What each stage does
A closer look at each of the seven agents in the pipeline.
Scout
Reads the Jira ticket and the target repository and produces its "what needs coverage" findings — what the ticket is actually asking for, and where in the repo that likely touches.
Analyst
Turns Scout's findings into a scoped testing plan, prioritizing which areas of the codebase most need coverage and reasoning over any historical context available.
TestWriter
Authors the actual test cases from Analyst's plan — one per scoped file, each with steps, an expected result, and a rationale. These test cases are saved into your workspace's Test Explorer, so they exist independently of the run that created them.
AutoCoder
Proposes the code changes needed to support TestWriter's test cases. This is PR/branch-only — AutoCoder never commits directly to a protected branch. Every proposal requires human review and merge through your repository's normal pull-request flow, regardless of gate mode.
QA Reviewer
Checks TestWriter's and AutoCoder's output before anything executes. It runs a set of deterministic structural checks (test coverage completeness, the size/scope of the proposed change, and detection of proposals that don't correspond to anything TestWriter asked for) — any one of these failing fails the run outright, regardless of anything else. When all three pass, an additional model-based judgment pass can still flag the proposal as a fail if it doesn't look like it addresses the ticket. This is what makes full-auto gate mode trustworthy rather than a rubber stamp.
Runner
Executes each eligible test case — one whose corresponding file AutoCoder actually wrote — inside an isolated, single-use sandbox created fresh for the run and guaranteed to be torn down afterward. A test case whose file was never written is reported as skipped, never silently passed. Runner reports pass/fail based on the sandbox's real exit code, never an assumed result.
Reporter
Aggregates every stage's output into a final run report: what ran, what passed, what failed, and a plain-language summary. Posts that summary to your connected Slack workspace if one is configured, and it's always available on the run's detail page regardless.
Next
See how gate mode changes what happens between QA Reviewer and Runner: Human gate & approvals.