ax evals
Read eval definitions, grader trends and case results.
| Command | Use |
|---|---|
evals list | List an agent's eval definitions. |
evals show | Read one eval definition. |
evals trend | Read grader history. |
evals results | Read case results with held-out redaction. |
Every command requires --agent <AGENT_ID>, the Context agent UUID, and accepts
--org <ORG_ID>. Eval ids are UUIDs from evals list. Your organization must
have agent evals enabled. Arguments are optional unless marked required.
evals list
List definitions for an agent.
ax evals list --agent <AGENT_ID> --json| Argument / flag | What it does |
|---|---|
--agent <AGENT_ID> | Required. Context agent UUID. |
--limit <N> | Rows per page, default 50, maximum 200. |
--page-token <TOKEN> | Continue from nextPageToken. |
--json | Print {rows, nextPageToken}. Otherwise print one record per row. |
These paging and output flags also apply to trend and results.
evals show
Read one eval definition as {record} JSON.
ax evals show <EVAL_ID> --agent <AGENT_ID>evals trend
Read paged grader history for an eval.
ax evals trend <EVAL_ID> --agent <AGENT_ID> --jsonevals results
Read paged case results for an eval.
ax evals results <EVAL_ID> --agent <AGENT_ID> --split test --json| Argument / flag | What it does |
|---|---|
--split <SPLIT> | Filter to train, validation or test. Omit for all splits. |
Unrevealed test results include redacted_held_out: true. Evidence, free-text
errors and trace pointers remain hidden. A reveal applies only to your user
and the exact dataset version. To record one explicitly, use
datasets pull --include-held-out,
the reveal API, or
dataset_reveal. Reading a result never records a reveal.