Description
agentcore export harness generates a bare Strands Agent with hand-written shell and file_operations tools. The harness builds its agents with Strands Harness (create_harness): the SDK's sandbox-routed shell, the Strands Harness read/write/edit tools, web_fetch, the todos and context-offloader plugins, and a subagent tool, with execution limits shared by the agent and its subagents. An exported agent therefore offers different tools and behaves differently from the harness it came from.
Acceptance Criteria
- The exported agent is built with
create_harness and the same built-in tools and plugins, each gated by allowedTools.
- A
subagent tool delegates to a child that can use the agent's built-in and MCP tools.
maxIterations/maxTokens cover the agent and its subagents together and end with the harness stop reasons.- Unset prompt and truncation take the harness defaults.
Additional Context
Follows #2456, which aligned allowedTools matching and model parameters.
Description
agentcore export harnessgenerates a bare StrandsAgentwith hand-writtenshellandfile_operationstools. The harness builds its agents with Strands Harness (create_harness): the SDK's sandbox-routedshell, the Strands Harnessread/write/edittools,web_fetch, the todos and context-offloader plugins, and asubagenttool, with execution limits shared by the agent and its subagents. An exported agent therefore offers different tools and behaves differently from the harness it came from.Acceptance Criteria
create_harnessand the same built-in tools and plugins, each gated byallowedTools.subagenttool delegates to a child that can use the agent's built-in and MCP tools.maxIterations/maxTokenscover the agent and its subagents together and end with the harness stop reasons.Additional Context
Follows #2456, which aligned
allowedToolsmatching and model parameters.