# Chief Of Staff By Stage > Diagnoses which operating stage a company is actually at, then routes the Chief of Staff role to the right playbook and writes the charter that scopes it. Produces a diagnosis from five observable signals rather than from the last round raised, a named archetype, a charter covering what the role owns, decides alone, escalates, stops and transfers, and a plain statement of any mismatch between the role being offered and the company that exists. Use this skill when scoping or hiring a Chief of Staff, when someone is starting in the role, when a round or a merger has changed the company's size, when a Chief of Staff cannot say what they own, when a founder asks what to use theirs for, or when someone says the role is not working and nobody can say why. - Skill: `ingridleiria/chief-of-staff-by-stage` (Agent Skill) - Install (CLI): `npx skillmds@latest add ingridleiria/chief-of-staff-by-stage` - Raw SKILL.md: https://api.skillmd.com/api/skills/ingridleiria/chief-of-staff-by-stage/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: Coding & Dev Tools - Author: ingridleiria (https://skillmd.com/u/ingridleiria) - Updated: 2026-09-17 - Page: https://skillmd.com/skills/ingridleiria/chief-of-staff-by-stage --- # Chief of Staff by Stage The title is stable and the job is not. A Chief of Staff at twenty people and one at eight hundred share a reporting line and almost nothing else. Most of the unhappiness in the role comes from doing the previous stage's job well after the company has moved past it, and most of the bad hires come from scoping the role against the round the company just raised rather than against the company that exists on Monday morning. The cost is rarely visible for six months and is then expensive. A senior person hired to synthesise across a leadership team that does not yet exist spends two quarters inventing work and leaves. A person hired to build the first systems at a four hundred person company drowns inside a month. In both cases the diagnosis was available on day one and nobody made it. This skill is the diagnostic. It does not tell you how to do the job; it tells you which job it is, and hands you to the playbook that covers it: `cos-at-seed`, `cos-at-series-a`, `cos-at-series-b`, `cos-at-series-c`, or `cos-at-series-d`. ## When to use this, and when not to Use it when the role is being scoped or advertised, when someone is in their first month in it, when a round or an acquisition has just changed the company's size, when the holder has accumulated work for eighteen months and cannot say what they own, or when a founder asks what they should be using this person for. Use it especially when the complaint is vague. "It is not working" and "I am not sure what she does" are almost always stage mismatches, and they are diagnosable in an afternoon. Do not use it to appraise an individual; that is a management conversation and this would give it false objectivity. Do not use it to redesign the reporting structure. Do not use it once the stage is settled and the question has become how to run a planning cycle or a board meeting: open the stage playbook, which carries the method. Everything operational belongs to one of the five: `cos-at-seed` builds what does not exist, `cos-at-series-a` rebuilds and hands over, `cos-at-series-b` synthesises across a leadership team, `cos-at-series-c` carries a named mandate, and `cos-at-series-d` acts under written authority. ## What you need before starting **Headcount and the reporting structure.** Not the number in the deck: the number of people paid this month, and how many layers sit between the chief executive and someone doing the work. Missing: pull reporting lines from a connected directory or human resources tool where one exists; otherwise draw the chart on one page and ask the chief executive to correct it, which is usually the most revealing ten minutes of the diagnosis. **Who owns each function, and whether they own its numbers.** A title is not ownership. The test is who explains a miss. Missing: ask the chief executive who they call when revenue is down, when the product is late, and when someone resigns. Three names, or one name three times, and the second answer is the finding. **The last two months of the leadership meeting.** Agendas, notes, decisions. Missing: attend the next one and take the notes yourself; whether the meeting produces decisions is a signal on its own. **Funding history and runway.** They set the pressure the role works under. Missing: ask for round size and months of runway. If nobody answers quickly, that is a seed signal whatever the headcount says. **What the chief executive spends their week on.** The strongest single input. Missing: reconstruct three weeks from their calendar and categorise it. Where calendar access is unavailable, ask them to list the last ten things they were pulled into and who else could have handled each. **The role as currently written.** A description, a charter, or the previous holder's objectives. Missing: proceed, and record that no written scope is the most common root cause of the complaint that brought you here. ## The method 1. **Count layers, not people.** Everyone reporting to the founder is seed. One layer of functional leads is Series A. Two layers with leaders who own their numbers is Series B. Three or more, with managers of managers, is Series C. Separate units or geographies with their own results is Series D. Where headcount and layers disagree, layers win: layers describe how information moves, headcount only describes cost. 2. **Test whether the leadership team is real.** Could the chief executive be uncontactable for two weeks without anything material stalling. No means there is no leadership team whatever the titles say, and the role is at most Series A. Yes for operations but not for decisions above a threshold is Series B. Yes entirely is Series C or later, where the value sits in mandate and governance rather than coordination. 3. **Locate the constraint.** Only one is true at a time: finding what works is seed, making it repeatable is Series A, scaling it across functions is Series B, running at scale without losing speed is Series C, governing a portfolio and satisfying outside obligations is Series D. Where two feel true, take the earlier one, because companies consistently believe they are a stage ahead of where they are. 4. **Test cadence durability.** A rhythm cancelled when the quarter is tight does not exist. No cadence is seed, a cadence that breaks under pressure is Series A, one that holds and produces decisions is Series B or later. 5. **Test the chief executive's memory span.** Naming every employee is seed. Every customer but not every employee is Series A. Neither, but every leader and the largest accounts, is Series B or C. Neither, and relying on reports, is Series D. 6. **Score the five and resolve conflicts with a rule.** Write the stage each signal indicates. Where they agree, you are done. Where they disagree, scope the operating work to the lowest stage that any two signals indicate, and take the governance work from the highest stage that any single signal indicates. A company that raised a large round while still being run out of the founder's head is a seed company with more money and a real board, so it needs seed-stage building and Series C governance. 7. **Route to the playbook explicitly.** Name the archetype and the folder. A diagnosis that does not end in a named playbook has not finished. 8. **Write the charter.** Five headings, none optional: owns, decides alone, escalates, stops, transfers, with a name and a date against every transfer. The transfer list is what stops the role becoming a permanent accumulation, and it is the heading people skip. 9. **Name the mismatch before the hire rather than after.** Where the offered role is a stage away from the company that exists, write one paragraph saying which stage the company is at, what the offered scope assumes, and what month four looks like if it is accepted unchanged. That paragraph is uncomfortable and it is the most valuable thing in the document. 10. **Set the re-diagnosis date and its triggers.** Six months by default, and immediately on any of three events: a round closes, headcount rises by half, or a new management layer appears. Stages do not announce themselves, which is why they are missed. ## The five archetypes in one table | Archetype | Playbook | Rough size | Constraint | Decision rights | Characteristic trap | | Builder | `cos-at-seed` | 5 to 29 | Finding what works | Informs, rarely decides | Permanent owner of everything built | | Rebuilder | `cos-at-series-a` | 30 to 99 | Making it repeatable | Decides on process | Rebuilding faster than anything is adopted | | Synthesist | `cos-at-series-b` | 100 to 299 | Scaling across functions | Advises, influences | Becoming a very senior project manager | | Mandate holder | `cos-at-series-c` | 300 to 999 | Scale without losing speed | Decides inside the mandate | Scope dissolving into leftover work | | Proxy | `cos-at-series-d` | 1,000 and up | Governing a portfolio | Decides in the chief executive's name | Power without accountability | The bands do not overlap, so no headcount belongs to two archetypes: a company of exactly three hundred reads as a mandate holder on that column and a company of ninety-nine as a rebuilder. The precision is there to stop two people reading the same number differently, not because it means anything. The headcount column is the least reliable row in the table, and where it disagrees with the layer count the layer count wins. The diagnosis comes from the five signals. ## The transitions, which is where the role breaks Nothing announces a transition. The previous version of the job stops working, and because the person is good at it they work harder at the thing that has stopped mattering. Three signals that one has been missed: the Chief of Staff still owns a function that now has a leader; they run a meeting a function should own; the chief executive asks for things a competent leadership team should handle. Any one warrants a re-diagnosis that week, and three questions asked out loud with the chief executive: what am I stopping, who is it going to, what am I picking up. A role that only accumulates ends in burnout for the holder and resentment from the leaders whose territory it occupies. ## Worked example **Situation.** Merrow Health, a company selling scheduling software to hospitals, had closed a 44 million Series B nine months earlier and grown from 55 to 168 people. The chief executive wanted a Chief of Staff and had written a scope built around board strategy, synthesis across the leadership team, and the annual planning cycle. Two candidates had already declined after meeting the leadership team, which nobody had treated as information. **Task.** Decide within a week whether that scope was correct, and produce a charter the next candidate could be held to. Good meant a charter that would still describe the job in month six. **Action.** The five signals took two days. Layers: six functional leaders on the chart, four hired in the previous five months, three still with the chief executive in their approval path for anything over 10,000, so effective layers, one. Leadership reality: the chief executive took four days off in March and returned to two stalled escalations and an unmade pricing decision. Constraint: 71 percent of closed revenue across two quarters had been touched personally by the chief executive, so nothing was repeatable. Cadence: the weekly leadership meeting had been cancelled five times in eleven weeks, every time in a board or customer week. Memory span: the chief executive named all 168 employees in one sitting. Four signals read Series A; the board read Series C, with an independent director, two committees, and a real audit calendar. The wrong turn came first. The initial charter accepted the offered scope and built the role around a monthly synthesis across functions, because headcount and round both pointed there. It was abandoned after one attempt to produce that synthesis: four of six functions had no numbers to reconcile, two defined an active customer differently by a factor of about two, and the resulting page told the chief executive nothing new. Synthesis needs functional reports to synthesise. The rewrite scoped operating work from `cos-at-series-a`: remove the three approval paths still running through the chief executive, rebuild the cadence with a rule that it is never cancelled, run one real planning cycle, and settle the metric definitions before anything is reconciled. Governance duties came from `cos-at-series-c` under the conflict rule, because a real board had arrived before the operating company had. **Result.** The role description lost the word strategic, gained the mismatch paragraph, and gained a first-year outcome about the chief executive no longer sitting in the approval path for standard deals. The person hired had run operations at a similar size rather than the strategy background originally sought. At the six-month re-diagnosis three signals had moved to Series B, the definitions were settled, and the synthesis that was impossible in month one started in month seven and was useful immediately. Scoping one stage too high would have produced an expensive person with nothing real to do, which was predictable from two candidates having already declined. ### A second scenario, where it goes differently A logistics business reached 260 people by merging two companies of roughly 130 each, eighteen months after a Series C. Layers said Series C. Leadership reality said Series B on one side and Series A on the other, because the acquired company's leaders had all left within a year. The constraint said Series A: nothing was repeatable across the halves and there were two of every system. The conflict rule sent operating scope to Series A, the lowest stage two signals indicated, and governance to Series C, the highest any signal indicated. The charter paired a `cos-at-series-a` rebuild, one cadence rather than two and one definition of a shipment rather than two, with the `cos-at-series-c` governance calendar and mandate structure, since integration is exactly the named mandate that stage describes. What changed the answer was not size. A merger resets the operating stage while leaving the governance stage where it was, and averaging the signals would have scoped the role to a company that existed on neither side of the business. ## Output One page, delivered as a document rather than a conversation. ``` STAGE DIAGNOSIS AND ROLE CHARTER Company: [name] Date: [date] Re-diagnose by: [date] Headcount: [number] Layers to an individual contributor: [number] Last round: [round, size, months ago] Runway: [months] SIGNALS | Signal | What was observed | Stage indicated | | Layers, not people | | | | Leadership team is real | | | | Where the constraint sits | | | | Cadence survives a busy week | | | | Chief executive memory span | | | DIAGNOSIS Operating stage: [stage] routed to [exact folder name] Governance stage: [stage] duties taken from [exact folder name] Conflict rule: [applied or not, and which signals disagreed] CHARTER Owns: [3 to 6 items, none of them a function that has a leader] Decides alone: [with a money threshold where the role can spend] Escalates: [always, without exception] Stops: [item, last date it happens] Transfers: [item, to whom, by when] MISMATCH [One paragraph, or: no material mismatch. Where there is one, say what month four looks like if the offered scope is accepted unchanged.] FIRST NINETY DAYS [Three outcomes taken from the routed playbook, each with a date.] ``` ## Failure modes **Diagnosing from the round.** Recognise it when the reasoning contains "they are a Series B company" and no observation about how decisions get made. Fix by running the five signals, which takes two days and is hard to argue with. **Averaging conflicting signals.** Recognise it when the diagnosis lands between two stages and the charter is correspondingly vague. Fix by applying the conflict rule rather than splitting the difference. **Diagnosing the company the chief executive wants.** Founders describe the company they are building, in the present tense, sincerely. Recognise it when every answer is one stage ahead of every document. Fix by weighting observation over self-report and reading the calendar. **A charter with no stop list and no transfer list.** Recognise it because everything in it is an addition. This is what produces the accumulating role that fails at the next transition. Fix by requiring a named person and a date against every transfer before the charter is agreed. **Naming the mismatch privately.** A mismatch raised in a corridor and left out of the written role will be discovered by the person hired, in month four. Put it in the document. ## Edge cases **A company shrinking rather than growing.** After a large reduction the operating signals move backwards and the governance obligations do not. Diagnose the operating stage where it now is, often two stages back, and keep governance where the company was. **Two founders who split the company between them.** The layer count misleads because there are two routing tables. Diagnose each half, scope to the earlier of the two, and say plainly that the split is the constraint. **A first Chief of Staff at a large company.** Where a 600 person company has never had the role, do not scope from headcount alone: there will be a backlog of unowned work characteristic of a much earlier stage. Scope the first six months as a rebuild with an end date, then re-diagnose. **Short runway.** Under nine months, every stage collapses into one job for a quarter: cash, the raise, and nothing else. Scope to `fundraise-readiness` and re-diagnose after the round closes. **The role is a title given to an existing employee.** Same diagnosis, but the charter must name what they are stopping, because their previous work will follow them and quietly consume the role. **Companies with no venture funding.** The round labels are irrelevant and the signals still work. Read layers, leadership reality, and the constraint, and route to the playbook whose operating description matches, ignoring the funding language in it. ## Quality bar - All five signals are recorded as observations of what was seen, not what was reported. - The diagnosis ends in a named playbook folder, and where signals conflicted the conflict rule is shown being applied. - The charter names what the role stops and transfers, each with a person and a date, not only what it owns. - Any mismatch appears in writing with the month-four consequence stated. - Decision rights carry a money threshold wherever the role can spend. - A re-diagnosis date and the three triggers are on the page. - Nothing in the charter is a function that already has a leader. ## Adapting this to your context The five signals were calibrated on venture-funded technology companies between five and a thousand people. The signals travel much further than the round labels do. - **The round names.** They are shorthand for headcount, decision layers and board formality, nothing more. A family business, an agency, an NGO, a university department or a bootstrapped company should read the five signals and ignore the funding language in whichever playbook it lands on. - **The headcount bands.** The least reliable column in the table, and taken from software companies. Labour-intensive operations run two or three times the headcount at the same number of layers; research, investment and design firms run a fraction of it. Layers win, always. - **The leadership reality test.** Two weeks of an uncontactable chief executive suits a company. For a partnership, a clinical department or an owner-managed business, ask instead whether a decision above the usual threshold can be made while the principal is away. - **Runway and the round.** Where nobody raises money, substitute the equivalent pressure: the grant cycle, the seasonal cash trough, the parent company's budget round, or a succession date. - **What not to change.** Diagnose from observed signals rather than from the round, and apply the conflict rule rather than averaging when the signals disagree. ## Related skills This skill routes to the five stage playbooks and does none of their work: `cos-at-seed`, `cos-at-series-a`, `cos-at-series-b`, `cos-at-series-c`, `cos-at-series-d`. Diagnose first, then work from exactly one of them. `onboarding-plan` turns the charter into a first ninety days. `operating-cadence-design` builds the rhythm the cadence signal tests for. `annual-planning-and-headcount` is where the layer count becomes a plan. `principal-simulator` predicts how the chief executive will react to the mismatch paragraph. `decision-memo` is the format for the choice the diagnosis often creates, including whether to hire at all.