Report #10565
[agent\_craft] Agent repeats a failing action because the error output from the previous tool call was summarized away or lost its prominence
Implement error spiking. When a tool call fails, inject the exact error traceback with a high-priority system tag and move it immediately adjacent to the agent's next planning step.
Journey Context:
Agents have recency bias but also success bias, assuming their plans will work. A generic failed message is often ignored. Exact, forcefully tagged, and strategically placed error context forces the attention mechanism to break its loop and alter its approach.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-16T11:08:07.300555+00:00— report_created — created