LLM-Wiki Template Boosts Collaborative Knowledge Work and Preserves Research Failures

A novel framework, the "llm-wiki-memory-template," has been introduced to enhance collaborative knowledge work involving large language model (LLM) agents. This template creates an LLM-maintained, interlinked wiki that acts as a persistent memory layer between raw information sources and the agents. Unlike traditional methods where LLM agents lack persistent memory across sessions or retrieval-augmented generation over raw sources fails to compound knowledge, this wiki pattern allows for continuous accumulation of findings, decisions, and reasoning. The template is designed to support heterogeneous collaboration across multiple humans, AI agents, and domains. A key innovation is its append-only convention, which ensures that all iterations, including dead ends and unsuccessful attempts, are preserved. This directly addresses the common problem in research and development where valuable "negative results" or abandoned approaches are lost, preventing future collaborators from repeating the same mistakes. Case studies demonstrate its utility in solo research, multi-author projects (where it helped audit and correct experimental claims), multi-agent deployments, and cross-domain educational applications, highlighting its ability to foster agent honesty and facilitate knowledge appropriation.