fix(agentchat): MessageFilterAgent count=0 returns no messages instead of all

[nolanchic]

Why are these changes needed?

In MessageFilterAgent._apply_filter, the per-source slice guards used a truthiness check on count:

if source_filter.position == "first" and source_filter.count:   # <- 0 is falsy
    msgs = msgs[: source_filter.count]
elif source_filter.position == "last" and source_filter.count:  # <- 0 is falsy
    msgs = msgs[-source_filter.count :]

Since 0 is falsy, count=0 skipped slicing entirely and returned all messages from that source — the opposite of the intended behavior.

The class docstring states:

If position is None, all messages from that source are included.

So the intended way to keep all messages is position=None; a count=0 with position="first"/"last" should mean zero messages. (PerSourceFilter.count is Optional[int], so 0 is a valid, explicit user input.)

There is a second, subtler issue: even after switching to is not None, the last branch would still return all messages for count=0, because msgs[-0:] is equivalent to msgs[0:] (the whole list) in Python.

Fix

Handle count <= 0 explicitly by returning an empty list, and switch the remaining guards from truthiness to is not None:

if source_filter.count is not None and source_filter.count <= 0:
    # count == 0 means "no messages" (and negative values are not meaningful).
    # Note: msgs[-0:] would return the whole list, so handle this explicitly.
    msgs = []
elif source_filter.position == "first" and source_filter.count is not None:
    msgs = msgs[: source_filter.count]
elif source_filter.position == "last" and source_filter.count is not None:
    msgs = msgs[-source_filter.count :]

Behavior summary:

count	position	before	after
0	first / last	all messages	no messages
None	any	unchanged (all messages)	unchanged (all messages)
>0	first / last	unchanged	unchanged

Related issue number

N/A — no existing issue. Edge-case correctness fix; happy to open a tracking issue if maintainers prefer.

Checks

• I've included any doc changes needed for https://microsoft.github.io/autogen/. (None — internal logic fix.)
• I've added tests (if relevant) corresponding to the changes introduced in this PR. (New tests/test_message_filter_agent.py covering count=0 for both first/last, count=None preserving all messages, and positive-count sanity. The count=0 last-position case fails on main and passes with the fix.)
• I've made sure all auto checks have passed. (Local verification below.)

Local verification

cd python
uv run pytest packages/autogen-agentchat/tests/test_message_filter_agent.py -q                  # 4 passed
uv run pytest packages/autogen-agentchat/tests/test_group_chat_graph.py -q                       # 76 passed (no regressions)
uv run ruff format --check packages/autogen-agentchat/src packages/autogen-agentchat/tests       # already formatted
uv run ruff check packages/autogen-agentchat/src packages/autogen-agentchat/tests                # all checks passed
uv run pyright packages/autogen-agentchat/src packages/autogen-agentchat/tests                   # 0 errors
uv run mypy --config-file pyproject.toml packages/autogen-agentchat/src packages/autogen-agentchat/tests  # no issues

[nolanchic]

Added a follow-up commit to harden the boundary validation:

• Constrain PerSourceFilter.count with Field(default=None, ge=0) so negative values are rejected at construction (fail-fast at the model boundary) instead of being silently coerced inside _apply_filter.
• Since negatives can no longer reach _apply_filter, tightened the defensive count <= 0 branch to count == 0 (still needed because msgs[-0:] returns the whole list).
• Added a docstring to PerSourceFilter clarifying the count semantics: 0 → no messages, None → do not slice.
• Added a test asserting count=-1 raises ValidationError at construction.

All checks still green: 81 passed (5 message-filter + 76 graph), ruff/pyright/mypy clean.