<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="zhang-etal-2026-stars">
    <titleInfo>
        <title>STARS: Skill Triage via Activation Risk Scoring for LLM Agents</title>
    </titleInfo>
    <name type="personal">
        <namePart type="given">Guijia</namePart>
        <namePart type="family">Zhang</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Shu</namePart>
        <namePart type="family">Yang</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Xilin</namePart>
        <namePart type="family">Gong</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Di</namePart>
        <namePart type="family">Wang</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <originInfo>
        <dateIssued>2026-10</dateIssued>
    </originInfo>
    <typeOfResource>text</typeOfResource>
    <relatedItem type="host">
        <titleInfo>
            <title>Findings of the Association for Computational Linguistics: EMNLP 2026</title>
        </titleInfo>
        <name type="personal">
            <namePart type="given">Hung-Yi</namePart>
            <namePart type="family">Lee</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Sunipa</namePart>
            <namePart type="family">Dev</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Jessy</namePart>
            <namePart type="family">Li</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Desmond</namePart>
            <namePart type="family">Elliott</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <originInfo>
            <publisher>Association for Computational Linguistics</publisher>
            <place>
                <placeTerm type="text">Budapest, Hungary</placeTerm>
            </place>
        </originInfo>
        <genre authority="marcgt">conference publication</genre>
        <identifier type="isbn">979-8-89176-462-0</identifier>
    </relatedItem>
    <abstract>The risk of an agent skill call depends not on the skill alone but on the invocation context: the same file-reading skill is benign when listing logs, yet dangerous when tainted content steers the agent toward secret exfiltration. We formalize invocation-time auditing: intercept a proposed skill call before execution and score its risk from the user request, the candidate skill, and the runtime context. STARS implements this as a three-stage pipeline that fuses a static capability prior with a request-conditioned invocation scorer. The primary configuration, STARS-LLM, uses an LLM judge; STARS-Rules is a deterministic, API-free alternative. On SIA-Bench, a 3,000-record synthetic benchmark with a 180-record human check at Fleiss \ensuremathąppa = 0.709, request-conditioned scoring improves high-risk retrieval on held-out indirect prompt injection, where static metadata cannot see the attack. On a frozen 500/100 scoring subset, STARS-LLM raises rules-only HR-AUPRC from 0.297 to 0.728 in-distribution and from 0.504 to 0.757 on held-out IPI. That gain is in ranking: under frozen thresholds, held-out classification accuracy does not improve. SIA-Bench is a ranking stress test, not an estimate of real-world prevalence.</abstract>
    <identifier type="citekey">zhang-etal-2026-stars</identifier>
    <identifier type="doi">10.18653/2026.findings-emnlp.891</identifier>
    <location>
        <url>https://aclanthology.org/2026.findings-emnlp.891/</url>
    </location>
    <part>
        <date>2026-10</date>
        <extent unit="page">
            <start>18874</start>
            <end>18890</end>
        </extent>
    </part>
</mods>
</modsCollection>

%0 Conference Proceedings
%T STARS: Skill Triage via Activation Risk Scoring for LLM Agents
%A Zhang, Guijia
%A Yang, Shu
%A Gong, Xilin
%A Wang, Di
%Y Lee, Hung-Yi
%Y Dev, Sunipa
%Y Li, Jessy
%Y Elliott, Desmond
%S Findings of the Association for Computational Linguistics: EMNLP 2026
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-462-0
%F zhang-etal-2026-stars
%X The risk of an agent skill call depends not on the skill alone but on the invocation context: the same file-reading skill is benign when listing logs, yet dangerous when tainted content steers the agent toward secret exfiltration. We formalize invocation-time auditing: intercept a proposed skill call before execution and score its risk from the user request, the candidate skill, and the runtime context. STARS implements this as a three-stage pipeline that fuses a static capability prior with a request-conditioned invocation scorer. The primary configuration, STARS-LLM, uses an LLM judge; STARS-Rules is a deterministic, API-free alternative. On SIA-Bench, a 3,000-record synthetic benchmark with a 180-record human check at Fleiss \ensuremathąppa = 0.709, request-conditioned scoring improves high-risk retrieval on held-out indirect prompt injection, where static metadata cannot see the attack. On a frozen 500/100 scoring subset, STARS-LLM raises rules-only HR-AUPRC from 0.297 to 0.728 in-distribution and from 0.504 to 0.757 on held-out IPI. That gain is in ranking: under frozen thresholds, held-out classification accuracy does not improve. SIA-Bench is a ranking stress test, not an estimate of real-world prevalence.
%R 10.18653/2026.findings-emnlp.891
%U https://aclanthology.org/2026.findings-emnlp.891/
%U https://doi.org/10.18653/2026.findings-emnlp.891
%P 18874-18890

Markdown (Informal)

[STARS: Skill Triage via Activation Risk Scoring for LLM Agents](https://aclanthology.org/2026.findings-emnlp.891/) (Zhang et al., Findings 2026)

• STARS: Skill Triage via Activation Risk Scoring for LLM Agents (Zhang et al., Findings 2026)

ACL

• Guijia Zhang, Shu Yang, Xilin Gong, and Di Wang. 2026. STARS: Skill Triage via Activation Risk Scoring for LLM Agents. In Findings of the Association for Computational Linguistics: EMNLP 2026, pages 18874–18890, Budapest, Hungary. Association for Computational Linguistics.