[Yarmoluk]

Description

Moves samples/python/agents/number_guessing_game from a2a-sdk==0.3.0 to a2a-sdk[http-server]==1.1.0, following the 1.0 migration guide and the helloworld 1.1.x sample. The game, the file layout and the agents' output stay the same.

Changes, all inside the sample folder:

• requirements.txt: a2a-sdk[http-server]==1.1.0. In 1.x, Starlette comes only with the http-server extra.
• Agent cards: AgentCard / AgentSkill / AgentInterface constructors. supported_interfaces replaces url, and stateTransitionHistory is dropped because 1.x no longer has it.
• Executors (Alice, Carol): a new conversation enqueues its Task first, as 1.x requires. For Alice this replaces updater.submit(). Also: Part(text=...), requires_input() (no final argument), and imports from a2a.helpers / a2a.server.*.
• utils/server.py: DefaultRequestHandler(..., agent_card), and create_agent_card_routes + create_jsonrpc_routes in a Starlette app (replaces A2AStarletteApplication). The JSON-RPC path stays /a2a/v1.
• utils/protocol_wrappers.py:
  • create_client() per call. In 1.x a ClientFactory keeps one httpx.AsyncClient, which fails with "Event loop is closed" because each sync wrapper runs its own asyncio.run.
  • SendMessageRequest and CancelTaskRequest.
  • The Task is read from the StreamResponse, since ClientTaskManager is gone.
• agent_Bob.py: TaskState.TASK_STATE_INPUT_REQUIRED. States are printed with TaskState.Name(...) so the log stays readable (state=TASK_STATE_INPUT_REQUIRED).
• README: SDK version in the coverage heading, SendMessage, and the task aggregation row.
• ruff format (repo .ruff.toml) on the touched files. This accounts for most of the line count in Bob and Carol.

SDK version: a2a-sdk 1.1.0, the same exact pin as helloworld, multitenancy and sign_and_verify_agent_card.

How it was tested

Python 3.12, fresh venv:

cd samples/python/agents/number_guessing_game
python -m venv .venv && source .venv/bin/activate
pip install -r requirements.txt
python agent_Alice.py &   # port 8001
python agent_Carol.py &   # port 8003
python agent_Bob.py       # guesses typed in; scripted by bisection for the transcripts

The game played to the end on 1.1.0. Last lines of Bob's output:

Alice says: correct! attempts: 6
...
=== Carol's visualisation (sorted) ===
Guesses so far:
  1.  25 -> Go higher
  2.  31 -> Go higher
  3.  34 -> Go higher
  4.  35 -> correct! attempts: 6
  5.  37 -> Go lower
  6.  50 -> Go lower
============================

You won! Exiting…

The "Try again" / "Well done!" multi-turn exchange between Bob and Carol completed as on 0.3.0. In a second run, the 400-attempt limit was reached and Bob's CancelTask reached Carol ([Carol] Task … canceled on request of peer agent). The same scripted game was played on 0.3.0 before the change for comparison. The only difference in output is the state name (TaskState.input_required → TASK_STATE_INPUT_REQUIRED).

ruff format --check --config .ruff.toml samples/python/agents/number_guessing_game/ passes. ruff check reports nothing new: 47 findings remain, all of which were already on main (48 there). They are mostly docstring style, EN DASH characters and missing __init__.py (INP001) in files this PR does not otherwise need to touch.

Not tested: Alice's cancel path (nothing calls it), Windows, Python 3.10/3.11, and a 0.3 client against the new servers (v0.3 compat is not enabled).

AI assistance was used to prepare this change. The game was run end to end locally on the new pin.

• Follow the CONTRIBUTING Guide.

Part of #462 🦕

🤖 Generated with Claude Code

https://claude.ai/code/session_015Nn19mXGNFoFZgk3SaYC7Y