--- title: "For me, an agent consists of four files — isla Studio" url: "https://isla-stud.io/en/ki-b2b/agent-besteht-aus-vier-dateien/" date: 2026-07-24 --- # For me, an agent consists of four files **What you’ll take away from this:** You’ll learn about the four core files I use to clearly separate an agent’s identity, mode of operation, limits, and memory. „SOUL“ sounds more elaborate than it actually is. In my Agents Brain, the word refers to a short persona file. Alongside it are MASTER, RULES, and MEMORY. Four files aren’t enough for every conceivable capability. But they’re enough to describe an agent in a way that allows another tool to load it, a human to review it, and the agent itself to maintain its knowledge. The separation is the real point. A single mega-prompt quickly mixes together tone, operating rules, project details, safety limits, and past decisions. Then every correction becomes risky: whoever changes a sentence doesn’t know which other areas they’re affecting in the process. That’s why I divide up the responsibility. SOUL: Who This Agent Is -------------------------- The SOUL file is the shortest of the four. It describes the agent’s role, attitude, and voice. It also contains the mandatory reading order for the start of the session. For Sol, for example, it states that, as an author and press spokesperson, she embodies Saskia’s public voice: professionally precise, clear, human, and without hype. Equally important are her boundaries: she does not make up numbers and does not publish anything herself. The SOUL must remain brief enough to be read in its entirety at every startup. It is not an encyclopedia. Its purpose is to provide guidance. A good short persona answers three questions: - What role does the agent embody? - How does he make linguistic and technical decisions? - What underlying files must they load before starting work? This ensures that SOUL serves as a stable starting point, not a catch-all repository. MASTER: How the Agent Works ------------------------------ The MASTER file is the complete operational specification. Here, details can be extensive. Depending on the role, it contains tone, output formats, subject-matter standards, source hierarchy, workflow, quality checks, and error handling. The rules for an author differ from those for a development agent. The structure remains similar, but the content is role-specific. The advantage of this separation becomes apparent when changes are made. If I refine the visual language for all future articles, I update the work instructions. The author’s core identity does not need to be rewritten for this. Conversely, a role can further develop its tone without blurring its technical delivery formats. The MASTER file thus serves as both a manual and a contract: detailed enough for difficult cases, yet separate from the brief starter persona. RULES: What the Agent Can and May Do ---------------------------------- RULES is the binding work contract for capabilities, limitations, and approvals. The file specifies: - Role and domain, - Capabilities, - Expected inputs, - Typical outputs, - Dependencies, - Memory scope, - Prohibited actions, - Approval gates. The last two points, in particular, are often addressed too late in agent setups. A capability alone does not indicate whether an agent is permitted to execute it independently. [Planning Roles, Authorizations, and Operations for AI in the Enterprise.](https://isla-stud.io/beratung/#ki-automatisierung) For example, Sol can prepare WordPress drafts. It is not allowed to publish them. This restriction isn’t just a friendly reminder at the end of a prompt—it’s part of its RULES file. This allows an orchestrator to verify, before delegation, whether the role is suitable for the task. And the next tool loads the same approval rules. MEMORY: What Remains Permanently Relevant ------------------------------------- The MEMORY file is not a log of all conversations. It contains the curated essence of daily logs and lessons. It lists ongoing projects, proven conventions, known pitfalls, and important stakeholder rules. Chronological details remain in the daily logs. Recurring correction patterns are maintained as separate lessons. This separation protects the initial phase from information overload. An agent doesn’t need to read a hundred old sessions to work effectively today. They load a compact, long-term file and, if necessary, supplement it with current daily logs or relevant lessons. Curation here means condensing and visibly updating the content. Daily logs are retained as episodic records. If a standalone „Memory“ or „Lesson“ file becomes technically obsolete, it does not disappear either: it is marked as superseded with the date, reason, and a reference to its successor. “Memory” is an editorial task, not a dumping ground. Why four files are better than one long prompt ---------------------------------------------------- The four files do not form a hierarchy. They answer different questions: FileKey QuestionType of ChangeSOULWho am I?rare, identity-relatedMASTERHow do I work?regular, operationalRULESWhat can and am I allowed to do?for skills and gatesMEMORYWhat do I need to know in the long term?ongoing, curatedThis division reduces coupling. A new capability changes the RULES file. An editorial rule belongs in the MASTER file. A correction based on practical experience becomes a Lesson and, later if necessary, a Memory. The persona does not need to be modified with every adjustment. This does not automatically make an agent good. It makes it maintainable. Approval rules are part of the identity ------------------------------------ To me, approval gates are not a downstream security filter. They define how an agent performs its role. An author who only submits drafts works differently than an autonomous publishing agent. A development agent with write access to the repository needs different boundaries than an agent with production access. If these differences are configured only within the respective tool, they disappear when switching tools. If, on the other hand, they are defined in RULES and MASTER, they travel with the role. This is especially important when it comes to delegation. When Nox hands off a task to Sol or another harness, the handoff does not remove a gate. The task can change hands. The chain of responsibility remains. A real-world example: Sol -------------------------- Sol is a good workshop example because it uses exactly this file architecture. Her SOUL file sets the brief identity and requires the full sources to be loaded before each task. The MASTER file contains her writing rules, verifiability, delivery structures, and image management. The RULES file describes capabilities and the “review-before-post” gate. The MEMORY file records ongoing editorial contexts and known pitfalls. When Saskia corrects a headline or a sourcing practice, the insight isn’t just incorporated into the current text. Sol receives a lesson that is reviewed again during the next relevant task. This is how feedback becomes systemic behavior. What this structure deliberately does not claim ------------------------------------------ Four files do not turn a language model into a human. Nor do they guarantee perfect continuity. An agent may overlook sources, apply a rule incorrectly, or document a session incompletely. The structure ensures that such errors are given a visible place: as a correction in the manual, as a refined gate, or as a new lesson. For me, this is the practical difference between “The prompt was good this time” and “The system can learn permanently from a correction.” The next section is therefore not about a fifth core file, but about the flow between the files: working memory, daily logs, curated long-term knowledge, and lessons. ### Sources - **Project source:** agents-brain specification, memory maintenance convention, and Sol’s canonical agent files, as of July 29, 2026 (private primary sources) - [CommonMark: open Markdown specification](https://spec.commonmark.org/0.31.2/) **Agents-Brain Series** [← Part 2: My Agents’ Brain Is a Git Repo](https://isla-stud.io/ki-b2b/gehirn-meiner-agenten-git-repo/) [Part 4: For my agents, remembering is a writing routine →](https://isla-stud.io/ki-b2b/agenten-erinnern-schreibroutine/)