HASP: Harnessing LLM Agents with Executable Skill Programs

ArticleHongjun Liu, Yifei Ming, Shafiq Joty, Chen ZhaoResearch

A research team introduces HASP (Harnessing LLM Agents with Skill Programs), a framework that converts past agent experiences into executable Program Functions (PFs). Instead of relying on passive textual prompts, PFs act as executable guardrails that trigger on failure-prone states to alter the next action or supply corrective context.

Key points

  1. Traditional agent skill libraries rely on passive textual advice that lacks reliable intervention mechanisms during execution.
  2. HASP upgrades learned skills into Program Functions (PFs) serving as executable guardrails in the agent loop.
  3. The framework supports inference-time intervention, structured post-training supervision, and self-improvement via reviewed PFs.
  4. In web-search tasks, inference-time PFs alone increased performance by 25% over a ReAct agent.
  5. Post-training combined with controlled PF evolution outperformed Search-R1 by 30.4% on search reasoning.
  6. Empirical gains were demonstrated across web search, mathematical reasoning, and coding benchmarks.

Why it matters

LLM agents often fail long-horizon tasks because natural language instructions are not strictly adhered to. Encapsulating reusable lessons as executable code provides deterministic safeguards directly inside the control loop, significantly improving task completion reliability.

Analysis

The authors (Liu, Ming, Joty, and Zhao) tackle a major limitation in agent architectures: prompt-based advice lacks enforceable execution guarantees. By converting operational lessons into program functions, HASP bridges statistical model output and deterministic software guardrails. The accompanying mechanism analysis sheds light on when functions should trigger, how agents internalise these capabilities during training, and what safeguards are required to evolve skill libraries without destabilising the agent.

Mentioned links

Summary generated by AI (Gemini 3.8 Flash, 27 September 2026). It may contain errors – the original source is authoritative.

Related entries

Similar in content, found by the AI search.

  • Article:Shuai Shao, Kangning Zhang, Qingyao Li, Shijian Wang, Hao Wang, Wenxiang Jiao, Yuan Lu, Yi Guo, Weiwen Liu, Weinan Zhang

    Harness-R1: Learning to Edit Executable Runtime Harnesses from Agent Failure Trajectories

    Researchers introduce Harness-R1, a framework that trains a dedicated agent to adapt and patch the runtime harness of LLM agents based on failure logs. Using online reinforcement learning, a separate 9B model learns to generate executable patches for context handling, tool mediation, and error recovery to directly improve task completion.

    AI & LLMs· Research

  • Article:Liyan Tang, Cyrus Rashtchian, Chun-Sung Ferng, Andrew Tomkins, Da-Cheng Juan, Tu Vu

    WikiSkill: Compiling Agent Experience into Persistent Knowledge for Skill Evolution

    This research paper introduces WikiSkill, a framework that co-evolves AI agent skills alongside a persistent knowledge base (wiki). By disentangling raw execution experience, accumulated knowledge, and executable skills, WikiSkill enables systematic reuse of insights across iterations.

    AI & LLMs· Research

  • Article:Siru Ouyang, Jun Yan, I-Hung Hsu, et al.

    ReasoningBank: Scaling Agent Self-Evolution with Reasoning Memory

    ReasoningBank is a memory framework for LLM agents that distils generalisable reasoning strategies from self-judged successful and failed experiences, enabling continuous self-evolution over time.

    AI & LLMs· Research

  • Link:Anna Gutowska

    What Are AI Agents? Fundamentals, Architecture, and Workflows by IBM

    An IBM guide explains AI agents as autonomous systems that leverage LLMs to plan workflows, break down tasks, and execute tool calls. Unlike traditional language models limited to static training data, agentic systems dynamically retrieve data via APIs, external tools, or other agents, adapting through iterative reflection and feedback loops.

    AI & LLMs· Resource

Back to the knowledge base

Let's talk about your project

Get in touch

Locations

  • Mattersburg
    Johann Nepomuk Bergerstraße 7/2/14
    7210 Mattersburg, Austria
  • Vienna
    Ungargasse 64-66/3/404
    1030 Wien, Austria

Parts of this content were created with the assistance of AI.