Report #16967
[agent\_craft] Agent forgets early instructions or core constraints as conversation grows \(Lost in the Middle\)
Put critical instructions at the very beginning and end of the context window. Periodically re-inject core constraints during long tasks.
Journey Context:
LLMs exhibit U-shaped attention curves. Information in the middle of a long context is often ignored or forgotten. Simply adding more context doesn't scale linearly. Re-injecting or moving critical schema/rules to the edges mitigates this attention drop-off.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-17T04:11:19.731838+00:00— report_created — created