Truncated Batch Tool Calls Corrupt Agent State; Paper Proposes Transaction Admission Control
Flunder707 · reddit · 2026-07-31
The author discovered that when a model hits its output limit mid-response, batched tool calls get truncated—causing some operations to execute while others are lost, leaving corrupted history.
Fuzz testing across 107 truncation positions confirmed that nearly all produce partial side effects. The proposed solution validates the entire batch before executing any call (transaction admission control), reducing partial execution to zero. Paper and code are open-sourced.
More from coding & agent
- 'LLM code is sloppy' rebutted with real ffmpeg code: identical if/else branches in a top library — kingslyj · 2026-10-16
- Office-editing tool calls need file, worktree and unit: lessons from dsh-univer-office — FunDistribution6537 · 2026-10-16
- Architect MD workflow: Gemini one-shot outputs that beat Opus, via 5-lens plan critique — oikk01 · 2026-10-16
- Prove a coding agent actually ran the tests: check artifacts, not approvals — ComprehensiveShake76 · 2026-10-16
- Building an acceptance harness for AI spreadsheet edits: allowlists, recalc checks, idempotency — ukdicha-modakkk · 2026-10-16
- Agent actions need checks outside the model: NVIDIA demos lean on guardrails amid wave of real agent incidents — Creamy-And-Crowded · 2026-10-16