operations — the operating manual

Every company runs on an operating system; most just never install it on purpose. Decisions get made in whatever room happens to be occupied, meetings accrete because nobody wrote down which ones are allowed to exist, and the org chart is whoever answered the last email. This skill writes the manual down. It interviews the founder through how work actually happens — the cadence, the decision rights, the meetings and the refusals, where truth lives, and every seat the business needs whether or not anyone sits in it — then ships the result as a Business OS module: canonical OPERATIONS.md, machine contract, a decision log that’s meant to be written in weekly, MCP server, standalone site.

This is stone 7 of 8. Read ../PROTOCOL.md before composing — the module contract, workspace layout, and definition of done live there and are not repeated here. This stone reads vision (values and refusals shape decision rights) and business-model (the metrics the planning cycle steers by); locate both in the workspace and extract before asking. If either is missing, flag it loudly and mark the dependent sections provisional.

When to use it

The user says any of: “operating manual”, “operations doc”, “how we work”, “decision framework”, “meeting cadence”, “roles map”, “hiring plan”, “/operations” — or asks who should decide something, which meetings to have, or what to hire for next.

The rubric — sections of OPERATIONS.md

Compose in order; each section feeds the next. Propose → confirm → lock, one section per exchange (the interview mechanics in ../../potatuhs-design/SKILL.md §Procedure apply verbatim).

00 · Cover — business name, one-line description of how this company works, version, date.

01 · Planning cycle & cadence — the rhythm work runs on: the quarterly cycle (what gets set, reviewed, killed) and the weekly beat inside it (what gets planned, what gets shipped, what gets looked at). Every recurring ritual in the company must trace to a slot in this cycle; this section is what the meeting protocol is later judged against.

02 · Decision framework — who decides what, by seat not by name. The reversibility test: reversible decisions get made fast by the nearest seat and logged; irreversible ones get written down before they’re made, with context and a revisit date. State the threshold for “write it down” — a decision framework that lives only in heads is the drift this document exists to prevent. Decisions land in the decisions table, not in chat scrollback.

03 · Meeting protocol — the complete list of meetings this company holds: each with its purpose, cadence, and which slot of the planning cycle it serves. Then the refusals: the meetings this company does not hold, named on paper (the status theater, the standing sync with no decision rights). A meeting that can’t point at a planning-cycle slot doesn’t get scheduled — it gets added to the refused list.

04 · Documentation & communication norms — where truth lives: which documents are canonical, what belongs in a doc versus a message, and the async default (write first, meet only when writing failed). One home per fact, per protocol; this section names the homes.

05 · Roles map — every seat the business needs to run, whether or not anyone occupies it: what the seat owns, what it decides (citing section 02), and its status — filled (by whom), unfilled, or agent-fillable (an agent could hold it today, with what supervision). Unfilled seats are listed with the same care as filled ones. This map feeds the master intranet’s ghost town and its hiring recommendations directly — it is the floor plan of the company the founder hasn’t hired yet.

06 · Hiring & onboarding — how a new seat gets defined (owns / decides / is measured by, citing business-model) and how a human or agent gets onboarded into it. The onboarding path IS the documentation system: a new occupant reads the workspace’s canonical documents and the module CLAUDE.md files, in a stated order, and is productive when they can update them correctly. If someone can’t be onboarded from the docs, the docs are the bug — name what’s missing.

The module

Per ../PROTOCOL.md (order 7, port 4007, upstream: vision, business-model):

• OPERATIONS.md — the canonical document, sections above.
• operations.json — contract: { name, cadence: { quarterly, weekly }, decisionFramework: { reversibleTest, writeDownThreshold, rights[{seat, decides}] }, meetings[{name, purpose, cadence, servesSlot}], refusedMeetings[{name, why}], norms: { canonicalDocs[], asyncDefault }, roles[{seat, owns, decides, status: "filled" | "unfilled" | "agent-fillable", occupant, notes}], hiring: { seatDefinition, onboardingPath[] } }.
• Core tables — decisions (decision, context, decided_by, reversible, decided_on, revisit_on) and meetings (meeting, purpose, cadence, notes_path): the operational half. The decision log is the living side of this stone — the document says how decisions get made; the table is where they actually land, week after week.
• MCP — the four standard tools (operations_get_document, operations_get_contract, operations_query, operations_record).
• Site — single-page-scroll document site, one section per rubric entry, per the architecture contract in ../../potatuhs-design/SITE.md (sidebar nav, scroll-sync, print-to-PDF). If a brand stone exists in the workspace, render with its tokens; otherwise ship the neutral default theme and note it in status.
• CLAUDE.md — states that the master intranet’s ghost town reads the roles map, so a change to section 05 changes what the intranet renders and recommends; and that a decision recorded in the decisions table which contradicts a canonical document must be flagged for the founder — never silently absorbed into the document, never silently ignored in the log.

Hard rules

• Unfilled seats stay visibly unfilled. The roles map never fakes occupancy — no founder listed in nine seats to make the map look staffed. Emptiness is the onboarding funnel, per protocol; an honest empty seat is this stone’s most valuable output.
• Every decision gets logged with its reversibility. A decision without a reversible flag and a decided_by seat isn’t logged, it’s gossip. Irreversible decisions also get a revisit_on date.
• Meetings that don’t serve the planning cycle get refused on paper. The refused list is part of the manual, not an aside — future occupants of any seat can point at it instead of relitigating.
• The manual describes how work happens, not how it sounds good. If the founder’s real cadence is one weekly review, lock that — an aspirational ritual nobody runs poisons the document’s authority.
• This stone defines process, not strategy. What to build lives in vision; what to measure lives in business-model. This document says who decides, when, and where it’s written — and cites upstream for the rest.
• Placeholders over fabrications, loudly, per protocol.