AWS bedrock-agentcore documentation change
Summary
Updated CLI command documentation with new flags including --kms-key for encryption and added job management commands.
Security assessment
Added --kms-key flag documentation enables customer-managed encryption keys for batch evaluation results, which is a security feature. However, no specific vulnerability is being addressed.
Diff
diff --git a/bedrock-agentcore/latest/devguide/batch-evaluations-start.md b/bedrock-agentcore/latest/devguide/batch-evaluations-start.md index 88b703e0e..275e391bc 100644 --- a//bedrock-agentcore/latest/devguide/batch-evaluations-start.md +++ b//bedrock-agentcore/latest/devguide/batch-evaluations-start.md @@ -34 +34 @@ With optional flags: - --name "baseline-eval" \ + --name my_baseline_eval \ @@ -49 +49,22 @@ With optional flags: -The CLI polls until the job reaches a terminal state (`COMPLETED`, `FAILED`, or `STOPPED`), displays per-evaluator average scores, and saves results to `.cli/eval-job-results/`. +By default the command starts the job and returns immediately. Pass `--wait` to block until the job reaches a terminal state (`COMPLETED`, `FAILED`, or `STOPPED`), after which the CLI displays per-evaluator average scores and saves results to `.cli/jobs/batch-eval-results/`. + +`agentcore run batch-evaluation` also supports the following flags: + + * `--wait` — block until the job reaches a terminal state. + + * `--json` — emit machine-readable JSON output. + + * `--kms-key <arn>` — encrypt batch evaluation results with a customer-managed KMS key. + + * `--dataset <name>` / `--dataset-version <version>` — invoke the agent with dataset scenarios before batch evaluation (omit the version for a local file, or use `N`/`DRAFT`). + + * `--endpoint <name>` — target a specific runtime endpoint (for example, `PROMPT_V1`); defaults to the `AGENTCORE_RUNTIME_ENDPOINT` environment variable, then `DEFAULT`. + + * `--evaluator-arn <arns…>` — reference evaluators by ARN instead of `-e`. + +Most flags have short aliases: `-r` (`--runtime`), `-e` (`--evaluator`), `-n` (`--name`), `-d` (`--lookback-days`), `-s` (`--session-ids`), and `-g` (`--ground-truth`). + +To manage a job after it starts, run `agentcore stop batch-evaluation -i <id>` to stop a running job and `agentcore archive batch-evaluation -i <id>` to archive a job record. + + +