Problem
When a user reveals personal info alongside another request (e.g., "you have platinum hair similar to me"), the agent only acts on the primary request (updating persona memory) but misses storing the user fact (that the user has platinum hair) in human memory.
Evidence
Agent trace from 2026-03-31: User said "from now on you have platinum hair similar to me." Agent called core_memory_append(label="persona", ...) but never called update_human_memory to record the user's platinum hair.
Possible fixes
- Add system prompt guidance: "When the user reveals personal information about themselves, even as part of another request, store it in human memory"
- Add a post-processing step that scans user messages for self-referential facts
- Include examples in the system prompt showing dual tool calls for mixed requests
Files
- System prompt construction in
services/agent/memory_blocks.py or services/agent/service.py
Problem
When a user reveals personal info alongside another request (e.g., "you have platinum hair similar to me"), the agent only acts on the primary request (updating persona memory) but misses storing the user fact (that the user has platinum hair) in human memory.
Evidence
Agent trace from 2026-03-31: User said "from now on you have platinum hair similar to me." Agent called
core_memory_append(label="persona", ...)but never calledupdate_human_memoryto record the user's platinum hair.Possible fixes
Files
services/agent/memory_blocks.pyorservices/agent/service.py