ComposalTests in plain English
Describe a journey the way you would brief a tester: who they are, what to do, and what they should see.
Reviewed like code
Each Scenario is a YAML file with a public schema. Lint it offline in CI, review it in a pull request, and upload in one step.
Only the ones that matter
On each pull request Verify selects the Scenarios your change affects and runs them in parallel. Uncertain coverage stays marked untested.
Write Scenarios as files
Download a project’s Scenarios, or have your coding agent write new ones. Lint checks every file against the schema without signing in; a dry-run upload also checks personas and versions. If any file is refused, upload saves nothing.
Verify picks what your change touches
When a pull request opens or updates, Verify waits for a preview of that exact commit and assesses the changed files against your saved Scenarios. It selects the ones it is confident about; uncertain coverage stays explicitly untested.
Results land on the pull request
The selected Scenarios run in parallel as one grouped Run. One maintained comment shows the commit, each result, why each Scenario was chosen, and every finding with what was expected and what happened. Fix with agent hands the failures to your coding agent.
Run them anytime, watch their health
Select Scenarios in Composal and run them against any environment, or run them from the CLI and wait for the result. Each row keeps a strip of its last 20 runs, Scenario Health charts daily success rates, and every session keeps its recording to replay.
What you get back
The run report
Each grouped Run reports every Scenario’s result with its evidence, in Composal and on the pull request. This is the real run for one of our own pull requests.
What you can do with it
Fix with agent
Hand the failure to Codex, Claude, or Cursor, or copy the prompt. It links the run and the CLI steps to read every result.
Retry Verification
Prepare a new attempt from the same screen, with the same selection, environment, and limits, using the current Scenario definitions.
Replay any step
Choose a step and the recording jumps to it, so you see exactly what the browser saw when an outcome failed.
Run them from the CLI
Run Scenario files as written, without saving them, and wait for the result. Each line reports one Scenario, and the run links back to its page.
