fix: scope the default session waiter identity to the sender

[icyaaaww]

Fixes #9377.

In a group chat, DefaultSessionFilter keyed a session waiter on event.unified_msg_origin alone. That value is identical for every member of the group, so a waiter registered by member A was matched by any other member's next message. handle_session_control_agent then stopped that message (other plugins never saw it), inserted a synthetic At component pointing at the bot, and re-dispatched it through the pipeline — the bot answered an unrelated message from member B as if it were A's follow-up input.

This is reachable in the default configuration: platform_settings.empty_mention_waiting defaults to True and platform_settings.unique_session defaults to False. The empty-mention waiter in astrbot/builtin_stars/astrbot/main.py registers a 60 s waiter without passing a session_filter, so it inherited the group-wide key.

It also contradicts the documented contract. docs/zh/dev/star/guides/session-control.md states that the default identity is based on sender_id and that a custom filter is required to treat a whole group as one session — the opposite of the actual behaviour.

Note that reverting to a sender_id-only key would bring back the defect #1326 fixed, where a waiter registered by one user captured that same user's messages in other groups and private chats. The key needs both parts, which is what this PR does.

Modifications / 改动点

• astrbot/core/utils/session_waiter.py: DefaultSessionFilter.filter() now returns f"{event.unified_msg_origin}!{event.get_sender_id()}" instead of event.unified_msg_origin. Session waiters are isolated per sender and per session, so neither cross-member interception nor cross-session interception is possible. Callers that genuinely want a group-wide session keep doing what the guide already describes: pass a custom SessionFilter.
• tests/unit/test_session_waiter.py (new): regression coverage for the default session identity.

Two lines of behaviour change, no public API change, no new dependency.

Scope note: this PR fixes the cross-member interception the issue reports, and deliberately leaves the two "连带问题" items listed at the end of #9377 alone, since they are separate defects in different code paths. Their blast radius does shrink a lot once the key includes the sender: the unconditional event.stop_event() in handle_session_control_agent can now only swallow a message from the member who registered the waiter (previously any member's image message), and the SessionWaiter._cleanup() pop can now only affect two waiters registered by the same member in the same session. I am happy to send follow-up PRs for either if you would like them handled separately.

• This is NOT a breaking change. / 这不是一个破坏性变更。

The default waiter becomes strictly more selective, and it moves toward the behaviour the developer guide already documents. Plugins that intentionally rely on a group-wide waiter should already be passing their own SessionFilter, as shown in the "自定义会话 ID 算子" section of the guide.

Screenshots or Test Results / 运行截图或测试结果

Verification steps

1. uv sync
2. uv run pytest tests/unit/test_session_waiter.py -v
3. uv run ruff format --check . and uv run ruff check .
4. uv run pytest tests -q to check for regressions

New tests, before the fix (DefaultSessionFilter still returning unified_msg_origin) — the group-chat interception is reproduced:

FAILED tests/unit/test_session_waiter.py::test_default_filter_separates_members_of_the_same_group
FAILED tests/unit/test_session_waiter.py::test_waiter_ignores_other_members_and_accepts_the_owner
2 failed, 2 passed

>       assert triggered == []
E       AssertionError: assert ['member_b'] == []
E         Left contains one more item: 'member_b'

member_b reaching a waiter that member_a registered is exactly the reported bug.

New tests, after the fix:

tests/unit/test_session_waiter.py::test_default_filter_separates_members_of_the_same_group PASSED [ 25%]
tests/unit/test_session_waiter.py::test_default_filter_is_stable_for_the_same_member PASSED      [ 50%]
tests/unit/test_session_waiter.py::test_default_filter_separates_sessions_of_the_same_member PASSED [ 75%]
tests/unit/test_session_waiter.py::test_waiter_ignores_other_members_and_accepts_the_owner PASSED [100%]
======================== 4 passed, 1 warning in 2.63s =========================

The four cases cover: two members of one group get distinct identities; the same member in the same group is stable across messages; one member across two groups and a private chat gets three distinct identities (the #1326 guard); and an end-to-end waiter run where another member's message does not trigger the waiter while the owner's follow-up does, with USER_SESSIONS cleaned up afterwards.

Lint:

$ uv run ruff format --check astrbot/core/utils/session_waiter.py
1 file already formatted
$ uv run ruff check astrbot/core/utils/session_waiter.py
All checks passed!

Full suite: 2132 passed, 34 failed, 1 skipped. The 34 failures are pre-existing and environment-specific to my Windows development machine (file:// URI drive-letter parsing, symlink creation, POSIX shell/python3 invocation). I confirmed they are unrelated by stashing this change and re-running the same selection — the failure set is byte-for-byte identical:

$ git stash push astrbot/core/utils/session_waiter.py
$ uv run pytest tests/test_media_utils.py tests/test_openai_source.py \
    tests/unit/test_computer.py tests/test_preprocess_stage.py \
    tests/unit/test_message_tools.py -q
FAILED tests/test_media_utils.py::test_file_uri_to_path_supports_localhost_and_encoded_paths
FAILED tests/test_openai_source.py::test_file_uri_to_path_preserves_windows_drive_letter
FAILED tests/test_openai_source.py::test_file_uri_to_path_preserves_windows_netloc_drive_letter
FAILED tests/test_openai_source.py::test_file_uri_to_path_preserves_remote_netloc_as_unc_path
FAILED tests/unit/test_computer.py::TestLocalShellComponent::test_exec_with_cwd
FAILED tests/unit/test_computer.py::TestLocalShellComponent::test_exec_with_env
FAILED tests/unit/test_computer.py::TestLocalPythonComponent::test_exec_simple_code
FAILED tests/test_preprocess_stage.py::test_preprocess_path_mapping_accepts_file_uri
FAILED tests/unit/test_message_tools.py::test_send_message_downloads_windows_sandbox_file_with_original_name
9 failed, 166 passed

---

Checklist / 检查清单

• 😊 If there are new features added in the PR, I have discussed it with the authors through issues/emails, etc.
  / 如果 PR 中有新加入的功能，已经通过 Issue / 邮件等方式和作者讨论过。

  No new feature — this is a bug fix for [Bug] 群聊中空提及等待器会截获其他成员的消息（session_waiter 未绑定发送者） #9377, which contains the maintainer-facing root-cause analysis.

• 👀 My changes have been well-tested, and "Verification Steps" and "Screenshots" have been provided above.
  / 我的更改经过了良好的测试，并已在上方提供了“验证步骤”和“运行截图”。

• 🤓 I have ensured that no new dependencies are introduced, OR if new dependencies are introduced, they have been added to the appropriate locations in requirements.txt and pyproject.toml.
  / 我确保没有引入新依赖库，或者引入了新依赖库的同时将其添加到 requirements.txt 和 pyproject.toml 文件相应位置。

• 😮 My changes do not introduce malicious code.
  / 我的更改没有引入恶意代码。

Summary by Sourcery

Prevent default session waiters from intercepting unrelated messages by isolating them per sender and message session.

Bug Fixes:

• Scope default session waiters to both the message origin and sender, preventing messages from other group members or sessions from triggering a waiter.

Tests:

• Add regression tests covering sender isolation within groups, session isolation across chats, stable identities, and end-to-end waiter behavior.

[sourcery-ai]

Hey - I've found 1 issue

Prompt for AI Agents

Please address the comments from this code review:

## Individual Comments

### Comment 1
<location path="tests/unit/test_session_waiter.py" line_range="107-116" />
<code_context>
+@pytest.mark.asyncio
</code_context>
<issue_to_address>
**suggestion (testing):** Add coverage for the timeout/cleanup path of SessionWaiter

This test only covers the success path where the owner’s follow-up arrives and the waiter fires, ending with `USER_SESSIONS` empty. Please add a separate async test for the timeout path: register a waiter with a short timeout, ensure no matching follow-up arrives, then assert that after the timeout (1) the waiter is removed from `USER_SESSIONS`, and (2) non‑owner messages during the wait do not trigger the waiter. That will exercise both success and timeout branches.

Suggested implementation:

```python
    @session_waiter(timeout=5)

@pytest.mark.asyncio
async def test_waiter_times_out_and_ignores_non_owner_messages():
    """A registered waiter times out cleanly and ignores non-owner messages during the wait."""
    USER_SESSIONS.clear()
    session_filter = DefaultSessionFilter()
    owner_event = make_event("member_a", "group_1", text="@bot")
    non_owner_event = make_event("member_b", "group_1", text="@bot")

    triggered: list[str] = []

    @session_waiter(timeout=0.1)
    async def handler(session, event):
        triggered.append(event.message.text)

    # Register the waiter via the owner event
    await handler(session_filter.filter(owner_event), owner_event)

    # Ensure a session is registered for the owner and no one else
    assert "member_a" in USER_SESSIONS
    assert "member_b" not in USER_SESSIONS

    # Send a matching follow-up from a non-owner member; it must not trigger the waiter
    await handler(session_filter.filter(non_owner_event), non_owner_event)
    assert triggered == []

    # Wait for the timeout to elapse
    await asyncio.sleep(0.2)

    # After timeout, the waiter is cleaned up
    assert "member_a" not in USER_SESSIONS

```

1. Ensure `asyncio` is imported at the top of `tests/unit/test_session_waiter.py`:
   ```python
   import asyncio
   ```
   If there is an existing import section, add it there rather than creating a new one.
2. Place the new test near the other `SessionWaiter` tests (e.g., immediately after `test_waiter_ignores_other_members_and_accepts_the_owner`) to keep related coverage together. If the indentation in the snippet is only due to doc formatting, this patch will already do that; otherwise, you may need to re-indent the `@session_waiter(timeout=5)` line so it remains at the correct scope.
3. If `session_waiter` registration semantics differ (e.g., if the first call does not register the waiter), adjust the test to use the correct way of creating and registering a waiter so that `USER_SESSIONS` is populated for `"member_a"` and then cleaned up on timeout.
</issue_to_address>

---

Sourcery is free for open source - if you like our reviews please consider sharing them ✨

• X
• Mastodon
• LinkedIn
• Facebook

_{Help me be more useful! Please click 👍 or 👎 on each comment and I'll use the feedback to improve your reviews.}

[sourcery-ai]

suggestion (testing): Add coverage for the timeout/cleanup path of SessionWaiter

This test only covers the success path where the owner’s follow-up arrives and the waiter fires, ending with USER_SESSIONS empty. Please add a separate async test for the timeout path: register a waiter with a short timeout, ensure no matching follow-up arrives, then assert that after the timeout (1) the waiter is removed from USER_SESSIONS, and (2) non‑owner messages during the wait do not trigger the waiter. That will exercise both success and timeout branches.

Suggested implementation:

    @session_waiter(timeout=5)

@pytest.mark.asyncio
async def test_waiter_times_out_and_ignores_non_owner_messages():
    """A registered waiter times out cleanly and ignores non-owner messages during the wait."""
    USER_SESSIONS.clear()
    session_filter = DefaultSessionFilter()
    owner_event = make_event("member_a", "group_1", text="@bot")
    non_owner_event = make_event("member_b", "group_1", text="@bot")

    triggered: list[str] = []

    @session_waiter(timeout=0.1)
    async def handler(session, event):
        triggered.append(event.message.text)

    # Register the waiter via the owner event
    await handler(session_filter.filter(owner_event), owner_event)

    # Ensure a session is registered for the owner and no one else
    assert "member_a" in USER_SESSIONS
    assert "member_b" not in USER_SESSIONS

    # Send a matching follow-up from a non-owner member; it must not trigger the waiter
    await handler(session_filter.filter(non_owner_event), non_owner_event)
    assert triggered == []

    # Wait for the timeout to elapse
    await asyncio.sleep(0.2)

    # After timeout, the waiter is cleaned up
    assert "member_a" not in USER_SESSIONS

1. Ensure asyncio is imported at the top of tests/unit/test_session_waiter.py:

   import asyncio

   If there is an existing import section, add it there rather than creating a new one.
2. Place the new test near the other SessionWaiter tests (e.g., immediately after test_waiter_ignores_other_members_and_accepts_the_owner) to keep related coverage together. If the indentation in the snippet is only due to doc formatting, this patch will already do that; otherwise, you may need to re-indent the @session_waiter(timeout=5) line so it remains at the correct scope.
3. If session_waiter registration semantics differ (e.g., if the first call does not register the waiter), adjust the test to use the correct way of creating and registering a waiter so that USER_SESSIONS is populated for "member_a" and then cleaned up on timeout.