Add user-visible skill loading warnings to conversation UI

Problem

When skills are loaded with configuration conflicts (e.g., a skill defines both paths: and triggers:), the conflict is currently resolved silently. The paths: trigger takes precedence and triggers: is ignored, but users have no visibility into this behavior.

Currently, logger.warning() calls during skill loading only appear in:

• Server logs (not visible to end users)
• Local terminal output (only for SDK developers)

These warnings do not appear in the conversation UI where users would actually see them.

Background

Path-triggered skills (PR #3994) introduce a precedence rule: if a skill defines both paths: and triggers:, the paths: wins and triggers: is silently ignored. This makes the skill path-triggered (not model-invocable).

We discovered that:

• Hooks already have HookExecutionEvent with a visualize property that displays in the UI
• Skills have no equivalent mechanism for surfacing loading warnings
• Users may create conflicting skill definitions without realizing one field is being ignored

Desired Feature

Add infrastructure to surface skill loading warnings/info in the conversation UI, similar to how HookExecutionEvent displays hook execution information.

Timing Considerations

Skills are loaded before the conversation exists:

1. Plugin.load() / AgentContext.__init__
   └─> Skill loading detects conflicts
   └─> No conversation yet to emit events to

2. Conversation.__init__()
   └─> Conversation starts
   └─> Could emit collected warnings here

3. Runtime
   └─> Path rules trigger on file operations

The warning collection and emission would need to be split across these phases.

Related

• PR #3994 - Path-triggered skills implementation
• HookExecutionEvent - Existing pattern for execution-time UI notifications

Start by reading the existing HookExecutionEvent visualization pattern, then trace skill loading through Plugin.load(), AgentContext.init(), and Conversation.init(). Determine how warnings can be collected before a conversation exists and emitted when it starts. Done means conflicting skill definitions produce a visible conversation UI warning without losing the loading-time information.