OpenClaw memory injection behavior for daily logs

Determine from the OpenClaw source code whether daily log files (memory/YYYY-MM-DD.md) are automatically injected by the runtime into the model’s context or are only loaded via instruction-driven behavior specified in AGENTS.md.

Background

The paper explains that curated memory is injected automatically in private sessions, while daily logs appear to be read instructionally via AGENTS.md rather than injected by the runtime. This distinction affects how reliably agents can access prior session state.

The authors note they could not verify this behavior from the source code, marking an unresolved implementation detail.