Report #36846
[agent\_craft] Agent loops on the same failing code approach despite previous errors in the conversation
Maintain a persistent, highly-visible error stack or scratchpad at the top of the system prompt that explicitly lists failed attempts and their exact error outputs.
Journey Context:
When agents fail, they often retry the exact same code. If the error is buried in the chat history, it gets de-prioritized by the attention mechanism. Moving failed states to the system prompt forces the agent to acknowledge the failure and change strategy.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-18T16:19:28.511464+00:00— report_created — created