Optimization: Increase Memory Search Trigger Frequency

[gaoteng-git]

Priority: P1

Summary

The current memory search trigger frequency in EverMemOS is insufficient, causing some relevant historical context to be missed during Agent
retrieval. This issue explores increasing search coverage by tuning the search skill and introducing a user prompt hook.

TODO

• Tune the trigger conditions and prompt of the existing search skill to encourage the Agent to invoke search more proactively
• Following the session start hook implementation, add search logic to the user prompt hook so that results are automatically injected into
  the system prompt

Proposed Approaches

Approach	Description
Tune skill	Modify the skill's trigger description or prompt to guide the Agent to call search more aggressively
User prompt hook	Automatically trigger search on every user message and inject results into the system prompt, without relying on the Agent
to call search explicitly

Notes

• Priority P1; does not affect current baseline functionality
• The user prompt hook approach is similar to the session start hook and has high reusability — recommended to try first
• Be mindful of the impact on token consumption and response latency as search frequency increases