Reviewer sessions
The review pipeline end to end: the enriched diff, the agents that ran, every finding and its severity, the reflection pass that cuts noise, and the comment merge that folds duplicate findings into one comment.
Every agent run has a record: what ran, what it found, what it skipped, and why. Open any session to see how the result was produced, and replay it.
Every run produces a session with the same shell: origin, status, environment, timeline, outcome, and cost. What changes is the stages inside it, because each agent works differently.
The review pipeline end to end: the enriched diff, the agents that ran, every finding and its severity, the reflection pass that cuts noise, and the comment merge that folds duplicate findings into one comment.
What the sandbox did: repo snapshot and setup, the patch the agent composed, the lint and build checks it had to clear, then the pushed commit. Validation failures end in diagnostics, so it's clear nothing was pushed and why.
Why Merger called it: the rule gates the PR cleared, the signals it weighed (CI, open findings, diff, blast radius, git history), the merge, reject, or escalate verdict, and whether it merged on its own.
How a plan got made before any code existed: the invocation, every tool call spent investigating the codebase, and the plan version it produced in the Plans workspace.
The one run type that needs no pull request. A full codebase scan walks the repository itself, and the session tracks it from request through scanning to findings persisted by severity.
Production first: the traces and logs that opened the run, how spans resolved to a file and function, the risk it found, and the Fixer session it handed the fix to. Signal, code, fix, PR, read as one chain.
Watch any of them as it runs: current stage, progress, findings so far, and anything blocked, without waiting for it to finish.
Start free and open a session on your first review, or have us walk your own pipeline.