Skip to main content
A QA agent is a saved reviewer configuration: one analysis template, one model, and the arguments that define what it should inspect. Run it repeatedly without rebuilding the review prompt for every trace.

​
What a QA agent reviews

Trace review is the base workflow. Environment, taskset, and task review appear for teams with resource QA enabled.

​
Create a QA agent

Open QA Agents and select New QA Agent:
  1. Choose the subject type.
  2. Select an analysis template compatible with that subject.
  3. Fill any template arguments you want to keep fixed across runs.
  4. Choose the model that will perform the review.
  5. Give the saved configuration a clear name.
The template owns the review method and output shape. The model supplies the analysis. Change either one and you have a different reviewer, even under the same name.

​
How analysis works

HUD freezes the subject into a review workspace before the QA agent begins. A trace review can include:
  • the task prompt and final reward
  • the model conversation and tool calls
  • environment and runner logs
  • screenshots and observations
  • file changes produced during the attempt
The QA agent explores this snapshot. It does not reconnect to the subject environment or alter the original trace.

​
Read the result

Each run keeps the source subject, reviewer configuration, model, status, and structured outputs together. Use those outputs to compare recurring failure modes or to decide which traces need human inspection. QA agents are reviewers, not graders in the task’s reward path. A review can challenge a suspicious reward, but it does not rewrite the recorded reward or silently remove the trace from an evaluation.