A method for managing AI agents
Operchief
Project Method
Two Operchief postulates“AI systems have crossed the threshold of human cognitive capability.”“An LLM-based AI agent is a highly erudite probabilistic system in which context loss and plausible fabrication are intrinsic operating modes, not malfunctions.”Operchief Project Method follows from these two postulates. It is a way to work with an AI agent so the project is led by a human, not by the flow of messages in a chat.
Operchief Project Method (OPM) is a method for non-professional builders working with AI agents in personal and hobby projects of any kind. Professionals use similar disciplines, but they are usually tuned to a specific kind of work, a specific agent, or a specific environment. OPM is simpler and more general: it works with any modern AI and in any working environment, from a browser chat to Cursor, Codex, Claude Code, or Antigravity.
At its core, OPM is a set of strict rules for the agent: do not change project files without authorization, plan the work in detail, and record results in logs. Its purpose is to turn work with an agent into a manageable project system: simple, inexpensive, and usable by non-professionals.
In that sense, OPM runs against the mainstream of vibe coding. It is not "build me a good project from one prompt." It is careful work by a human and an agent: splitting work into stages, keeping control at every stage, and making the result dependable.
OPM is not only for coding. It is for any AI-assisted work that requires iteration: writing an article, building an Arduino prototype, translating a book, or carrying a personal project through several rounds of thought and revision.
It also makes AI work cheaper. Long chats force models to keep rereading old context; people avoid starting a new chat because they are afraid of losing the thread, and thousands of tokens are spent on old conversation history. Step away from the computer and the server-side cache may expire, so the model reads the same history again at your expense. OPM moves context into project files, making it easier to start fresh sessions, bring in several models at once, and separate planning from execution.
Four supports of OPM
- I
Authority stays with the human
The human is the only source of decisions in the project. In OPM, this role is called the Operchief: the project's chief operating officer (COO). The agent analyzes, plans, proposes, and executes what the Operchief approves. The agent has no authority of its own.
- II
PIN separates words from action
The agent works in two modes, and the boundary between them is absolute. The guard at that boundary is the PIN: a numeric code explicitly typed by the Operchief. Discussion and analysis mode is the default: reading, analysis, planning, risk review, questions, and clarifications are allowed, but changing any file is forbidden. Execution mode begins only after the PIN and only inside the approved plan.
- III
Logs are the project's external memory
Decisions, plans, and work history are stored in project documents and logs. A new chat restores context from those files, so the human does not need to explain the project rules, the purpose of the project, or the current task to every new agent from the beginning.
- IV
StepLoop turns control into work
Every action that changes project files goes through the cycle: task statement, analysis of task-related files, plan, plan entry in the log, execution, report, review, return to planning if the review fails, or result logging and archiving if the result is accepted. The cycle is not shortened.
A human is present at every stage, but not always in the same role: sometimes as the subject-matter expert, sometimes as the router between agents, and sometimes simply as the keeper of the protocol.
Even when the Operchief cannot personally judge the code, they can bring in another agent for review, rely on a well-specified task, and keep the executing agent working under the assumption that its output will be checked. That assumption alone helps prevent errors and strange behavior.
The StepLoop cycle for solving any task with an AI agent
A StepLoop is the basic unit of any work that changes project files. It is not a "planning unit" or a generic "task"; it is a fixed order the agent must follow for every change. Shortening the cycle because a task looks small is exactly where the agent stops being an executor.
A PIN stands between the agent's words and its actions. In its simplest form, StepLoop requires three PINs.
The first authorizes writing the plan into the log. That may seem excessive, but once a plan is approved and written down, the Operchief can open a new chat and drop the old context, bring in a consulting agent, or hand the recorded plan to a simpler, cheaper execution agent.
The second PIN authorizes execution: the Operchief has completed the necessary consultation and is ready for the agent to carry out the plan. The third PIN closes the StepLoop: the Operchief has verified that the work was done correctly and closes the task.
If the result is not accepted, the loop returns to the plan and starts again.
Templates and documentation
Templates are the operational form of the method. They are plain text, Markdown, with no build step and no dependencies. The bootstrap prompt deploys the core into a clean project: it asks for the briefing, generates Node IDs, and puts the files in place.
Download OPM_template_en.zip| № | File | Template | Role |
|---|---|---|---|
| 01 | AGENTS_TEMPLATE_en.md | Main Protocol | The project constitution: authority, operating modes, StepLoop, PIN, and conflict priorities |
| 02 | AGENT_BRIEF_TEMPLATE_en.md | Agent Brief | Fast agent onboarding: document navigation, working rules, and project terms |
| 03 | PROJECT_BRIEF_TEMPLATE_en.md | Project Brief | Project meaning: audience, MVP, metrics, acceptance, constraints, and risks |
| 04 | AGENT_LOG_TEMPLATE_en.md | Working Log | TODO, open StepLoops, commits, and the latest closed StepLoop headers |
| 05 | AGENT_LOG_ARCHIVE_TEMPLATE_en.md | Archive Log | Project history: closed first-level StepLoops and completed TODO items |
| 06 | BRIEFING_SOURCE_TEMPLATE_en.md | Bootstrap Questionnaire | Structured answers that the bootstrap prompt uses to create the project core |
| 07 | PROMPT_0_IDE_en.md | Bootstrap Prompt (IDE) | Instructions for an agent with direct file access: Cursor, Codex, Claude Code, Antigravity |
| 08 | PROMPT_0_CHAT_en.md | Bootstrap Prompt (Chat) | Instructions for an agent without file access; the Operchief applies changes manually |
| 09 | HUMAN_NOTE_en.md | Human Note | A plain-language note for the Operchief on the meaning and use of the method |
| 10 | README_en.md | Archive Guide | Archive contents and deployment order for the EN template archive |
The OPM template is distributed as a single English ZIP archive. Deployment is handled by the bootstrap prompt: the IDE version is for an agent with file access, and the chat version is for work through conversation. Individual template files are not published as separate downloads.