View experiments and results

The eval experiment command group (alias exp) is used to view experiments submitted by eval run: list experiments, inspect an experiment’s status and aggregate scores, and fetch per-case results. On TEA, details include targets, evaluator versions, aggregate scores, and field mappings; on Coze, the commands continue to query by project.

--project applies only to the Coze evaluation backend; the TEA backend ignores it.

eval experiment list

List experiments in a project or workspace: ID, name, status, start time, creator, and aggregate score.

Flag / Argument	Description	Default
`-p, --project <name>`	Coze project name; ignored on TEA.	`default`
`--json`	Output raw JSON.	`false`

agentkit eval experiment list

agentkit eval exp list --json

eval experiment show

Show an experiment’s details: status, dataset, target, evaluators, aggregate scores, and field mappings. get is an alias of show.

Flag / Argument	Description	Default
`<id>`	Experiment ID.	Required
`-p, --project <name>`	Coze project name; ignored on TEA.	`default`
`--json`	Output raw JSON.	`false`

agentkit eval experiment show 75901xxxxxxxxxxxxx
agentkit eval experiment get 75901xxxxxxxxxxxxx

ID              75901xxxxxxxxxxxxx
Name            qa-set-<timestamp>
Status          Success
Dataset         qa-set (75900xxxxxxxxxxxxx)
Target          my-agent
Overall score   1.00

Evaluators:
  relevance @0.0.1  avg 1.00
    1.00: 2 (100%)

Field mappings:
  target: user_input <- dataset.input
  relevance: output <- target.actual_output, reference_output <- dataset.reference_output

eval experiment results

Fetch per-case results: each dataset case, the target output, and each evaluator’s score and reasoning.

Flag / Argument	Description	Default
`<id>`	Experiment ID.	Required
`-p, --project <name>`	Coze project name; ignored on TEA.	`default`
`--limit <n>`	Items per page; capped at 20.	`20`
`--page <n>`	Page number.	`1`
`--json`	Output raw JSON.	`false`

agentkit eval experiment results 75901xxxxxxxxxxxxx --json

Experiment Status

Status	Meaning
`Success`	All cases finished executing.
`Failed`	Some cases failed to execute, such as when the target did not respond.
`Draining` / `Processing`	Still running.

When a case fails but there is no experiment-level error message, it is usually because the target Runtime did not respond correctly (not deployed, not ready, or authentication failed), rather than an evaluation configuration issue. First confirm that the Runtime pointed to by --target is running.

Getting Started

AI-Friendly

Command reference

eval experiment list

eval experiment show

eval experiment results

Experiment Status