Paper tests resume semantics in agent workflow frameworks
The RESUME CONTRACT uses TLA+ and a deterministic harness to check persistence behavior.
Why it matters
Persistent agent workflows increasingly mediate external effects, so inconsistent resume behavior can lead to duplicated work, invalid state or divergent recovery after crashes. The paper gives framework authors and users a concrete contract for testing these failure modes.
The key points
- 1.RESUME CONTRACT defines six persistence properties for workflow resume semantics.
- 2.The TLA+ reference model was checked across 7.4 million states.
- 3.The harness found differing resume behavior across five agent workflow frameworks.
A new arXiv paper proposes RESUME CONTRACT, a machine-checkable conformance contract for checkpoint, interrupt and resume semantics in workflow persistence layers. The paper defines six persistence properties plus fork-intent and liveness obligations, checks a reference semantics with a TLA+ model across 7.4 million states, and uses a deterministic LLM-free harness on pinned framework releases. It reports that five widely deployed agent workflow frameworks handle resume behavior differently and do not expose machine-checkable contracts.
⚡ Try this today
Audit any agent workflow system that persists state for exactly-once effects, crash recovery and checkpoint validity before using it for effect-bearing tasks.
Sources & original reporting
This brief summarizes and links to reporting from the publishers below.
- arXiv cs.LGResume Means Resume: A Machine-Checked Conformance Contract for Checkpoint, Interrupt, and Resume Semantics in Workflow Persistence LayersAug 5, 12:00 PM↗
- HF Daily PapersResume Means Resume: A Machine-Checked Conformance Contract for Checkpoint, Interrupt, and Resume Semantics in Workflow Persistence LayersAug 4, 4:00 AM↗
Enjoyed this brief? Get the next one in your inbox.
More in Research
GLM-5.3 Artificial Analysis Benchmarks
87 points, 39 comments on Hacker News.
Prior Audit-Repair Context Shifts LLM Verifier Thresholds Toward Leniency
arXiv:2608.
Improving the matrix multiplication exponent with modern optimization and AlphaEvolve
arXiv:2608.