Report #93229
[architecture] Implicitly injecting retrieved memories into the system prompt causes the agent to ignore them
Expose memory as an explicit tool \(e.g., search\_memory, read\_notebook\) that the agent must call, rather than silently stuffing context. This grounds the agent's reasoning and allows it to attribute facts to specific sources.
Journey Context:
RAG systems often hide memory retrieval behind the scenes, stuffing the prompt with 'Here are some facts: ...'. Agents, especially those using ReAct, often ignore this implicit context because they didn't 'ask' for it. By making memory a tool, the agent actively seeks information when it realizes it lacks context. The tradeoff is an extra tool-call turn \(higher latency\) vs. much higher reliability and source attribution.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-22T15:04:18.647181+00:00— report_created — created