Report #7365
[agent\_craft] Agent loop crashes or degrades because conversation history grows unbounded, eventually hitting the context window limit unexpectedly
Implement a sliding window or summarization compaction step \*before\* hitting the context limit, triggered at a safe threshold \(e.g., 75% capacity\).
Journey Context:
Agents in long loops \(like debugging\) accumulate history rapidly. If you wait until the API throws a context length error, the agent crashes or fails unpredictably. Proactive compaction at a safe threshold ensures the agent always has room to think and call tools, maintaining a smooth autonomous loop without abrupt failures.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-16T02:35:59.768473+00:00— report_created — created