feat(export): build exported harness agents with Strands Harness

[amhegazy-mp4]

Description

agentcore export harness now builds the exported agent the way the harness runtime does: strands_harness.create_harness() with its defaults turned off and each capability passed explicitly, instead of a bare Strands Agent with hand-written tools. The construction lives in a new generated module, harness_runtime.py.

• Built-in tools. The SDK's sandbox-routed shell, the Strands Harness read/write/edit, and web_fetch in markdown mode, each gated by allowedTools (file_operations still grants read/write/edit).
• Plugins. todo_write (todos) and retrieve_offloaded_content (a context offloader that keeps large tool results on disk per session and never evicts them).
• Subagent. A subagent tool delegates to a generalist child that can use the agent's built-in and MCP tools, gets only the plugins it selects, has no session manager, and cannot delegate further.
• Shared limits. One budget, passed in the invocation state, covers the agent and its subagents. maxIterations/maxTokens end the invocation with max_iterations_exceeded / max_output_tokens_exceeded. With the subagent tool enabled, usage is reported once per invocation including subagents; otherwise per model call.
• MCP. Tools are prefixed with their server name (<server>_<tool>), as in the harness, so they cannot collide with the built-ins.
• Defaults when unset. The Strands Harness contract prompt and summarization truncation (summary_ratio 0.3, preserve_recent_messages 10). Bedrock Converse models cache the system prompt and tools unless the model parameters already place a cache point.
• Dependencies. Pins strands-agents ~= 1.57.1 and strands-harness == 0.1.2.

Known differences from the harness: shell and file tools run on the runtime host rather than in a container sandbox, offloaded results have no per-session size cap, and tool names are not shortened to fit Bedrock's 64-character limit.

Related Issue

Closes #2485

Documentation PR

Not applicable.

Type of Change

• Bug fix
• New feature
• Breaking change
• Documentation update
• Other (please describe):

Testing

• I ran bun test
• I ran the relevant end-to-end tests with bun run test:e2e, or explained why they are not applicable: there are no export end-to-end tests. Instead I rendered 14 harness shapes (all built-ins with limits, none, subagent only, each plugin alone, inline functions with and without MCP and a timeout, S3 skills, memory, no truncation); each passes pyflakes and py_compile and builds an agent against the pinned packages. I also ran exports against Bedrock with a local MCP server: @mcp/tool filtering, subagent delegation with context offloading, the shared iteration limit, a timeout cancelling a running subagent, and an inline function handoff.
• I ran bun run typecheck
• I ran bun run lint:check
• I ran bun run format:check
• I ran bun run build
• If I modified src/assets/, I updated affected snapshots with bun test <test-file> --update-snapshots and committed them (no snapshots changed)

Checklist

• I have read the CONTRIBUTING document
• I have added any necessary tests that prove my fix is effective or my feature works
• I have updated the documentation accordingly
• I have added an appropriate example to the documentation to outline the feature, or no new docs are needed
• My changes generate no new warnings
• Any dependent changes have been merged and published

---

By submitting this pull request, I confirm that you can use, modify, copy, and redistribute this contribution, under the
terms of your choice.

[codecov-commenter]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 97.40%. Comparing base (04b45c3) to head (4c46a27).

Additional details and impacted files

@@            Coverage Diff            @@
##           refactor    #2486   +/-   ##
=========================================
  Coverage     97.39%   97.40%           
=========================================
  Files           642      642           
  Lines         46807    46854   +47     
=========================================
+ Hits          45590    45638   +48     
+ Misses         1217     1216    -1     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.