Co-founder of Yempik. AI agents and automations in production, not in slides. Building cowork-os in public.

github

Raffaele Zarrelli  • Peerlist

<h3>About</h3>Most people use AI like a disposable chat: every session starts from zero and you become the memory. cowork-os gives your Claude Cowork project a memory you can read and correct. Decisions (each with a status), context, and open questions live in plain Markdown files the agent reads when a task starts and writes back when it ends. No database, no black box. Because the state lives on disk and stays inspectable, the agent runs real work while staying grounded and under your control.It is not a standalone application but rather a folder structure combined with prompts and conventions that can be integrated into a Claude Cowork project or any agentic-coding workspace. The setup is designed to be adaptable to specific business needs within approximately 30 minutes.<h3>Key Features</h3><ul><li>Living Memory: Unlike typical chat interfaces, decisions, assumptions, and open questions are stored in files, providing a persistent knowledge base.</li><li>Evidence-Based Operations: The system prompts the AI to verify information, distinguishing between facts, assumptions, and recommendations.</li><li>Outcome-Oriented Workflow: Focuses on achieving defined outcomes rather than just completing tasks.</li><li>Rapid Adoption: Can be set up in about 30 minutes by copying the folder structure and integrating the project instructions.</li><li>Guided Setup: An installer guides users through setup by pasting instructions into the chat and answering a few questions, resulting in a pre-configured workspace.</li><li>Inbox Processing: Unsorted materials can be placed in the _inbox/ folder for organization.</li><li>Pre-built Modules: Includes specialized modules like LinkedIn Growth OS, Missions, and Scheduled Tasks for various business functions.</li></ul><h3>Structure</h3>The project is organized into several key directories and files:<ul><li>INSTALL.md: For guided setup.</li><li>capabilities.md: Defines what Claude can set up.</li><li>PROJECT_INSTRUCTIONS.md: The core system prompt.</li><li>workflows/: Contains the Relentless Outcome Workflow.</li><li>context/: For overview, positioning, services, and tone of voice.</li><li>marketing/: For strategy, campaigns, and content.</li><li>website/: For notes, backlog, and homepage copy.</li><li>decisions/: For decision logs and open questions.</li><li>reviews/: Templates for weekly and maintenance reviews.</li><li>missions/: A template for outcome-driven missions.</li><li>linkedin/: The LinkedIn Growth OS module.</li><li>scheduled-tasks/: Recurring automations and setup guides.</li><li>_inbox/: For processing unsorted materials.</li><li>examples/yempik/: A complete, sanitized example workspace.</li></ul><h3>License</h3>This project is licensed under the MIT license.