SkCC: A Portable and Secure Compiler for Cross-Framework LLM Agent Skills

Forum topic · 小凯 · 2026-07-03

Summary

This paper presents SkCC, a compiler designed for LLM agents that brings classical compilation design into agent skill development. LLM agents increasingly depend on reusable skills (e.g., SKILL markdown files) to execute complex tasks, but these artifacts lack portability because agent frameworks are highly sensitive to prompt formatting, causing the same skill to perform inconsistently across frameworks. Most skills are authored once as format-agnostic Markdown, requiring costly per-framework rewrites, while security vulnerabilities remain largely unaddressed. SkCC introduces SkIR, a strongly-typed intermediate representation that decouples skill semantics from framework-specific formatting, enabling portable deployment. A static optimizer enforces security constraints before deployment, blocking vulnerabilities proactively. Implemented as a four-phase pipeline, SkCC reduces adaptation complexity from O(m*n) to O(m+n) across m skills and n frameworks. Experiments on SkillsBench show pass rate increases from 21.1% to 33.3% on Claude Code and from 35.1% to 48.7% on Kimi CLI, with sub-10ms compilation latency, 94.8% proactive security trigger rate, and 10-46% runtime token savings.

Paper Overview

Research Area: Agent / Skill Compilation / Cross-Platform Authors: Yihao Wang, Yuheng Ji, Mingyu Cao et al. (Sun Yat-sen University team) Posted: 2026-05-05 arXiv: 2605.03353 Code: Open-source compiler code included with the paper

Abstract

LLM agents increasingly rely on reusable skills (e.g., SKILL markdown files) to execute complex tasks, yet these artifacts lack portability: agent frameworks are highly sensitive to prompt formatting, leading to large performance variation for the same skill across frameworks. Most skills are authored once as format-agnostic Markdown, necessitating costly per-framework rewrites, while security remains largely unaddressed, with widespread vulnerabilities observed in practice.

To address this, the authors propose SkCC, a compiler for LLM agents that introduces classical compilation design into agent skill development. SkCC centers on SkIR, a strongly-typed intermediate representation that decouples skill semantics from framework-specific formatting, enabling portable deployment across agent frameworks. Atop this IR, a static Optimizer enforces security constraints, blocking vulnerabilities before deployment.

Key Contributions

SkIR: A strongly-typed intermediate representation that separates skill semantics from framework-specific prompt formatting.
Static Optimizer: Performs security validation and blocks vulnerabilities before deployment.
Four-Phase Pipeline: Reduces adaptation complexity from **O(m*n) to O(m+n) across m skills and n frameworks.

Experimental Results (SkillsBench)

| Framework | Original Pass Rate | SkCC Pass Rate | Improvement | |---|---|---|---| | Claude Code | 21.1% | 33.3% | +12.2pp | | Kimi CLI | 35.1% | 48.7% | +13.5pp |

Additional performance metrics:

Compilation latency: <10 ms

Proactive security trigger rate: 94.8%

Runtime token savings**: 10%–46% across frameworks

--- *Auto-collected 2026-07-03*