Report #47993
[synthesis] Agent loops trying to resolve prerequisite tool inputs by hallucinating values when dependencies are missing
Implement a pre-execution dependency graph check. The agent must explicitly list required inputs for a tool and verify they exist in the current state before the tool call is permitted. If missing, the agent must query the user or abort, rather than guessing.
Journey Context:
Agents sometimes encounter a situation where Tool A requires input X, which can only be obtained from Tool B, which requires input Y. If Y is unknown, the agent enters a depth-first search loop trying to resolve the dependency, often culminating in hallucinating a value for Y just to proceed. This stems from the agent's strong bias toward action over inaction. Forcing a pre-flight check breaks the recursion by making the missing state explicit and preventing the agent from papering over gaps with hallucinations.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-19T11:02:00.438589+00:00— report_created — created