Think Before You Act: Decision Gate for LLM Agents

Iftach Shoham; Yuval David; Offer Akrabi; Erez Bilgory; Ido Levy; Sami Marreed; Sergey Zeltyn; Segev Shlomov

cs.AIOct 28, 2026

Think Before You Act: Decision Gate for LLM Agents

Authors: Iftach Shoham, Yuval David, Offer Akrabi, Erez Bilgory, Ido Levy, Sami Marreed, Sergey Zeltyn, Segev Shlomov

Organizations: IBM SIL

Abstract

LLM agents increasingly perform consequential actions through external tools, yet their proposed actions are often executed without independent verification of policy compliance or task completion. We introduce JevAct, a runtime verification framework that intercepts selected agent actions before execution and enables feedback-driven correction. JevAct compiles domain policies into executable state checks and structured questions evaluated by a prefill-only decision model, which produces probability-based judgments without relying on LLM generation. By combining deterministic verification of tool-state facts with probabilistic assessment of conversational context, the framework identifies potentially invalid actions and premature task termination, providing targeted feedback for bounded revision. JevAct operates independently of the underlying agent model and requires no modifications to its parameters or tools. We evaluate JevAct across eleven LLM agents on $τ^2$ -bench Airline and Telecom tasks and BFCL. Results show improvements in agent reliability across multiple models and benchmarks, with particularly consistent gains on Telecom and BFCL, while performance on Airline is more mixed. Overall, these findings suggest that combining structured decision models with executable verification can improve the reliability of tool-using LLM agents.

Figures & tables

Panel of Figure 1 — Figure 1: pass^3 results from Table 2 .

Panel of Figure 2 — Figure 2: JevAct. Offline (top), an LLM compiles the policy into checks. Online (bottom), the agent’s proposal goes through four stages: (1) propose, (2) trigger, (3) check when required, and (4) commit or revise. Proposals without a write call or hand-off are executed as in ReAct. A flagged proposal is discarded, and the LLM proposes again with temporary feedback.

Benchmark	Check	Compiled items
$\tau^{2}$ Airline	write gate	6 write tools, 86 questions (17)
$\tau^{2}$ Telecom	write gate	6 write tools, 75 questions (5)
	hand-off check	8 problem types, checklist steps (7 dropped)
BFCL	end-of-turn check	–
	requested-action gate	–

Table 1: Checks per benchmark, with the number of compiled items (dropped items in parentheses).

		pass@3		pass^3
Benchmark	Model	ReAct	JevAct	ReAct	JevAct
$\tau^{2}$ Airline	GPT-4.1	66.0	82.0	36.0	46.0
	Opus 4.6	84.0	92.0	72.0	76.0
	gpt-oss-20b	66.0	70.0	48.0	44.0
	gpt-oss-120b	68.0	66.0	42.0	50.0
	Qwen3.8-27B	90.0	90.0	74.0	76.0

Table 2: Main results (%). pass@3: tasks solved in at least one of three runs. pass^3: tasks solved in all three runs. Both arms use the same LLM and differ only in the harness; the higher value of each pair is in bold .

	pass^1	Flags per task
ReAct	50.0	–
JevAct (same configuration as Table 2 )	72.0	0.38
Code checks only	68.0	0.28
Kev-27B questions only	57.1	0.29
Kev-9B instead of Kev-27B	58.0	0.22
One revision instead of two	68.0	0.32

Table 3: Single-trial write-gate ablation on Airline with GPT-4.1 as the agent. The experiment uses the same configuration as Table 2 (code checks, Kev-27B, and

R=2

Appendix figures & tables3 assets

Supplementary material from the paper’s appendix.

Appendix

Step	Reads	Writes	Labels
1. Questions	policy; write-tool schemas	questions per write tool	none
2. Rule	policy; step-1 questions	allow rule per write tool	none
3. Code	policy, schemas; one read result per tool	derive_<tool> function per write tool	none
4. Repair	code and flags on logged calls	repaired code	none
5. Filter	Kev answers on logged write calls	questions to drop	none
6a. Prerequisites	policy; tool names (Telecom only)	prerequisites	none

Appendix

Table 6: The offline compilation workflow. “Logged” calls and hand-offs come from development runs. Telecom step 6 also includes a separate prerequisite-extraction call.

Hand-offs flagged	371	369	342
	Kev-27B	Kev-9B	Kev-0.8B
Same flag/pass decision as Kev-27B	–	97.0%	90.6%
Required fixes caught (of 533)	319	316	299
Unneeded fixes flagged (of 245)	193	195	170
Failed episodes incorrectly marked fixed (of 380)	0	0	11

Appendix

Table 7: Offline replay of the same 609 telecom hand-off states for each Kev size.

		ReAct		JevAct
		Model		Model		Kev	Drafts	Not
Benchmark	Model	in	out	in	out	in	sent back	rebuilt
$\tau^{2}$ -bench airline 50 tasks $\times$ 3	GPT-4.1	58.4	0.9	60.1	0.9	9.0	0.39	0
	Opus 4.6	64.6	1.5	64.1	1.5	6.7	0.08	0
	gpt-oss-20b	50.4	3.0	56.8	3.4	5.2	0.19	0
	gpt-oss-120b	44.1	2.5	43.8	2.6	6.5	0.36	0

Appendix

Table 9: Mean tokens per run, in thousands, for the runs in Table 2 . Agent usage is logged; Kev input is logged on

\tau^{2}

-bench and rebuilt on BFCL. Drafts sent back and checks not rebuilt are counts per run.

LLM Agent Verification

LLM agent verification checks whether agents’ actions, evidence, policies, and completion claims satisfy explicit requirements.

See all 148 papers

Verified Tool Calls Improve LLM Agent Reliability Under Non-Atomic Failures
Jul 31, 2026Isham Kalappurackal Mansoor, Abhishek Phadke, Pratip RanaAI Agent Reliability LLM Tool Use
Department of Computer Science Old Dominion University Norfolk, VA 23529 · School of Engineering and Computing Christopher Newport University Newport News, VA 23606
Look Before You Leap: Pre-Action Verification for LLM Agents
Sep 14, 2026Asaad AlthoubiLLM Agent Verification AI Agent Benchmarks
Oklahoma State University Stillwater, OK USA
Evaluating Tool-Using Language Agents: Judge Reliability, Propagation Cascades, and Runtime Mitigation in AgentProp-Bench
Apr 17, 2026Bhaskar GurramAI Agent Reliability Tool-Augmented Language Model Agents
Zasti Inc., Ashburn, VA, USA

Jul 31, 2026J/K move · Enter open · S save

Verified Tool Calls Improve LLM Agent Reliability Under Non-Atomic Failures

Isham Kalappurackal Mansoor, Abhishek Phadke, Pratip Rana

Department of Computer Science Old Dominion University Norfolk, VA 23529 · School of Engineering and Computing Christopher Newport University Newport News, VA 23606

AI Agent Reliability LLM Tool Use Tool Use LLM Agent Reliability

Open paper AskPDF arXiv

Large Language Model (LLM) agents rely on external tools to perform multistage tasks. Existing agent frameworks typically assume that tool calls are atomic and return binary success or failure signals. However, real-world systems exhibit non-atomic behaviors such as timeouts after dispatch, delayed visibility, and partial state updates. These mismatches lead to reliability issues including duplicate actions, task success, and unnecessary tool executions. A lightweight, verification-aware tool wrapper is introduced that augments tool calls with postcondition verification, verify-before-retry logic, and idempotency keys. The approach is evaluated in a controlled simulated environment with injected non-atomic failures across multiple task templates. The results demonstrate that the proposed method significantly reduces duplicate actions, while maintaining comparable task success rates. Overall, the findings suggest that strengthening tool interaction semantics is a promising direction for improving LLM agent reliability without requiring modifications to the underlying language model.

Abstract

LLM Agent Verification

Explore similar work

Verified Tool Calls Improve LLM Agent Reliability Under Non-Atomic Failures

Look Before You Leap: Pre-Action Verification for LLM Agents

Evaluating Tool-Using Language Agents: Judge Reliability, Propagation Cascades, and Runtime Mitigation in AgentProp-Bench