Reading
A privilege-level instruction encoding system based on the Routine framework, which significantly improves the execution accuracy of multi-step tasks through structured execution contracts and runtime verification mechanisms.
Section 01
A privilege-level instruction encoding system based on the Routine framework, which significantly improves the execution accuracy of multi-step tasks through structured execution contracts and runtime verification mechanisms.
Section 02
Section 03
Current large language models (LLMs) often have unsatisfactory accuracy when executing multi-step tasks. Studies show that the execution accuracy of complex task chains can be as low as around 41%. The root cause of this problem is that LLMs tend to deviate from the original goal or lose key constraints during execution when handling long contexts and complex instructions.
Traditional prompt engineering methods can improve output quality to some extent, but lack a systematic execution guarantee mechanism. When tasks involve multiple steps, tool calls, or need to strictly follow specific rules, the model may call tools at inappropriate times or ignore important safety constraints.
Section 04
Imperative Workflow Engine is an open-source project implemented based on the Routine framework (arXiv:2507.14447), aiming to solve the above problems through Privilege-Level Instruction Encoding and structured execution contracts. This project increases the execution accuracy of multi-step tasks from about 41% to about 96%, an improvement of more than twice.
The core idea of the project is to draw on the concept of privilege levels in operating systems, assigning different priorities and constraint strengths to different types of instructions. By explicitly marking the privilege level of instructions, the system can enforce these constraints at runtime to ensure that high-priority rules are not overridden by low-priority requests.
Section 05
The project defines five privilege levels, each corresponding to different types of instructions:
| Level | Label | Purpose |
|---|---|---|
| 0 | Safety rules | Safety rules, absolutely non-deviable |
| 1 | System imperatives | Core system instructions |
| 2 | Skill-loaded rules | Default rules when skills are loaded |
| 3 | User requests | Default level for user requests |
| 4 | Tool outputs | Tool output results |
This layered design ensures that the most critical safety constraints (Level 0) have the highest priority and will not be overridden under any circumstances. User requests (Level 3) can be executed flexibly without violating higher-level constraints.
Section 06
The routine_decomposer.py script in the project is responsible for decomposing complex tasks into Routine-style execution contracts. These contracts clearly define the input, output, preconditions, and postconditions of each step, forming a formal execution protocol.
The advantages of execution contracts are:
Section 07
contract_verifier.py provides runtime verification functionality to ensure that the execution process strictly follows predefined contracts. This verification is not a simple result check, but continuous monitoring during execution to ensure each step meets expectations.
Section 08
tool_filter.py implements semantic tool filtering based on context relevance. This means the system does not blindly expose all available tools to the model, but intelligently selects the most relevant subset of tools based on the context of the current task. This filtering mechanism reduces the probability of the model selecting the wrong tool and also reduces noise in the context window.
Keep going with more reads from the same topic.
Nornir MCP Server is an enterprise-level server based on the Model Context Protocol (MCP). It seamlessly integrates large language models (such as Claude) with the Nornir network automation framework, supporting natural language orchestration for multi-vendor network devices (Cisco, Arista, Juniper, etc.), and providing production-grade features like a dual-engine architecture (NAPALM + Netmiko), intelligent filtering, and a secure sandbox.
Bibliothèque Française LLM is a structured indexing and annotation project for French public domain literature designed specifically for large language models (LLMs). It integrates multiple authoritative sources such as DraCor, Common Corpus, and Wikisource, providing metadata indexing categorized by genre, author, and era, as well as in-depth annotations for dramatic texts (including characters, lines, stage directions, etc.). Its aim is to enable LLMs to efficiently read and understand classic French literary works.
Splinter is a minimalist, high-performance key-value (KV) and vector storage system enabling zero-latency inter-process communication via shared memory and atomic operations. With only 766 lines of core code, it supports millions of operations per second and 768-dimensional vector storage, offering a new architectural approach for local LLM inference and data-intensive applications.
Folkering OS is the world's first AI-native bare-metal operating system, entirely written in Rust no_std without relying on Linux, POSIX, or libc. It can generate commands from scratch, compile them into WASM, and run them in 10 seconds, achieving true self-evolution.