[FEATURE][Steering] Expand steerable events

[dbschmigelski]

Problem Statement

Currently, the steering system in Strands Agents only supports steering on BeforeToolCallEvent. While the hook system provides many other events (AfterInvocationEvent, BeforeModelCallEvent, AfterModelCallEvent, MessageAddedEvent, etc.), steering handlers cannot hook into these additional events to provide contextual guidance.

This limitation means steering can only influence:

• Tool selection and execution via BeforeToolCallEvent

But cannot provide guidance on:

• Model calls before they happen (BeforeModelCallEvent)
• Response content after model generation (AfterModelCallEvent)
• Message flow as conversations develop (MessageAddedEvent)
• Overall request handling at invocation boundaries (BeforeInvocationEvent, AfterInvocationEvent)

The lack of broader event support prevents steering from addressing use cases like tone control, content validation, conversation flow management, and comprehensive quality assurance across the entire agent lifecycle.

Proposed Solution

No response

Use Case

Response Quality Control (AfterModelCallEvent) allows tone validation to ensure responses meet communication standards, content filtering for sensitive information, and response regeneration when quality criteria aren't met.

Conversation Flow Management (MessageAddedEvent) provides context tracking to monitor topic drift, message validation for format standards, and conversation limits to prevent runaway interactions.

Alternatives Solutions

No response

Additional Context

No response

[clareliguori]

Another use case: I have a workflow where the last thing the agent should do is send a confirmation message via a tool. Often the agent neglects to do this. I can't steer the agent about the LACK of a tool call, and prompt it to send the confirmation prior to the end of the turn.