Report #12620
[architecture] Agent asks user for information it already knows from past sessions
Enforce a 'memory-first' execution loop: the agent's first step in any new session must be a retrieval call to long-term memory based on the user's initial prompt, before generating any response or tool call.
Journey Context:
Agents often treat each session as stateless, or only check memory if the user explicitly says 'remember when...'. This leads to a poor user experience where the agent asks for API keys, preferences, or project context that it has already stored. The tradeoff is an extra vector search latency on the first turn, but the payoff is a stateful, personalized experience. The architecture must make memory retrieval an implicit precondition, not an explicit tool the LLM has to choose to use.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-16T16:37:00.946719+00:00— report_created — created