Claude is the best colleague you have ever hired. It is also asleep. Every night it forgets your name, your prices, and the thing you have now explained four times, twice using diagrams. Skeleton Staff is the house it wakes up in: your voice, your rates, your settled arguments, and 26 specialists who read the lot before they touch your work.
No launch date. No newsletter. No man with a headset called Brad.
Runs on Claude Code. Built by two people and a lot of skeletons.
Skeleton Staff & Co.
open all night
The ledger
At the heart of the house. Staff read it before they lift a pen, and it survives every session.
built for Claude, not bolted onto it
This is not a wrapper. It is not a chatbot in a hat. It installs into Claude Code and speaks its actual language: skills, subagents, memory, MCP. Claude is the only model with somewhere to put things. Almost nobody puts anything there. We put an entire company in, including the arguments.
You bring your own Claude subscription. We do not resell tokens, mark up your API bill, or stand between you and Anthropic like a middleman in a beige raincoat.
what it has
the average of the internet
what it is missing
absolutely everything about you
Ask a raw model for your ad, your proposal, your pricing, and it answers from the average of the internet. It has not read your brand voice. It does not know your margins. It has no idea you settled this in March, in writing, twice, and that the second time was quite tense.
What comes back is competent, confident and completely forgettable. Then you spend the afternoon reintroducing yourself to a machine that will have forgotten you again by teatime. This is not a small tax. This is most of your week.
Prompt packs do not fix this. A smarter prompt is still a stranger holding a nicer script.
not prompts in a trench coat
There is a whole industry selling you fifty AI agents in a zip file. It is fifty strangers. They have never read your handbook, because you were never sold one. They are all, in a sense, on their first day. Our specialists read your house first, so they know how you sound and what you charge before they write a word. They run a ship gate before you see anything, so nothing invented and nothing banned reaches your inbox. Then they write down what they settled, so the next one already knows. That is the whole trick. Not a smarter instruction. A colleague with a memory.
An installer, ten department manuals, two ledgers. Your voice, your economics, your compliance rules, your settled decisions, your real numbers. It sets the doctrine and does none of the work, like a very good chairman.
26 of them, across 9 departments. Each is a Claude subagent with a job, a handbook and a conscience. They read the house, do the work, check themselves, write down what they decided, and never once mention that they are working nights.
Act IThe house
Every Claude session dies. Close the tab and it is gone: your voice, your prices, and the thing you settled in March and have since explained four separate times while quietly losing the will to live. The house is what they come home to. The ledgers survive. The doctrine survives. The handbook survives. Tomorrow it walks in already knowing, and does not ask how your weekend was.
A reply to a lead asking what you charge.
Thanks for getting in touch. Straight answer: the full build is £4,800, fixed.
skills/economics/SKILL.md your rates and cost rules, written once
We don't do discovery calls before a price. We decided that in March and it's served us well.
DECISIONS.md the settled call, cited not re-argued
If the scope grows, the ledger keeps score and you'll see every line.
METRICS.md costs tracked where the buyer can check
Plain words, no exclamation marks. That's how we sound.
skills/brand/SKILL.md voice and never-words, enforced
We did not exaggerate this to make a point. It really does put two exclamation marks in. Illustrative example: no real client. Each highlight names the pack file it came from.
❯ install the skeleton staff operating system
claude code detected. installing as skills + subagents
reading the repo... found a payments SDK and an email SDK
prefilling economics: 2 paid vendors detected
? six questions: your voice, never-words, rates, locale, rituals, owners
writing 10 department manuals...
seeding DECISIONS.md and METRICS.md...
verification audit:
✗ claim found with no source: 'saves 40% of costs'
→ corrected: costs are tracked per vendor in METRICS.md, claims cite rows
❯ install complete. The house remembers.
Illustrative transcript. It mirrors the recorded install demo, which follows before launch.
❯ install the skeleton staff operating system
claude code detected. installing as skills + subagents
reading the repo... found a payments SDK and an email SDK
prefilling economics: 2 paid vendors detected
? six questions: your voice, never-words, rates, locale, rituals, owners
writing 10 department manuals...
seeding DECISIONS.md and METRICS.md...
verification audit:
✗ claim found with no source: 'saves 40% of costs'
→ corrected: costs are tracked per vendor in METRICS.md, claims cite rows
❯ install complete. The house remembers.
Illustrative transcript. It mirrors the recorded install demo, which follows before launch.
Act IIThe staff
Every specialist is labelled by how it was actually made, because the alternative is lying and the alternative is extremely popular. Core ones were lifted out of a workflow that was already shipping real things. Contract-built ones were built to the contract and tested on a real business before we let them out of the house. Nobody padded this roster to reach a rounder number. Every one works alone too, if you are not ready for the full staff. It will tell you exactly what it is missing, out loud, which we find less annoying than the alternative.
26 specialists · 12 Core · 14 Contract-built · 9 departments
Growth · card 2
Short-form Scripter
A full vertical-video script under 35 seconds, with shot notes
Growth · card 4
Email Sequencer
A 5-email sequence with two subject lines each
Growth · card 5
SEO Brief Builder
An SEO brief: intent, outline, entities, internal-link slots
Growth · card 6
Repurposer
One artefact recut for up to three platforms, claim intact
Growth · card 7
Competitor Analyst
A competitor teardown with one argued, exploitable weakness
Growth · card 8
Market Scout
A market scan: segments, demand signals, risks, white space
Brand · card 9
Voice Guardian
A voice verdict on a draft, with minimal-edit fixes
Sales · card 10
Proposal Writer
A structured proposal that frames the price, never discounts unprompted
Sales · card 11
Outreach Composer
Three escalating outreach touches, personalised from real detail
Sales · card 12
Objection Coach
A response framework per objection, no manipulation
Sales · card 13
Onboarding Architect
Welcome email, kickoff agenda and an expectations doc
Sales · card 14
Case Study Builder
A before/after/how case study, quotes marked draft until approved
Memory · card 15
Meeting Distiller
Decisions, owned actions and open questions from any notes
Memory · card 16
Performance Analyst
A plain-English readout: what moved, why, one action each
Memory · card 17
Feedback Synthesiser
Themes ranked by frequency and severity, quotes traced to source
Planning · card 18
Project Planner
A plan with milestones, dependencies and a week-one task list
Money · card 19
Pricing Analyst
Three priced options with a test plan and a decision rule
Money · card 20
Cashflow Explainer
A narrative of what moved and what to watch, every figure traced
Money · card 21
Quote & Invoice Drafter
Quote and invoice text with explicit terms and VAT handling
Legal · card 22
Contract Explainer
A plain-English, clause-by-clause walkthrough with risk flags
Operations · card 23
SOP Writer
Numbered steps with actor, tool and done-signal, plus edge cases
Operations · card 24
Process Auditor
A bottleneck map with fixes ranked by effort versus impact
Team · card 25
Hiring Writer
A job description, five screening questions and a scorecard
Team · card 26
Update Composer
A wins/numbers/risks/asks update, numbers traced to source
Whichever ones you hire, they clock on the same way. They read the house, then they start. Nobody asks you twice. Nobody asks what your brand voice is at half four on a Friday.
what the ledger remembers
DUPI Technology Ltd is a two-person consumer-app company that runs its entire operation on this system: planning, shipping, metrics, marketing, all of it through Claude Code. Every figure below is a dated row in its ledger. This product was extracted from that same workflow.
Every figure traces to a dated row in our own metrics ledger. There are no revenue numbers and no user numbers here, because the product has not launched. Anyone showing you those before launch is showing you fiction.
Single specialist
£19
One specialist, yours to keep. Post-launch additions are new £19 SKUs.
Get notifiedDepartment set
£49
Any one department of three or more: Growth, Sales, Memory or Money.
Get notifiedAll specialists
£99
All 26 specialists, every department.
Get notifiedThe Operating System
£199
The installer, 10 department templates and both ledgers.
Get notifiedFull package
Most complete£299
The house and everyone in it.
Get notifiedInstalled with you
We install and calibrate the whole system on your business. On request.
£1,950on request
EnquireRuns on your own Claude subscription. Prices are indicative while we finish arguing about them.
Straight answers. If yours isn’t here, ask us when you join the list.
No. We are a two-person company in England that uses Claude for approximately everything and got fed up of introducing ourselves to it every morning like a very polite goldfish. Skeleton Staff is an independent product built to run on Claude Code. Anthropic has not endorsed it, sponsored it, or, as far as we can tell, heard of us.
Yes, on purpose. This is built on how Claude genuinely remembers things: skills, subagents, project memory. We built it for one model and did it properly rather than for five and did it badly. You bring your own subscription. We never sit between you and Anthropic.
Because free packs are more prompts, and prompts are still strangers. You are buying the house they live in. If a free pack genuinely solves your problem, use it and keep your money. This is for people who have already tried that and noticed.
You type one command into Claude Code. Then you answer a few questions and read the output. If you can order a takeaway, you can hire a skeleton.
Honesty about where each one came from. Core was extracted from a workflow that ran. Contract-built was built to the contract and validated on a real business first. Every specialist wears its label. Nobody gets to claim an extraction they did not do.
No. Buy it once, own it, and we keep improving it for nothing, forever. We are aware this is a terrible business decision. We are doing it anyway.
No, and we have written a whole section about it further up, which is either reassuring or a warning sign.
Día de los Muertos is a festival about remembering people. Memory is the entire product. Also they work nights and they have never once asked for a pay rise.
DUPI Technology Ltd. A two-person company that built this while running itself on it. No engineering team. One extremely well-briefed AI. The proof section above is our own ledger, not a testimonial we commissioned.
There is no launch date, because nothing ships until it has proved itself on our own company first. Leave your email and you will hear at first light.
Opens your email app. No account, no spam, no nurture sequence written by a stranger who is not us.
