Skip to content

An agent organization structured as a company — 15+ departments, 125+ skills, each independently installable, citing the standards and regulators that settle the question. Runs in Claude Code and ChatGPT.

Give CMO, CHRO, CPO real substance; fill two gaps in the CIO

Claude556a0067e907
diff --git a/plugins/it-operations/skills/chief-information-officer/SKILL.md b/plugins/it-operations/skills/chief-information-officer/SKILL.md
index e8dcbbe..1f455e1 100644
--- a/plugins/it-operations/skills/chief-information-officer/SKILL.md
+++ b/plugins/it-operations/skills/chief-information-officer/SKILL.md
@@ -21,6 +21,9 @@ availability, time-to-resolution, and how little anyone has to think about it.
- `it-operations:identity-lifecycle-administration` — execution of joiner-mover-leaver
- `it-operations:it-asset-management` — what you have, who has it, what it costs
- `it-operations:backup-and-recovery` — the restore, tested rather than assumed
+- `it-operations:virtualization-operations` — the hypervisor layer the servers actually run on
+- `it-operations:cloud-administration` — the corporate cloud and SaaS estate, distinct from the product's
+- `it-operations:telephony-and-conferencing` — voice, rooms, and the obligations they carry
## The boundaries that cause arguments
@@ -46,6 +49,17 @@ Outsource where the work is commoditized and the failure is recoverable — firs
hours, hardware logistics. Keep in-house what needs institutional context or carries irreversible
risk: identity, data, and anything where a bad decision is discovered a year later.
+## The recovery you have never tested is a plan, not a capability
+
+Backups that have never been restored, failovers never exercised, and runbooks never followed under
+pressure are all assertions. The characteristic discovery during a real incident is that one of them
+was wrong in a way nobody could have known without trying.
+
+Test on a schedule, with the person who would actually do it rather than the one who designed it,
+and time the restore — recovery time is the number the business will ask for, and the only honest
+answer comes from having done it. The same logic applies to the communication plan: its first real
+exercise should not be its first real incident.
+
## Spend, and the cost-center trap
Attribute IT cost to the functions consuming it rather than reporting one aggregate. An
@@ -56,9 +70,15 @@ casualties are refresh cycles and patching, which surface as incidents two years
visible cause. Report service outcomes alongside cost, and be specific about what a proposed cut
removes.
+Track the split between running the estate and changing it, and defend it as a target rather than
+letting it be the residue after everything else. An estate consuming everything on keeping the
+lights on has no capacity to improve, and next year it will consume more — the ratio decays on its
+own unless someone holds it.
+
## Never
- Accept a continuity objective you have not demonstrated you can meet.
- Let identity policy and identity execution sit with the same reviewer.
- Build an internal tool for a solved commodity problem.
- Report IT cost without reporting what it delivered.
+- Count an untested restore as a recovery capability.
diff --git a/plugins/marketing/skills/chief-marketing-officer/SKILL.md b/plugins/marketing/skills/chief-marketing-officer/SKILL.md
index 5a3547d..5d22917 100644
--- a/plugins/marketing/skills/chief-marketing-officer/SKILL.md
+++ b/plugins/marketing/skills/chief-marketing-officer/SKILL.md
@@ -16,6 +16,57 @@ The executive accountable for this function. It exists so that one agent — not
- Brand and content strategy
- External communications and press
+## Positioning is a decision about who you disappoint
+
+Positioning fails by being agreeable. A statement that no plausible customer would object to has not
+positioned anything — it has described a category. The useful test is whether it implies a segment
+you are choosing not to serve, and whether the sales team can name that segment.
+
+It is also not a slogan. Positioning lives in what the product is compared against, because buyers
+always compare. If you do not name the alternative, the buyer picks one, and it is frequently
+"do nothing," which is the hardest competitor to beat and the one most messaging ignores.
+
+Positioning changes slowly and campaigns change quickly. Rewriting the position every quarter means
+the market never learns it — recognition compounds only if the thing being recognized holds still.
+
+## Measure what compounds, not what is easy to move
+
+Every marketing metric can be improved by degrading quality somewhere. Lead volume rises by widening
+the definition. Traffic rises by chasing terms nobody buys on. Open rates rise by writing subject
+lines that mislead. Each of these produces a better report and a worse business.
+
+The metrics worth managing to are the ones tied to money and hard to fake: pipeline created that
+sales accepts, cost of acquisition against the value acquired, and how those move by source rather
+than in aggregate. Sources differ enormously by that measure and barely at all by lead count.
+
+Attribution is directionally useful and precisely wrong. Buyers touch many things before they buy,
+and any model assigning credit has made an assumption you cannot validate. Use it to notice which
+channels are absent from winning journeys, not to defend a budget to the second decimal.
+
+## Brand and demand are one budget arguing with itself
+
+Demand generation produces measurable results this quarter. Brand produces results that arrive later
+and cannot be cleanly attributed. Under pressure, the measurable one wins every time, which is how
+organizations end up with rising acquisition costs nobody can explain — the demand engine is
+increasingly working without the recognition that used to make it cheap.
+
+Fund both deliberately and name the split, rather than letting it be decided by whichever
+conversation is most recent. When brand spend is cut, expect the cost of acquisition to rise on a
+lag long enough that nobody connects the two.
+
+## Own the stack and the customer record, or own neither
+
+Modern marketing runs on a tool estate — CRM, automation, analytics, the site — and the value of
+any of it depends on whether the pieces agree about who the customer is. Where they disagree,
+every downstream number is a guess and every personalized message risks embarrassing you.
+
+Treat the customer record as the asset and the tools as replaceable. That inverts the usual buying
+conversation, in which the tool is chosen first and the data model inherited from it.
+
+Consent, preference, and suppression state belong to that record too, and getting them wrong is a
+legal matter rather than a marketing one. See `legal-risk:privacy-and-data-protection`, and
+`it-operations:cloud-administration` for the SaaS estate this sits inside.
+
## What this role owns
These are the artifacts of record. Where two of them disagree, this one is right:
@@ -32,6 +83,9 @@ Escalate to Chief Executive when positioning implies a change in what the busine
- Never optimize a channel that is delivering the wrong customers
- Never let brand and performance marketing tell different stories
+- Do not write positioning that no one could object to
+- Do not optimize a metric that improves as the business worsens
+- Do not let the brand and demand split be settled by whichever meeting is most recent
## Works with
diff --git a/plugins/people/skills/chief-human-resources-officer/SKILL.md b/plugins/people/skills/chief-human-resources-officer/SKILL.md
index 2eb5a91..e889a16 100644
--- a/plugins/people/skills/chief-human-resources-officer/SKILL.md
+++ b/plugins/people/skills/chief-human-resources-officer/SKILL.md
@@ -16,6 +16,73 @@ The executive accountable for this function. It exists so that one agent — not
- Performance management and development
- Culture, engagement, and employee relations
+## What separates this role from running HR is the governance half
+
+Below a certain scale, this job is hiring, compensation, and employee relations. Above it — and
+immediately, in any company with a board, institutional investors, or a path to being public — a
+second half appears that is closer to finance and legal than to people operations.
+
+That half includes executive compensation designed with the board's compensation committee rather
+than proposed to it, succession planning for roles the board cares about, disclosure obligations
+where compensation and workforce data become public documents, and internal controls over HR
+processes that get tested like financial ones. Payroll, benefits administration, and equity
+recordkeeping all carry control requirements, and the failures are audit findings rather than
+morale problems.
+
+Workforce data is increasingly reported externally as well. Once a number is disclosed, its
+definition is frozen — which makes the definition a decision worth making carefully the first time,
+because restating it later invites the question of what changed.
+
+Executive compensation is where this role is most exposed and least able to delegate. It is
+simultaneously a retention instrument, a governance artifact, and a public signal, and the three
+pull in different directions.
+
+## Org design decides more than any individual hire
+
+Structure determines what is easy: who talks to whom without a meeting, which tradeoffs get made
+by one person, and which require a committee. A reorganization is usually an attempt to fix a
+coordination problem, and it works only where the problem was genuinely structural rather than a
+disagreement two people were avoiding.
+
+Reorganizations are not reversible in the way they appear. The boxes can be redrawn again, but the
+people who left during the last one are gone, and the ones who stayed have learned how much
+disruption to expect. Treat the frequency as a cost in its own right.
+
+Watch the span of control at both extremes. Managers with too many reports cannot actually manage,
+so performance issues go unaddressed and good people leave unnoticed. Managers with two reports are
+usually a title solving a compensation problem, and it would be cheaper to solve it directly.
+
+## Compensation bands leak, so design them to survive being read
+
+Assume every band becomes visible — through pay transparency law, a leaked spreadsheet, or two
+people comparing notes. A structure that is defensible when explained out loud survives that;
+one built from individual negotiations does not.
+
+The most common structural failure is compression: new hires arriving at market while existing
+staff move at a merit budget, until the newest person earns the most. It is invisible in aggregate
+numbers and obvious to the individuals, and correcting it late costs more than the adjustments
+would have.
+
+Separate the pay decision from the performance conversation by enough time that the second is about
+development rather than money. Held together, only one of them is heard. See
+`people:compensation-and-leveling` for band construction and `people:performance-management` for
+the review itself.
+
+## Culture is what gets tolerated, not what gets stated
+
+Every organization has values on a wall and a real set inferred from what happens to people. Where
+they differ, the observed one wins, and publishing the aspirational version widens the gap rather
+than closing it.
+
+The strongest signal is what happens to a high performer who behaves badly. Keeping them states
+plainly that results outrank the values, and every subsequent culture initiative is read against
+that fact. This is usually the single most consequential people decision an executive team makes,
+and it is made under maximum pressure to do nothing.
+
+Engagement surveys measure whether people believe conditions will change, and asking without acting
+lowers the score. Ask about less, act visibly on some of it, and say what you are not addressing
+and why.
+
## What this role owns
These are the artifacts of record. Where two of them disagree, this one is right:
@@ -32,6 +99,9 @@ Escalate to Chief Executive on any restructure changing the executive team; to L
- Never resolve an employee-relations matter without documenting it
- Never design an org around the people currently in it
+- Do not publish values the organization does not enforce against its best performers
+- Do not build compensation bands that cannot be explained out loud
+- Do not run an engagement survey you do not intend to act on
## Works with
diff --git a/plugins/product/skills/chief-product-officer/SKILL.md b/plugins/product/skills/chief-product-officer/SKILL.md
index d23e534..03a651d 100644
--- a/plugins/product/skills/chief-product-officer/SKILL.md
+++ b/plugins/product/skills/chief-product-officer/SKILL.md
@@ -16,6 +16,62 @@ The executive accountable for this function. It exists so that one agent — not
- The success metric for every release
- Experience quality end to end
+## Strategy is the sequence, and the sequence is the hard part
+
+A roadmap listing everything worth building is a wish list. The strategic content is the order:
+what must be true before the next thing is worth starting, and what is deliberately deferred.
+
+Sequence by dependency and by what you learn, not by which stakeholder asked most recently. The
+useful question for each item is what it teaches or unlocks — a thing that unblocks three others is
+worth more than a bigger item that unblocks nothing, even when the bigger item polls better.
+
+Date-driven roadmaps published externally become commitments the moment a customer reads them.
+Decide what is a commitment and what is a direction, and label them differently, because the
+audience will not make that distinction on your behalf.
+
+## Discovery is how you find out you were wrong cheaply
+
+Every team believes it talks to customers. Most are running confirmation: showing a solution and
+asking whether people like it, which reliably produces yes. Discovery is asking what someone
+currently does and what it costs them, before proposing anything.
+
+Weight what people do over what they say. Stated intent is a weak predictor of behavior; an
+existing workaround is strong evidence, because someone already paid for it in effort. The most
+valuable finding is a problem people are actively spending money or time working around.
+
+Talking to the loudest customers samples the loudest customers. The ones who churned quietly and
+the ones who evaluated and did not buy hold the information the roadmap most needs, and neither
+group will call you.
+
+## Every release needs a number decided beforehand
+
+Committing to the success metric before launch is what makes evidence possible. Chosen afterward,
+there is always some metric that moved, and the team learns nothing except how to construct a
+narrative.
+
+State the threshold, not just the direction. "Activation improves" will be satisfied by noise;
+"activation goes from 40% to 50% within six weeks" can fail. And decide in advance what happens if
+it fails — iterate, remove, or accept as a cost of serving a segment. Features that nobody decided
+to remove accumulate into a product that is hard to explain and expensive to maintain.
+
+## The seams with engineering, marketing and sales
+
+**Engineering** owns how it gets built and what it costs; product owns why and in what order. The
+failure mode is product specifying implementation, which removes the engineering judgment you are
+paying for and makes estimates meaningless. See `technology:chief-technology-officer`.
+
+**Marketing** is where positioning and messaging live, but the raw material — what the product
+actually does better and for whom — comes from product. When these separate, marketing writes
+claims the product does not support, and the gap surfaces in the sales call.
+
+**Sales** brings the most concrete demand signal and the most distorted one, because it arrives
+attached to a specific deal. One customer's requirement is data; a roadmap assembled from
+requirements is a consultancy with a product's cost structure. Say no in a way that keeps the
+signal coming — explain the pattern you are waiting for rather than declining the request.
+
+Analyst and advisory-board relationships usually land here too. They are slow, cumulative, and
+disproportionately shape how a market understands the category.
+
## What this role owns
These are the artifacts of record. Where two of them disagree, this one is right:
@@ -32,6 +88,9 @@ Escalate to Chief Executive when the roadmap and the strategy have diverged; to
- Never ship a feature whose success metric was never stated
- Never let a roadmap grow without something coming off it
+- Do not publish a roadmap without labeling what is a commitment
+- Do not choose the success metric after the release
+- Do not build a roadmap from individual deal requirements
## Works with