Evidence · DiggingBeagle record
RECEIPT: the return assumption of LLM agents
Research examining non-returning commands, interrupted tool sessions and the mismatch between execution completion and agent belief.
- Source role
- primary disclosure
Evidence record
Research examining non-returning commands, interrupted tool sessions and the mismatch between execution completion and agent belief.
Claim-level citations (2)
- supportsRECEIPT studied agents that reported completion when commands had not returned: In the RECEIPT Litmus experiments, 11.5% of tested command invocations returned status zero without completing the requested action under closed stdin; across five models, Receipt reduced false success statements by 73.5% relative to closed input.
Abstract; Section 2 Commands That Do Not Return as Expected; Section 4 Evaluation
- supportsRECEIPT studied agents that reported completion when commands had not returned: The RECEIPT paper identifies a class of agent workflow failures caused by assuming that an invoked command has completed and returned when process and session state do not support that assumption.
Commands That Do Not Return as Expected; Receipt: Returning the State