feat(templates): add agui-python-strands template

[Hweinstock]

Spec

Add support for an AG-UI Strands Python template to the refactor architecture.

Problem

The refactor branch does not yet support AGUI templates.
AGUI + runtime: https://docs.aws.amazon.com/bedrock-agentcore/latest/devguide/runtime-agui.html · AGUI: https://docs.ag-ui.com/introduction

Definition of Done

• Customers can use the template and have it render the Strands AG-UI assets into their project on both project creation and add runtime.
• The template supports containers and memory, and no other context (to keep it simple).

Verification

• Create a project with the template, add a runtime with the container version, run the dev server with both, curl them locally, then deploy and invoke them e2e.
• Verify the template shows up in the TUI create flow, and take a screenshot.

Notes

Based on aws#2170.

Naming note: the spec drafted the name as agui-strands-python, but this PR is stacked on aws#2170, whose whole purpose is a consistent <category>-<language>-<framework> scheme (agent-python-strands, a2a-python-strands, mcp-python-fastmcp). To stay consistent with the convention aws#2170 establishes, the template ships as agui-python-strands (confirmed with the requester).

What changed

• src/assets/templates/agui-python-strands/ — new template assets: an AG-UI Strands agent (create_strands_app serving /invocations + /ping on 8080), Bedrock-only model/load.py, memory wiring (memory/session.py), pyproject.toml, and Dockerfile/.dockerignore/.gitignore (byte-identical to the a2a-python-strands siblings).
• src/core/project/templates/runtime.ts — resolver for strands/Python/AGUI, mirroring the A2A resolver but with only the simple context (name, memory, otel, entrypoint) — no OS/filesystem-mount/identity context, per the spec.
• src/handlers/project/shortcuts.ts — agui-python-strands shortcut (Python · strands · AGUI · Bedrock · long+short-term memory, CodeZip default).
• src/handlers/project/create/screen.tsx — the template appears in the interactive create wizard.
• Tests — manager.test.ts and add/runtime/index.test.ts cover create, add runtime, --build Container, and --memory for the new template.

AGUI was already a first-class protocol in the refactor branch (ProtocolModeSchema, DEV_PORTS.AGUI=8080, and the dev inspector), so no changes were needed there.

Verification evidence

All account IDs / ARNs redacted.

1. project create renders the template (AGUI protocol + memory)

$ agentcore project create --name AguiTest --template agui-python-strands --skip-git
Created project 'AguiTest' in ./AguiTest
# agentcore.json runtimes[0]:
{ "name": "agui_python_strands", "build": "CodeZip", "entrypoint": "main.py",
  "codeLocation": "app/agui_python_strands", "runtimeVersion": "PYTHON_3_14", "protocol": "AGUI" }
# memories[0].name = "agui_python_strandsMemory" (SEMANTIC, USER_PREFERENCE, SUMMARIZATION, EPISODIC)

2. add runtime with the container version

$ agentcore project add runtime --name agui_container --template agui-python-strands --build Container
added runtime 'agui_container' to 'AguiTest'
# runtimes[1]: { "name": "agui_container", "build": "Container", "dockerfile": "Dockerfile", "protocol": "AGUI", ... }
# app/agui_container/ contains a rendered Dockerfile

3. Dev server + local curl — CodeZip runtime (full Bedrock round-trip)

$ agentcore project dev --mode headless --agent agui_python_strands --port 8090 --no-traces &
[agui_python_strands] Uvicorn running on http://0.0.0.0:8090
$ curl localhost:8090/ping                 ->  {"status":"healthy"}
$ curl -N -X POST localhost:8090/invocations -d '{"threadId":"t1","runId":"r1","state":{},
    "messages":[{"id":"m1","role":"user","content":"Reply with exactly: hello from agui"}],
    "tools":[],"context":[],"forwardedProps":{}}'
data: {"type":"RUN_STARTED",...}
data: {"type":"TEXT_MESSAGE_CONTENT","delta":"hello"}
data: {"type":"TEXT_MESSAGE_CONTENT","delta":" from agui"}
data: {"type":"RUN_FINISHED","outcome":{"type":"success"}}

4. Dev server + local curl — Container runtime — image builds, container serves AG-UI, /ping healthy and the SSE stream flows. (The local container uses the desktop instance-profile role, which lacks Bedrock, so the model call is blocked locally; the deployed container below proves the Bedrock path end-to-end.)

5. Deploy + invoke both runtimes end-to-end

$ agentcore project deploy
Deployed project 'AguiTest' to target 'default'
# CodeZip runtime:
$ aws bedrock-agentcore invoke-agent-runtime --agent-runtime-arn <codezip-arn> --payload '{...RunAgentInput...}' out.json
{ "contentType": "text/event-stream; charset=utf-8", "statusCode": 200 }
  -> data: {"type":"RUN_FINISHED","outcome":{"type":"success"}}
# Container runtime:
$ aws bedrock-agentcore invoke-agent-runtime --agent-runtime-arn <container-arn> --payload '{...}' out.json
{ "contentType": "text/event-stream; charset=utf-8", "statusCode": 200 }
  -> data: {"type":"TEXT_MESSAGE_CONTENT","delta":"hello"} ... {"delta":" deployed"}
  -> data: {"type":"RUN_FINISHED","outcome":{"type":"success"}}

6. TUI create flow — the template appears and is selectable

 ✓ name ── ✓ type ── ● template ── ○ review
 choose a template
 ╭──────────────────────────────────────────────────────────────────────────────╮
 │ ○ agent-python-strands (recommended)  Strands agent on Bedrock with memory     │
 │ ○ agent-python                        minimal Python agent on Bedrock          │
 │ ○ mcp-python-fastmcp                  MCP server exposing tools via FastMCP     │
 │ ○ a2a-python-strands                  Strands agent speaking the A2A protocol   │
 │ ● agui-python-strands                 Strands agent speaking the AG-UI protocol │
 ╰──────────────────────────────────────────────────────────────────────────────╯

How to reproduce

bun install
bun run compile:linux-x64          # or your platform target
BIN=./dist/bin/agentcore-linux-x64

# scaffold + container runtime
$BIN project create --name AguiTest --template agui-python-strands --skip-git
cd AguiTest
$BIN project add runtime --name agui_container --template agui-python-strands --build Container

# run locally + curl (repeat with --agent agui_container)
$BIN project dev --mode headless --agent agui_python_strands --port 8090 --no-traces &
curl localhost:8090/ping
curl -N -X POST localhost:8090/invocations -H 'Content-Type: application/json' \
  -d '{"threadId":"t1","runId":"r1","state":{},"messages":[{"id":"m1","role":"user","content":"hi"}],"tools":[],"context":[],"forwardedProps":{}}'

# deploy + invoke (needs AWS creds with Bedrock access)
$BIN project deploy

Tests: bun test (2740 pass), bun run typecheck, bun run lint:check, bun run format:check all green.

[Hweinstock]

we should advertise that our cli can invoke with agentcore project invoke runtime ... lets verify a command on a real deployed agent and update the docs here.