-backed result

Ready0 / 3

Coding agent · jtest

Agent runtime

Current testerJason Arbon · Overall CARBON

/carbon

The default example uses published jtest/.carbon results. The natural-language example demonstrates how CARBON structures a request; specialized examples replay observed evidence or preserve explicit not-run boundaries instead of inventing passes.

Inside the harness

A full cast of AI testers.

CARBON routes the specialist whose testing perspective fits the risk in front of you. Each one has a defined domain, concrete checks, and the same obligation to show evidence and limits—not just a generic testing prompt.

Sharon

Security Tester

Tests authentication, authorization, sessions, secrets, exposed data, dependencies, and practical vulnerability evidence.

Security

Marcus

OWASP Security Tester

Challenges access control, injection, insecure design, misconfiguration, vulnerable components, integrity, logging, and SSRF risk.

OWASP

Alejandro

Accessibility Tester

Tests keyboard use, assistive technology, labels, feedback, responsive access, and barriers affecting disabled users.

Accessibility

Hiroshi

WCAG Compliance Tester

Gathers criterion-level WCAG evidence without overstating an automated scan as accessibility conformance.

WCAG

Mia

Usability Tester

Tests discoverability, task clarity, responsive behavior, interaction friction, visual hierarchy, and inclusive UX.

Usability

Tariq

Performance Tester

Measures latency, Web Vitals, payload and request cost, caching, API timing, mobile constraints, memory, and leaks.

Performance

Sophia

Content Quality Tester

Reviews page identity, copy clarity, information architecture, credibility, status communication, readability, and localization.

Content

Richard

Forms Tester

Exercises input contracts, validation, boundaries, state, submission, recovery, data quality, conversion, and accessible interaction.

Forms

Fatima

Error Message Tester

Tests truthful status, clear messages, consistent severity, actionable recovery, safe logs, failure loops, and user trust.

Error UX

Mei

Search Box Tester

Tests query handling, suggestions, empty states, input accessibility, relevance feedback, and resilient result navigation.

Zara

News Content Tester

Reviews article identity, readability, supporting media, metadata, context, advertising boundaries, and multi-device behavior.

Publishing

Yuki

Homepage & Landing Page Tester

Tests value clarity, trust, navigation, calls to action, responsive composition, dead ends, conversion, and credibility.

Landing Pages

Hassan

Checkout Tester

Tests order accuracy, address and payment input, trust, retry safety, pricing truth, completion, and conversion-blocking failures.

Checkout

Priya

Shopping Cart Tester

Tests line-item state, quantities, promotions, totals, inventory changes, persistence, accessibility, and checkout transitions.

Shopping Cart

Mateo

Pricing Page Tester

Tests plan clarity, comparison, currency and locale, hidden conditions, billing cadence, conversion paths, and truthful claims.

Pricing Pages

Pete

Privacy Tester

Finds PII exposure, consent and tracking problems, excess collection, unsafe storage or transfer, debug leakage, and AI privacy risk.

Privacy / PII

Zanele

GDPR Compliance Tester

Tests lawful consent, minimization, retention, international transfer, data-subject rights, transparency, security, and accountability.

GDPR

Rajesh

Cookie Consent Tester

Verifies prior consent, purpose and vendor choices, reject symmetry, preference persistence, withdrawal, tracking, and regional behavior.

Cookie Consent

Sundar

Legal Policies Tester

Checks policy discoverability, consistency, versioning, contact details, user rights, product alignment, accessibility, and credibility.

Legal Policy

Jason Arbon

GenAI Code Tester

Challenges AI-generated logic, boundaries, null and empty states, failure handling, API use, security, privacy, tests, and misleading shortcuts.

AI Code Review

Diego

AI Chatbot Tester

Tests task success, conversation quality, memory, recovery, privacy, safety, injection, tool use, latency, integration, and nondeterminism.

AI Chatbot

Optional Cloud add-on

Download locally. Add Cloud when you need it.

The CARBON plugin runs inside your coding agent and stays local by default. Connect testers.ai Cloud only for selected work that benefits from parallel capacity, recurring runs, saved account history, API automation, or a shared web results console.

Keep the downloaded client local.

Install CARBON and use /carbon normally. Your project context, local tests, credentials, evidence, and reports stay in your coding-agent environment unless you explicitly choose a Cloud workflow.

Connect your Cloud account.

Create a testers.ai account, then store the premium API key through CARBON’s protected settings flow. Keys and test credentials never need to appear in chat or reports.

Send only approved runs.

Use explicit testers_plan, testers_run, testers_results, testers_import, or testers_schedule workflows. CARBON never silently switches a local run to paid Cloud execution.

Explore CARBON Cloud Open Web Console Read Cloud setup details

FAQ

The practical questions.

CARBON stays inside your AI coding-agent harness, works with the tools you already use, records what it actually observed, and keeps missing evidence visible.

What does /carbon do?

By itself, /carbon runs the full testing cycle: it evaluates the project, target, and change; maps risk; creates the right tests; executes safe checks; collects evidence; and analyzes confidence with explicit limits and next actions. You can also add anything you want in plain language—what to focus on, which kinds of tests or frameworks to use, a particular flow, risk, environment, or constraint—and CARBON turns that request into a robust, repeatable testing system.

Can CARBON use my existing test frameworks?

Yes. CARBON can work with and control any classic test framework your coding agent can run, including multiple frameworks in the same project. It can help interpret, import, repair, and migrate tests between frameworks and implementation languages while keeping the durable test intent and evidence visible. Nothing is rewritten until you approve the change.

Can I use my own API keys?

Yes. CARBON can use the model and provider credentials already configured in your AI coding-agent environment. You keep control of those keys; CARBON does not require you to hand them to testers.ai.

Is CARBON private by default?

Yes. CARBON runs locally inside your AI coding-agent harness and writes its state and evidence to your project. testers.ai does not receive your code, prompts, test data, credentials, screenshots, or reports. We do not need or want your private data. Your chosen coding agent and model provider still handle data according to the configuration and terms you selected.

Which coding agents are supported?

The current packaging supports Claude Code, Codex, Cursor, Antigravity, and MCP-compatible coding agents. The same evidence boundaries apply regardless of host.

Can CARBON test more than browser UI?

Yes. Top-level testing covers functionality, UI, UX, usability, accessibility, privacy, security, performance, networking, reliability, compatibility, content, data integrity, APIs, and localization—when the required access and evidence are available.

Does a CARBON report mean everything passed?

No. Reports separate observed results from blocked, deferred, not assessed, and not measured areas. A completed-looking report is not treated as proof of untested coverage.

Who makes the release decision?

You do. CARBON assembles the evidence, severe findings, unresolved evidence gaps, and rollback context, but preserves human authority over ship, canary, hold, or rollback decisions.