Verify each finding against the current code and only fix it if needed.

Inline comments:
In `@interface/src/components/ChannelEditModal.tsx`:
- Around line 573-581: The checkbox label in ChannelEditModal (the input bound
to bindingForm.require_mention and updated via setBindingForm) is not associated
with its input for accessibility; add an explicit association by giving the
checkbox an id (e.g., requireMentionCheckbox) and reference that id from the
label's htmlFor attribute, or wrap the input element inside the label element so
clicking the text toggles the checkbox and screen readers correctly link the
label to the input.

In `@interface/src/components/ChannelSettingCard.tsx`:
- Around line 1175-1183: The checkbox label in ChannelSettingCard isn’t
associated with its input; update the JSX so the label is programmatically bound
to the input (either give the input a unique id and set the label’s htmlFor to
that id, or wrap the <input> with the <label>) for the checkbox controlling
bindingForm.require_mention; ensure the onChange and checked props remain
unchanged and that the id is unique within the component to preserve
accessibility and click-target behavior.

---

Nitpick comments:
In `@src/agent/channel.rs`:
- Around line 1431-1447: Extract the duplicate "suppress-path" behavior
(injecting MentionOnly entries into in-memory history and performing passive
memory capture) into a shared helper and call it from both places where it
currently appears; specifically, create a helper function (e.g.,
inject_mention_only_and_capture) that takes &mut self, an iterator over
pending_batch_entries (or a single formatted_text), and message_count, performs
the history.push into self.state.history.write().await for
ResponseMode::MentionOnly, increments self.message_count, and calls
self.check_memory_persistence().await; then replace the duplicated blocks that
check for ResponseMode::MentionOnly and update
message_count/check_memory_persistence with calls to this helper (references:
ResponseMode::MentionOnly, pending_batch_entries, state.history, message_count,
check_memory_persistence).
- Around line 1444-1447: Add regression tests covering the new MentionOnly async
suppression behavior: write unit/integration tests that exercise both batched
and single-message flows to assert history ingestion occurs and passive memory
capture is triggered (i.e., message_count increments and
check_memory_persistence() is invoked) when ChannelMode::MentionOnly is used;
target the async/stateful paths exercised by the code that updates
self.message_count and calls check_memory_persistence(), and include assertions
that ensure no suppression bypasses these calls and that behavior matches Quiet
mode for passive capture. Ensure tests simulate relevant async timing/race
conditions (awaits, multiple concurrent messages) and cover the other affected
lines (around the similar block at 1870-1873) to prevent regressions.

Verify each finding against the current code and only fix it if needed.

Inline comments:
In `@interface/src/components/ChannelSettingCard.tsx`:
- Line 1184: Update the Mention Only helper text in ChannelSettingCard
(component ChannelSettingCard, the paragraph currently saying "Blocks messages
entirely — the agent won't see them at all. To let the agent read all messages
but only respond to mentions, use Mention Only mode in Channels settings (cog
icon).") to clarify that Mention Only still allows command invocations in
addition to mentions — e.g., change the copy to indicate the agent will only
reply to `@mentions` or slash/command invocations while still reading all
messages, so users can distinguish channel-level Mention Only from binding-level
require_mention.

In `@src/agent/channel.rs`:
- Around line 1431-1453: Avoid awaiting compaction/persistence on the
suppression fast-path: instead of awaiting
self.compactor.check_and_compact().await and
self.check_memory_persistence().await inline (inside the
ResponseMode::MentionOnly/suppression branch after pushing pending_batch_entries
and incrementing self.message_count), spawn a detached background task to run
those checks so the suppression path remains non-blocking; clone/move the
minimal pieces needed into the task (e.g., Arc::clone(&self) or
self.compactor.clone()) and call compactor.check_and_compact().await and
self.check_memory_persistence().await inside tokio::spawn (or equivalent) so
errors can be logged but do not block the channel.