Problem
The Claude Code MCP plugin does not expose the metadata field when creating RELEASE_READINESS_REVIEW backlog tasks. The DevOps Agent API (https://docs.aws.amazon.com/boto3/latest/reference/services/devops-agent/client/get_backlog_task.html) includes a metadata document field on task objects, which can carry reviewer context (what changed, what to focus on, what's already been accepted).
Without the ability to pass metadata on review tasks, every review is effectively greenfield — there's no mechanism to make iterative re-reviews converge instead of re-litigating prior decisions.
Kiro's DevOps Agent integration already exposes this field.
Expected Behavior
When creating a RELEASE_READINESS_REVIEW backlog task through the Claude Code plugin, users should be able to provide structured metadata (e.g., reviewerContext) so the agent can scope its review to what actually changed rather than starting from scratch each time.
Problem
The Claude Code MCP plugin does not expose the metadata field when creating RELEASE_READINESS_REVIEW backlog tasks. The DevOps Agent API (https://docs.aws.amazon.com/boto3/latest/reference/services/devops-agent/client/get_backlog_task.html) includes a metadata document field on task objects, which can carry reviewer context (what changed, what to focus on, what's already been accepted).
Without the ability to pass metadata on review tasks, every review is effectively greenfield — there's no mechanism to make iterative re-reviews converge instead of re-litigating prior decisions.
Kiro's DevOps Agent integration already exposes this field.
Expected Behavior
When creating a RELEASE_READINESS_REVIEW backlog task through the Claude Code plugin, users should be able to provide structured metadata (e.g., reviewerContext) so the agent can scope its review to what actually changed rather than starting from scratch each time.