Problem
Current reminder implementation relies on 'prompt-driven' notification instructions (e.g., ). This leads to:
- Tool Hallucinations: LLMs attempting to use with invalid targets (like ) to satisfy the prompt.
- Silent Failures: The system 'eagerly acknowledges' reminders, meaning if a notification fails due to routing errors, the system thinks the job was a success.
- High Management Overhead: Users must coach the LLM on how to deliver results.
Proposed Solution: Structured Delivery Contract
Shift from 'prompt-based' delivery to 'schema-based' delivery to enable reliable Session Re-entry (Mode B).
1. Strengthen the Tool Schema
Replace/augment with a strongly-typed enum to remove decision-making from the LLM.
- : Automatically routes the LLM's final response back to the originating via the gateway's inbound routing path.
- : Targets a specific channel/user ID.
- : Targets an email address.
2. Fix the 'Eager Ack' Gap
Modify to hold the Akka.Reminders envelope open until the target gateway (Slack, SignalR, etc.) acknowledges receipt of the message.
- This restores the at-least-once delivery guarantee for the actual notification.
3. AI Skill Guardrails
Create a to provide machine-readable instructions:
- "Always use for tasks intended to continue the current conversation."
- "Do not attempt to manually call notification tools for reminder results."
Related Issues
Problem
Current reminder implementation relies on 'prompt-driven' notification instructions (e.g., ). This leads to:
Proposed Solution: Structured Delivery Contract
Shift from 'prompt-based' delivery to 'schema-based' delivery to enable reliable Session Re-entry (Mode B).
1. Strengthen the Tool Schema
Replace/augment with a strongly-typed enum to remove decision-making from the LLM.
2. Fix the 'Eager Ack' Gap
Modify to hold the Akka.Reminders envelope open until the target gateway (Slack, SignalR, etc.) acknowledges receipt of the message.
3. AI Skill Guardrails
Create a to provide machine-readable instructions:
Related Issues