Problem 1: --evaluator-arn accepts non-ARN values
The --evaluator-arn flag accepts both actual ARNs and managed evaluator IDs like Builtin.Correctness. The flag name implies it only takes ARNs, which is confusing for users.
# "Builtin.Correctness" is not an ARN, but --evaluator-arn accepts it
agentcore run eval \
--runtime-arn arn:aws:bedrock-agentcore:us-west-2:123456:runtime/my-agent \
--evaluator-arn "Builtin.Correctness"
Problem 2: Users without a project can't use Builtin evaluators via --evaluator
Users who don't have a CLI project (e.g. running evals against a runtime ARN directly) can't use --evaluator "Builtin.Correctness" — it fails with:
No agentcore project found.
Run agentcore create to fix this.
This is because the project requirement is only skipped when both --runtime-arn and --evaluator-arn are provided:
const isArnMode = !!(cliOptions.runtimeArn && cliOptions.evaluatorArn);
Users have to know to use --evaluator-arn instead of --evaluator for Builtin evaluators when outside a project — but nothing tells them this.
Suggestion
- Rename
--evaluator-arn to --evaluator-id to reflect that it accepts both ARNs and managed IDs
- Only require
--runtime-arn to skip the project check, so --evaluator works with Builtin evaluators outside a project
Problem 1:
--evaluator-arnaccepts non-ARN valuesThe
--evaluator-arnflag accepts both actual ARNs and managed evaluator IDs likeBuiltin.Correctness. The flag name implies it only takes ARNs, which is confusing for users.Problem 2: Users without a project can't use Builtin evaluators via
--evaluatorUsers who don't have a CLI project (e.g. running evals against a runtime ARN directly) can't use
--evaluator "Builtin.Correctness"— it fails with:This is because the project requirement is only skipped when both
--runtime-arnand--evaluator-arnare provided:Users have to know to use
--evaluator-arninstead of--evaluatorfor Builtin evaluators when outside a project — but nothing tells them this.Suggestion
--evaluator-arnto--evaluator-idto reflect that it accepts both ARNs and managed IDs--runtime-arnto skip the project check, so--evaluatorworks with Builtin evaluators outside a project