CARBON · AI Agentic Verification Harness

One command.
A complete
testing loop.

/carbon

Run this command in your coding agent after installing CARBON. Host prefixes may vary. This preview does not start a test or send data.

Full help & usage example →

An automated testing loop from a single command. CARBON understands your project, chooses high-value tests, runs them, and reassesses confidence as the evidence comes in.

Install CARBON →See it in action

Runs inside your coding agent, locally by default.
Scoped by your budget. Consequential actions need approval.
No project yet? Start with /carbon-demo.

CARBONILLUSTRATIVE EXAMPLE
Coding-agent chat
/carbon
  1. I’m mapping your change.Checkout, retries, stored state.
  2. I’m testing the risky paths.12 checks planned. 10 run.
  3. I found two failures.Evidence linked to each issue.
  4. Here’s what to test next.Reassess. Prioritize. Continue.
CARBON report
68/ 100

Confidence judgment
Evidence strength, not release probability.

8 passed2 failed2 not run
Still needs attention

Retry can lose the cart.
A stale total survives reload.

Next useful check

Verify recovery across two tabs.

Illustrative chat trace and report—not measured results. Actual confidence depends on the available evidence and tested scope.
WHAT’S NEW

CARBON Map

An interactive way for AI QA experts to understand product risks and guide AI testing. Explore pages, journeys and elements; see the evidence, spot gaps, and steer the next investigation.

Explore what Map can do
  • Inspect one page at a time, with highlighted elements and recorded journeys.
  • Compare functional and business risks, then prioritize the next useful checks.
  • Add notes, priorities and custom tests that remain attached across builds.
  • Review added, changed, removed and restored surfaces without losing history.

/carbon-map opens the map for your project. Captures show what was observed—not proof that every behavior passed.

Captured CARBON Map example: page explorer, selected screenshot, results and suggested checksExpand the captured example →

Optional cloud workflows

Your expertise.
Your direction.

Steer cloud testing toward what matters: the user journeys, business risks and questions that deserve attention.

Cloud steering & deployment details
  • Define the target, scope, priorities and test depth before a run.
  • Agree budgets, concurrency and stop conditions for approved remote work.
  • Review findings and feed expert context into the next investigation.

First, steer the plan.

Choose features, checks, personas and test cases. Add the context that makes the investigation useful.

Illustrative cloud steering UI · click to enlarge

Then, inspect the evidence.

Review the findings and use the report to guide your next pass.

Illustrative cloud report · sample values, not a measured CARBON run

Cloud execution is separately configured. Enterprise/private-cloud availability and capacity are agreed with the team; installing the local plugin does not provision cloud workers or upload your project.

Sharing artifacts, using external services and paid execution require explicit approval.

Discuss cloud steering with the team →