YOUR AI CODING AGENT. WITH A QA TEAM.
Add the AI test harness with sub-agents to your AI coding agent, or run them in the cloud.
Claude
Codex
Cursor
Antigravity
MCPThe CARBON plugin brings testing knowledge, repeatable QA workflows, and specialist AI sub-agents into the tools where you build.
Inside Claude, Codex, Cursor, and Antigravity. Connect via MCP.
Your agent subscription or API usage is separate.
You bring the judgment.
CARBON helps with the repetition, exploration, and evidence.
Inside your AI coding agent
Select a profile to explore its testing expertise
›/carbon
I’ll inspect your coverage, explore the risky journeys, and collect evidence for you.
AI
AI
AI
AI
AISpecialist sub-agentsYour AI coding agent can also ask CARBON to test changes automatically, in real time.
Books by Jason Arbon · Testing AI is a work in progress
Google
Microsoft
ChromeGoogle Search
Bing
ApplauseIn your coding agent first. In the cloud when you need it.
Runs in your coding agent
Your coding agent. Now with a test team.
One command. An entire verification loop.
Specialist AI QA sub-agents, right where you build. CARBON reads your code and existing tests, follows the risk, tests your app, and explains what’s ready—and what isn’t.
In your coding agent · Your tools · You stay in control
/carbonA preview of the CARBON workflow.
Start with /carbon, or focus on the area that matters to you. Add your feature, URL, or question after a command. Names and invocation syntax can vary by coding agent.
54 commands
/carbonFree/carbon-helpFree/carbon-demoFree/carbon-testFree/carbon-morePro/carbon-autoPro/carbon-foreverPro/carbon-issuesFree/carbon-browserPro/carbon-mapFree/carbon-personasPro/carbon-confidencePro/carbon-humanPro/carbon-full-reportPro/carbon-statsPro/carbon-accessibilityFree/carbon-securityPro/carbon-privacyPro/carbon-functionalityPro/carbon-uiPro/carbon-uxPro/carbon-usabilityPro/carbon-performancePro/carbon-reliabilityPro/carbon-compatibilityPro/carbon-contentPro/carbon-dataPro/carbon-apiPro/carbon-networkingPro/carbon-localizationPro/carbon-intlPro/carbon-agenticPro/carbon-geoPro/carbon-statePro/carbon-loadPro/carbon-stressPro/carbon-failurePro/carbon-generatePro/carbon-testsPro/carbon-ai-upgradePro/carbon-chatbotPro/carbon-integrationsPro/carbon-settingsFree/carbon-confidence-initPro/carbon-confidence-planPro/carbon-evalPro/carbon-eval-design-reviewPro/carbon-reviewPro/carbon-security-reviewPro/carbon-skeptical-reviewPro/carbon-statistical-reviewPro/carbon-releasePro/carbon-release-reviewPro/carbon-incidentProNo matching commands. Try another topic or clear your search.
Read the full command guideINSIDE YOUR AI CODING AGENT
Real testing, real evidence. Choose a useful assessment today—or an ongoing quality system for your team.
Free · no account required
A complete, bounded assessment inside the coding agent you already use.
Use it at work. No run quota or hidden findings.
Your agent subscription or API usage is separate.
Access by conversation
Continuous coverage, deeper investigations and your team in the workflow.
Tell Jason what you need. Your request goes to our private Google Sheet. We’ll talk before providing access. No automatic checkout.
Free is source-available under PolyForm Perimeter: internal commercial use is permitted; competing-product use is restricted. Pro is separately licensed. CARBON Cloud is a separate service.
Runs in the cloud
Find release risks before your users do.
Fully autonomous, steerable
web page testing and QA.
Give CARBON a URL. It explores your site, exercises real journeys, and brings back the bugs—with evidence. Set the direction. Let your AI QA sub-agents do the work.
Real cloud execution. The illustration beside it is a demo.
Freshly brewed. Made for you.
Find your coffee1 × Medium oat latte · $5.25
USD · The free sample runs right away. Paid plans are arranged with our team: execution allowance, supported checks, and spending cap are agreed before activation. No checkout or automatic billing here.
USD · Deliverables, included effort, response expectations, and any additional costs are agreed before work begins. No unlimited consulting or automatic subscription.
A FEW THINGS WORTH KNOWING
CARBON Light is free. It uses the coding agent you already have; your subscription limits or provider API charges still apply. Start with a focused request such as /carbon-test check the sign-in form; spend at most 5 minutes. Time and check budgets guide the agent—they are not a guaranteed billing cap. Set hard spending limits with your provider where supported.
No. Your coding agent may send authorized context to its model provider. CARBON keeps project evidence locally, but that is not an air-gap or zero-retention guarantee. Check your agent and provider settings before using sensitive code. Read the data-handling details.
Look for the reproduction steps, observed behavior and supporting evidence. CARBON distinguishes suspected issues from reproduced failures and calls out checks it could not run. AI can make mistakes; review consequential findings and retest fixes rather than treating a score as certification.
CARBON can inspect existing tests and use the browser, terminal and API tools available in your coding agent. Light provides focused assessments; Pro adds advanced test engineering and ongoing workflows. Automated CI execution needs a configured runner and appropriate permissions—it is not installed by visiting this page. Explore Pro workflows.
One team. Cloud or local.
Different perspectives. One clearer picture of quality.
Select a profile to see what each specialist looks for.
AI perspectives, not human reviewers. Coverage depends on your target and the checks performed.
AIAI QA sub-agent
Deliberately hunts for the bugs AI-generated code can introduce and the gaps AI coding agents can overlook. Goes beyond the happy path: missing boundary and null checks, invented API assumptions, incomplete error handling, async races, state lost across multi-step journeys, privacy leaks, and tests that pass without checking the intended behavior. Challenges plausible-looking code with targeted failure cases and evidence.
Save a draft, sign out, then switch accounts. Does the previous user’s draft appear?
An AI QA sub-agent, not a human reviewer. Actual coverage depends on your target, access, and the checks performed.
Explore CARBON’s testing commandsAIAI QA sub-agent
Web-application security specialist focused on authentication, authorization, session safety, exposed data, dependencies, and practical vulnerability evidence.
Change a record ID as a second test user. Can they read a record they do not own?
An AI QA sub-agent, not a human reviewer. Actual coverage depends on your target, access, and the checks performed.
Explore CARBON’s testing commands
AIAI QA sub-agent
OWASP-focused security specialist who challenges access control, injection, insecure design, misconfiguration, vulnerable components, integrity, logging, and SSRF risks.
Submit harmless markup in a comment. Is it displayed as text rather than interpreted as HTML?
An AI QA sub-agent, not a human reviewer. Actual coverage depends on your target, access, and the checks performed.
Explore CARBON’s testing commandsAIAI QA sub-agent
Accessibility specialist focused on keyboard use, assistive technology, inclusive interactions, responsive access, labels, feedback, and barriers affecting disabled users.
Open an editor with the keyboard. Does focus move inside, and return when it closes?
An AI QA sub-agent, not a human reviewer. Actual coverage depends on your target, access, and the checks performed.
Explore CARBON’s testing commands
AIAI QA sub-agent
WCAG specialist focused on criterion-level evidence across perceivable, operable, understandable, and robust behavior, without overstating automated scan results as conformance.
Increase text size to 200%. Can you still read the price and reach the checkout button?
An AI QA sub-agent, not a human reviewer. Actual coverage depends on your target, access, and the checks performed.
Explore CARBON’s testing commands
AIAI QA sub-agent
User-centered design and usability specialist focused on discoverability, task clarity, responsive behavior, interaction friction, visual hierarchy, and inclusive UX.
Try to undo an accidental deletion. Is recovery visible when you need it?
An AI QA sub-agent, not a human reviewer. Actual coverage depends on your target, access, and the checks performed.
Explore CARBON’s testing commands
AIAI QA sub-agent
Compatibility specialist focused on browser, device, viewport, operating-system, assistive-technology, and support-matrix evidence, with explicit coverage gaps rather than assumed portability.
Open checkout on a narrow phone screen. Does the keyboard cover the payment action?
An AI QA sub-agent, not a human reviewer. Actual coverage depends on your target, access, and the checks performed.
Explore CARBON’s testing commandsAIAI QA sub-agent
Performance specialist focused on user-visible latency, Core Web Vitals, payload and request cost, caching, API timing, scalability, mobile constraints, memory, and leaks.
Filter a long product list on a slow connection. Does the interface stay responsive?
An AI QA sub-agent, not a human reviewer. Actual coverage depends on your target, access, and the checks performed.
Explore CARBON’s testing commands
AIAI QA sub-agent
Content and UX-writing specialist focused on page identity, clear copy, information architecture, credibility, navigation, status communication, readability, and content quality.
Compare the offer headline with its checkout terms. Do they promise the same thing?
An AI QA sub-agent, not a human reviewer. Actual coverage depends on your target, access, and the checks performed.
Explore CARBON’s testing commands
AIAI QA sub-agent
Localization specialist focused on translation accuracy, locale formats, pluralization, truncation, bidirectional layout, cultural fit, regulatory copy, and recovery behavior across languages.
Switch languages on a booking form. Do dates, currency, and validation messages agree?
An AI QA sub-agent, not a human reviewer. Actual coverage depends on your target, access, and the checks performed.
Explore CARBON’s testing commands
AIAI QA sub-agent
Data-integrity specialist focused on agreement between rendered values, APIs, events, schemas, exports, storage, calculations, retention, and downstream state.
Export records, then reimport them in an isolated copy. Are IDs, dates, and totals preserved?
An AI QA sub-agent, not a human reviewer. Actual coverage depends on your target, access, and the checks performed.
Explore CARBON’s testing commands
AIAI QA sub-agent
Forms specialist focused on input contracts, validation, boundaries, state, submission, recovery, data quality, conversion barriers, and accessible interaction.
Submit a form with whitespace-only required fields. Does validation explain what to fix?
An AI QA sub-agent, not a human reviewer. Actual coverage depends on your target, access, and the checks performed.
Explore CARBON’s testing commands
AIAI QA sub-agent
Error and recovery specialist focused on truthful status, clear messages, consistent severity, actionable recovery, safe logs, failure loops, and user trust.
Interrupt a save. Does the error preserve your work and offer a safe retry?
An AI QA sub-agent, not a human reviewer. Actual coverage depends on your target, access, and the checks performed.
Explore CARBON’s testing commands
AIAI QA sub-agent
Search-interface specialist focused on discoverability, query handling, suggestions, empty states, input accessibility, relevance feedback, and resilient result navigation.
Type two searches quickly. Can a slow, older response replace the newer results?
An AI QA sub-agent, not a human reviewer. Actual coverage depends on your target, access, and the checks performed.
Explore CARBON’s testing commands
AIAI QA sub-agent
Digital publishing specialist focused on article identity, readability, supporting media, metadata, context, advertising boundaries, and multi-device content behavior.
Open an older news item. Is its publication date clear enough to avoid mistaking it for current news?
An AI QA sub-agent, not a human reviewer. Actual coverage depends on your target, access, and the checks performed.
Explore CARBON’s testing commands
AIAI QA sub-agent
First-impression and conversion specialist focused on value clarity, trust, navigation, calls to action, responsive composition, dead ends, and page credibility.
Follow the primary call to action. Does the next screen deliver what the headline promised?
An AI QA sub-agent, not a human reviewer. Actual coverage depends on your target, access, and the checks performed.
Explore CARBON’s testing commands
AIAI QA sub-agent
Checkout and payment specialist focused on order accuracy, address and payment input, trust, retry safety, pricing truth, completion, and conversion-blocking failures.
Retry a checkout after a timeout in a test environment. Could the order be submitted twice?
An AI QA sub-agent, not a human reviewer. Actual coverage depends on your target, access, and the checks performed.
Explore CARBON’s testing commands
AIAI QA sub-agent
Shopping-cart specialist focused on line-item state, quantities, promotions, totals, inventory changes, persistence, accessibility, and safe transition to checkout.
Change quantity in one tab, then return to another. Does the cart show the correct total?
An AI QA sub-agent, not a human reviewer. Actual coverage depends on your target, access, and the checks performed.
Explore CARBON’s testing commands
AIAI QA sub-agent
Pricing and subscription specialist focused on plan clarity, comparison, currency and locale, hidden conditions, billing cadence, conversion paths, and truthful claims.
Switch between monthly and annual billing. Are the billing interval and total charge explicit?
An AI QA sub-agent, not a human reviewer. Actual coverage depends on your target, access, and the checks performed.
Explore CARBON’s testing commandsAIAI QA sub-agent
Privacy specialist focused on personal-data exposure, consent, tracking, collection minimization, storage and transfer, debug leakage, user expectations, and AI-specific privacy risks.
Enter synthetic personal data. Does any of it end up in URLs, logs, or analytics requests?
An AI QA sub-agent, not a human reviewer. Actual coverage depends on your target, access, and the checks performed.
Explore CARBON’s testing commands
AIAI QA sub-agent
GDPR specialist focused on lawful consent, personal-data exposure, international transfer, minimization, retention, security, data-subject rights, transparency, and accountability.
Request deletion using a test account. Is the data also removed from search and exports?
An AI QA sub-agent, not a human reviewer. Actual coverage depends on your target, access, and the checks performed.
Explore CARBON’s testing commands
AIAI QA sub-agent
Cookie-consent specialist focused on prior consent, purpose and vendor choices, reject symmetry, preference persistence, withdrawal, tracking verification, and regional behavior.
Reject optional cookies. Do optional trackers stay inactive after navigation and reload?
An AI QA sub-agent, not a human reviewer. Actual coverage depends on your target, access, and the checks performed.
Explore CARBON’s testing commands
AIAI QA sub-agent
Policy-content specialist focused on discoverability, internal consistency, versioning, contact details, user rights, actual-product alignment, accessibility, and credible presentation.
Compare the privacy notice with observed storage. Does the notice describe the data actually retained?
An AI QA sub-agent, not a human reviewer. Actual coverage depends on your target, access, and the checks performed.
Explore CARBON’s testing commands
AIAI QA sub-agent
Conversational-AI specialist focused on task success, conversation quality, state, memory, recovery, privacy, safety, injection, tools, latency, integration, and nondeterminism.
Ask the chatbot a question it cannot answer from its sources. Does it admit uncertainty rather than invent a policy?
An AI QA sub-agent, not a human reviewer. Actual coverage depends on your target, access, and the checks performed.
Explore CARBON’s testing commands