SoK: Agentic Skills -- Beyond Tool Use in LLM Agents
SoK: Агентные навыки — за пределами использования инструментов в LLM-агентах
2026-02-24
SCID: 54.1/4bxtzjc9
Discuss with AI
LLM agentsagent securityagentic skillsprocedural knowledgeskill libraries
Figures from the paper
Abstract (AI)
Agentic systems increasingly rely on reusable procedural capabilities, \textit{a.k.a., agentic skills}, to execute long-horizon workflows reliably. These capabilities are callable modules that package procedural knowledge with explicit applicability conditions, execution policies, termination criteria, and reusable interfaces. Unlike one-off plans or atomic tool calls, skills operate (and often do well) across tasks. This paper maps the skill layer across the full lifecycle (discovery, practice, distillation, storage, composition, evaluation, and update) and introduces two complementary taxonomies. The first is a system-level set of \textbf{seven design patterns} capturing how skills are packaged and executed in practice, from metadata-driven progressive disclosure and executable code skills to self-evolving libraries and marketplace distribution. The second is an orthogonal \textbf{representation $\times$ scope} taxonomy describing what skills \emph{are} (natural language, code, policy, hybrid) and what environments they operate over (web, OS, software engineering, robotics). We analyze the security and governance implications of skill-based agents, covering supply-chain risks, prompt injection via skill payloads, and trust-tiered execution, grounded by a case study of the ClawHavoc campaign in which nearly 1{,}200 malicious skills infiltrated a major agent marketplace, exfiltrating API keys, cryptocurrency wallets, and browser credentials at scale. We further survey deterministic evaluation approaches, anchored by recent benchmark evidence that curated skills can substantially improve agent success rates while self-generated skills may degrade them. We conclude with open challenges toward robust, verifiable, and certifiable skills for real-world autonomous agents.
Key Findings
1
Agentic skills are reusable procedural modules that extend beyond one-off plans or atomic tools by encoding applicability conditions, execution policies, termination criteria, and interfaces.
2
An orthogonal representation-by-scope taxonomy classifies skills by form—natural language, code, policy, or hybrid—and operating environment, including web, operating systems, software engineering, and robotics.
3
Benchmark evidence indicates that curated skills can substantially improve agent success rates, whereas self-generated skills may reduce performance, motivating deterministic evaluation and stronger verification mechanisms.
4
Skill-based agents create major security and governance risks, illustrated by the ClawHavoc campaign, where nearly 1,200 malicious marketplace skills exfiltrated API keys, cryptocurrency wallets, and browser credentials.
5
The paper maps the complete agentic-skill lifecycle and introduces seven system-level design patterns spanning packaging, execution, self-evolution, and marketplace distribution.
Research Object
agentic skills in LLM-based autonomous agents
Research Subject
the lifecycle, representations, execution patterns, evaluation, security, and governance of agentic skills
Publication Details
Publication Date
2026-02-24
Journal
Publisher
ISSN
Cited by
157
Open access PDF
Access Type
Author Information
Download PDF
Subscribe to digest
References available in scid.ai4
A survey on large language model based autonomous agents2024
Generative Agents: Interactive Simulacra of Human Behavior2023
Not What You've Signed Up For: Compromising Real-World LLM-Integrated Applications with Indirect Prompt Injection2023
AgentTuning: Enabling Generalized Agent Abilities for LLMs2024