feat(delegate): wire agent connections through Agent.as_tool

.github/workflows/ci.yml

@@ -26,7 +26,7 @@ jobs:
     steps:
       - uses: actions/checkout@v7
 
-      - uses: astral-sh/setup-uv@v9.0.0
+      - uses: astral-sh/setup-uv@v10.0.1
         with:
           enable-cache: true
 
@@ -59,7 +59,7 @@ jobs:
     steps:
       - uses: actions/checkout@v7
 
-      - uses: astral-sh/setup-uv@v9.0.0
+      - uses: astral-sh/setup-uv@v10.0.1
         with:
           enable-cache: true
           python-version: ${{ matrix.python-version }}

.github/workflows/publish.yml

@@ -40,7 +40,7 @@ jobs:
     steps:
       - uses: actions/checkout@v7
 
-      - uses: astral-sh/setup-uv@v9.0.0
+      - uses: astral-sh/setup-uv@v10.0.1
         with:
           enable-cache: true
 

.gitignore

@@ -1,6 +1,7 @@
 .env
 .claude
 .kiro/specs
+.kiro/settings
 .sessions
 .converter
 .hypothesis

.kiro/skills/library-development/SKILL.md

@@ -35,7 +35,7 @@ load it whenever you are unsure where something goes.
 3. **The pipeline flows one way** — text -> dict -> validated schema -> live objects. Never resolve during parsing; never parse during resolution (see The Pipeline).
 4. **Explicit over implicit** — no auto-registration, no global singletons, no hidden state. Every object is wired by hand and passed as an argument.
 5. **Single responsibility** — each module does one thing; one resolver per config concept, one builder per orchestration mode.
-6. **Composition over inheritance** — small functions and focused modules that compose. The only base classes are the strands-facing ones (`MCPServer`, `StreamConverter`, `HookProvider`).
+6. **Composition over inheritance** — small functions and focused modules that compose. The only base classes are the strands-facing ones (`MCPServer`, `HookProvider`).
 7. **Smallest reasonable change** — don't refactor unrelated code to land a feature.
 
 ---

.kiro/skills/library-development/references/project-map.md

@@ -45,13 +45,12 @@ src/strands_compose/
 ├── tools/
 │   ├── loaders.py       # resolve_tool_spec(s) — module/file/dir → AgentTool
 │   ├── extractors.py    # extract_last_message · serialize_multiagent_result
-│   └── wrappers.py      # node_as_tool / node_as_async_tool — wrap a node as a delegate tool
+│   └── wrappers.py      # multiagent_as_tool — wrap a Swarm/Graph as a delegate tool
 ├── hooks/               # reusable HookProvider implementations
 │   ├── event_publisher.py    # EventPublisher — strands hook events → StreamEvent (the key one)
 │   ├── stop_guard.py         # StopGuard / MultiAgentStopGuard — external cancel signal
 │   ├── max_calls_guard.py    # MaxToolCallsGuard — tool-call circuit breaker
 │   └── tool_name_sanitizer.py# ToolNameSanitizer — repair model-mangled tool names
-├── converters/          # StreamEvent → protocol chunks (base ABC · openai · raw)
 ├── renderers/           # terminal output (base ABC · ansi)
 └── startup/             # opt-in health checks (validator.py) + report (report.py)
 ```

.pre-commit-config.yaml

@@ -1,8 +1,7 @@
 # https://pre-commit.com
 # https://pre-commit.com/hooks.html
 
-# Use the active venv Python — avoids path issues on Windows where
-# pre-commit looks for python3.12 at ~/.local/bin/python.exe
+# Use the active venv Python
 default_language_version:
   python: python3
 
@@ -29,7 +28,7 @@ repos:
       - id: trailing-whitespace
 
   - repo: https://github.com/astral-sh/ruff-pre-commit
-    rev: 'v0.15.20'
+    rev: 'v0.16.3'
     hooks:
       - id: ruff          # lint  — commit + push
       - id: ruff-format   # format — commit + push
@@ -42,7 +41,7 @@ repos:
         files: \.(py|yaml|yml|md|toml|json|env)$
 
   - repo: https://github.com/commitizen-tools/commitizen
-    rev: 'v4.16.4'
+    rev: 'v4.17.0'
     hooks:
       - id: commitizen         # validates commit message format
         stages: [commit-msg]   # only on commit-msg event, not push

AGENTS.md

@@ -7,6 +7,33 @@ subclasses.
 
 ---
 
+## Response Style
+
+Answer with a short executive summary. Lead with the answer or decision in the
+first line.
+
+- Direct question -> 1-3 sentences. Yes/no question -> start with yes or no.
+- Expand only when asked for details, code, a full review, or a rationale.
+- Prefer a compact table or 3-5 bullets over prose sections.
+- Report the outcome, not the journey.
+
+### Do not
+
+- No multi-section reports, headers, or background unless requested.
+- No narrating what was searched, read, or what sub-agents found.
+- No restating earlier turns, no recaps of prior decisions.
+- No storytelling, no "what settled it" / "where I was wrong" essays when one line will do.
+- No praise, filler, or hedging.
+
+### Still required
+
+Brevity never overrides correctness. Keep these even when short:
+
+- State uncertainty plainly, and say what was verified vs assumed.
+- Correct a wrong earlier statement if it would change a decision.
+- Flag risk before a destructive or hard-to-reverse action.
+
+
 ## Read the Skill First — MANDATORY
 
 Before touching any code, load the skill for the area you are working in. Skills

README.md

@@ -80,7 +80,7 @@ Strands Compose is an ecosystem that includes the following packages:
 |-------|---------|-------------|
 | **Define the agents** | [**strands-compose**](https://github.com/strands-compose/sdk-python) | Developers |
 | Run / deploy the agents | [strands-compose-agentcore](https://github.com/strands-compose/bedrock-agentcore) | Developers, operations |
-|*Put the agents in front of people | [strands-compose-chat](https://github.com/strands-compose/chat-ui) | **End users** |
+|Put the agents in front of people | [strands-compose-chat](https://github.com/strands-compose/chat) | **End users** |
 
 ---
 

docs/configuration/Chapter_10.md

@@ -43,7 +43,9 @@ orchestrations:
 entry: team
 ```
 
-**How it works**: strands-compose **forks** a new agent from the `entry_name` agent's blueprint (model, system_prompt, hooks, tools) and adds delegate tools for each connection. The original `coordinator` agent is **never mutated**. Each connection becomes an async tool that the coordinator can call.
+**How it works**: strands-compose **forks** a new agent from the `entry_name` agent's blueprint (model, system_prompt, hooks, tools) and adds delegate tools for each connection. The original `coordinator` agent is **never mutated**.
+
+> **Two limits.** A delegate cannot be called twice concurrently — the second call comes back as a tool error, so declare a second agent to fan out. And an interrupt (such as an approval) can be answered and resumed only for an agent connection; raised inside a nested Swarm or Graph it reaches the coordinator as a tool error.
 
 **Fields**:
 
@@ -54,6 +56,7 @@ entry: team
 | `connections` | list | Yes | Sub-agents to wire as tools |
 | `connections[].agent` | string | Yes | Name of the target agent or orchestration |
 | `connections[].description` | string | Yes | Tool description the LLM sees |
+| `connections[].preserve_context` | bool | No | Keep the delegate's history between calls (default `true`). Set `false` for a stateless delegate that starts from its construction-time baseline every call. Rejected for a nested orchestration, or for an agent carrying a session manager |
 | `session_manager` | dict | No | Override session manager for the forked agent |
 | `hooks` | list | No | Additional hooks for the forked agent |
 | `agent_kwargs` | dict | No | Override agent kwargs (merged with entry agent's kwargs) |

docs/configuration/Chapter_12.md

@@ -53,7 +53,7 @@ entry: manager
 1. strands-compose collects all orchestration dependencies.
 2. It performs a **topological sort** — inner orchestrations are built before outer ones.
 3. Built orchestrations become nodes in the node pool, available for outer orchestrations to reference.
-4. For delegate mode, inner orchestrations are wrapped as async tools (just like regular agents).
+4. For delegate mode, inner orchestrations become callable tools on the coordinator.
 
 ## Circular Dependencies
 

examples/12_streaming/README.md

@@ -57,8 +57,8 @@ bracketing all per-agent activity.
 `invoke_async` so both the agent and the queue consumer share the same event loop.
 
 **`AnsiRenderer` is optional.** It's a convenience for terminals. In production you'd
-consume the queue and convert events to SSE chunks (see `OpenAIStreamConverter`) or
-NDJSON (`RawStreamConverter`).
+consume the queue and serialize each event into whatever wire format your transport
+needs — SSE chunks, NDJSON, or your own envelope.
 
 **`queue.flush()`** resets the queue between turns so events from one invocation
 don't leak into the next. It also resets the `session_start` / `session_end` guards.

pyproject.toml

@@ -22,7 +22,7 @@ classifiers = [
     "Typing :: Typed",
 ]
 dependencies = [
-    "strands-agents>=1.48.0,<2.0.0",
+    "strands-agents>=1.52.0,<2.0.0",
     "pydantic>=2.12.5",
     "pyyaml>=6.0.0",
     "mcp>=1.24.0",
@@ -36,16 +36,16 @@ agentcore-memory = [
     "bedrock-agentcore>=1.4.0",
 ]
 ollama = [
-    "strands-agents[ollama]>=1.48.0,<2.0.0",
+    "strands-agents[ollama]>=1.52.0,<2.0.0",
 ]
 openai = [
-    "strands-agents[openai]>=1.48.0,<2.0.0",
+    "strands-agents[openai]>=1.52.0,<2.0.0",
 ]
 gemini = [
-    "strands-agents[gemini]>=1.48.0,<2.0.0",
+    "strands-agents[gemini]>=1.52.0,<2.0.0",
 ]
 anthropic = [
-    "strands-agents[anthropic]>=1.48.0,<2.0.0",
+    "strands-agents[anthropic]>=1.52.0,<2.0.0",
 ]
 
 [project.urls]
@@ -57,16 +57,16 @@ Changelog = "https://github.com/strands-compose/sdk-python/blob/main/CHANGELOG.m
 
 [dependency-groups]
 dev = [
-    "ty>=0.0.29",
+    "ty>=0.0.72",
     "bandit>=1.9.2",
-    "coverage>=7.12.0",
+    "coverage>=7.15.4",
     "pytest-asyncio>=1.2.0",
     "pytest-cov>=7.0.0",
     "pytest-mock>=3.15.1",
     "pytest-xdist>=3.8.0",
-    "pytest>=9.0.2",
-    "ruff>=0.14.8",
-    "rust-just>=1.42.4",
+    "pytest>=9.1.1",
+    "ruff>=0.16.3",
+    "rust-just>=1.58.0",
     "commitizen>=4.8.4",
     "pre-commit>=4.3.0",
     "hypothesis>=6.155.7",

src/strands_compose/__init__.py

@@ -24,8 +24,7 @@
 from .mcp import MCPLifecycle, create_mcp_client, create_mcp_server
 from .renderers import AnsiRenderer
 from .tools import (
-    node_as_async_tool,
-    node_as_tool,
+    multiagent_as_tool,
     serialize_multiagent_result,
 )
 from .types import EventType, StreamEvent
@@ -59,8 +58,7 @@
     "load_config",
     "load_session",
     "make_event_queue",
-    "node_as_async_tool",
-    "node_as_tool",
+    "multiagent_as_tool",
     "resolve_infra",
     "serialize_multiagent_result",
 ]