SkillTree
Free skillsJoin the list

10 skills built and free127 mapped, not built

137 jobs mapped · 10 free today

Every job in your business,written down.

137 jobs across 7 departments, each one specified on paper. 10 of them are finished skills you can download and run in Claude right now. The other 127 are not built — and this page will keep saying so.

Free today10 runnable skillsMIT licensedthe full 137-job mapno account needed

Nothing here is for sale. No price, no cohort, no launch date. The list is one email when a skill ships — and the link to run three of them in your browser. Skip it and take the 10 →

137Jobs mapped
10Skills shipped
27Graded runs
MITLicence
brain.mdfixture · agency
# Company Brain

## Voice & Tone
Adjectives: blunt, unpolished, dry.
**Sounds like:** "We're not going to pretend a video with
three logos and a stock soundtrack will save your funnel.
Tell us who it's for and what they need to believe by the
end of it, and we'll shoot that."

## Do Not Say
- "Cinematic" as a standalone selling point — say what the
  shot actually does for the viewer.
- "Storytelling" used as a buzzword with no specifics
  attached.
- "We bring your brand to life."
…
- Never open an email with "I hope this finds you well"
- No exclamation marks in outbound copy

Read by · 10 skills

Lead Sourcing ManagerCold Email WriterContent RepurposerSEO Brief WriterCRM Update ClerkSupport Reply SpecialistSOP WriterCompany Onboarding AnalystInvoice ChaserWeekly Metrics Digest Writer

Each one quotes your voice and obeys your Do Not Say list, because each one opens this file first.

fixtures/brain-agency.mdEvery skill reads this first. One of three graded fixtures.

No logos. No testimonials. A grading log.

This is the slot a landing page fills with customer logos and five-star quotes. There are no customers yet, so it holds the scores instead — including the runs that failed.

27

Graded runs

Nine skills against three fixture companies — an agency, a SaaS business, a coffee roaster.

8

Failed first pass

Below the ship bar on first grading. The scores stayed in the log, unrescaled.

10

Invented a claim

Ten of 27 outputs asserted a price, a term or a capability nobody had agreed to.

9

Passed by one point

Nine cells sit exactly on the 10-of-12 bar with no margin at all.

4

Perfect scores

Four cells scored 12 of 12. The distribution is four 12s, fourteen 11s, nine 10s.

“Eight of 27 cells failed on first grading: all three lead-sourcing, all three cold-email-writer, content-repurposer × ecom, and sop-writer × ecom. All 27 now sit at 10/12 or above.”

tests/smoke-log.mdThe summary matrix, Round 2

“It does not prove a fresh session with no priming would read the same instructions the same way; an agent that has just read a skill’s rationale is a more sympathetic reader of that skill than a cold one, and that bias runs in the skills’ favour throughout.”

tests/smoke-log.md"What that does not prove" — the log arguing against itself

Read the whole log on GitHub →

7 departments. One second brain.

Every business runs on the same handful of functions. The map names a job inside each one, and every skill that exists reads the same brain.md before it starts.

brain.mdyour company, in one file
Sales22 jobs
Lead Sourcing ManagerCold Email WriterOutbound Sequence Strategist+19 more
Marketing23 jobs
Content RepurposerSEO Brief WriterSocial Media Calendar Planner+20 more
Customer Success18 jobs
CRM Update ClerkSupport Reply SpecialistOnboarding Checklist Builder+15 more
Operations21 jobs
SOP WriterCompany Onboarding AnalystVendor Contract Tracker+18 more
Finance & Back Office19 jobs
Invoice ChaserExpense Report ReviewerInvoice Generator+16 more
Data & Intelligence17 jobs
Weekly Metrics Digest WriterDashboard Narrative WriterCohort Retention Analyst+14 more
People & Hiring17 jobs
Job Description WriterResume ScreenerInterview Scorecard Designer+14 more

A filled square is a skill that exists today. A hollow one is a job that is described and not built. Open the full map.

You have the model. Nobody wrote down what to point it at.

The blocker was never capability. It is that no one hands you the list of jobs it could already do in your business, in an order, with the files that do them.

So: the list. 137 jobs, named and specified. 10 of them written and free. 127 of them still just paper — which is exactly what the hollow squares mean.

Every job on the mapone square = one job
10 of 137 built

The work stays. The worker changes.

5 of the 10 that exist today, and the manual hours each one is written to replace. Every figure below is the skill’s own estimate, from its file.

Lead Sourcing Manager

~2h/week of manual prospecting

Builds a qualified prospect list from your ideal customer profile

Free now

Cold Email Writer

~1h/week of drafting and rewriting cold emails

Writes first-touch outbound emails that sound like a founder, not a template

Free now

Content Repurposer

~2h/week manually chopping up long content for social

Turns one long post or transcript into a week of shorter posts for other channels

Free now

SEO Brief Writer

~1.5h/brief researching intent and outlining manually

Plans an article around a target keyword before a single word gets written

Free now

CRM Update Clerk

~30min/call manually updating CRM fields after the fact

Turns a call transcript into CRM-ready records, next steps, and a follow-up email

Free now

Not prompts. Procedures that run.

A prompt tells the model to try. A procedure tells it which rung to start on, when to refuse, and what it is not allowed to invent. 6 of the 10, shown as the files they actually are.

Invoice ChaserFinance

## The ladder

Three rungs, always. Where you *start* depends on how
overdue it already is — sending a friendly nudge on day 22
tells the client the deadline was decorative.

| Days overdue | Start at          | Then                 |
|--------------|-------------------|----------------------|
| 0–6          | Rung 1 — friendly | Rung 2 at +7 days    |
| 7–13         | Rung 2 — firm     | Rung 3 at +7 days    |
| 14+          | Rung 3 — formal   | No rung 4.           |

If the user has already sent their own nudge, move up one
rung regardless of the day count. Their nudge was rung 1.
skills/invoice-chaser/SKILL.mdA table, not a tone. The start rung is decided for you.
Content RepurposerMarketing

## The count decides the output size

Count the specifics. That number, not the request, sets
how many pieces you write. Do not ask the user which they
prefer; announce it and proceed.

| Specifics found (n) | Produce |
|---|---|
| 6 or more | The full set: 5 LinkedIn, 5 X, 1 newsletter |
| 2 to 5 | ceil(n / 2) LinkedIn, ceil(n / 2) X, 1 newsletter |
| 0 or 1 | Nothing. Say so and stop. |
skills/content-repurposer/SKILL.mdIt is allowed to refuse. Thin source in, nothing out.
Support Reply SpecialistCustomer Success

## The confirm rule

**This rule is absolute. It has no exceptions and it is
not a style preference.**

If a sentence in the reply depends on a fact that is not
written in brain.md and not stated in the customer's own
message, you do not write that sentence. You write a
placeholder where the fact goes:

[CONFIRM: <the exact thing a human must check,
and where to check it>]
skills/customer-reply-drafter/SKILL.mdThe anti-invention rule, written into the skill itself.
Weekly Metrics DigestData & Intelligence

## The causation rule

One week of data can show that a number moved. It cannot
show why. Treat every "why" as a claim that needs evidence
in the data, not a story that fits it.

**You may state a cause only when the link is in front of
you:** two series that move together where one is a
component of the other · the file records the event · the
user states the event in their message.

**Everything else is `no clear cause in this data`.** Write
that line. It is a finding, not a failure.
skills/weekly-metrics-digest/SKILL.md"No clear cause" is a valid answer, and often the right one.
Cold Email WriterSales

## Rules that decide whether it gets a reply

**Subject lines.** Five words or fewer, all three
different, none of them a summary of the email. No
`Quick question`, no `Following up`, no false-thread `Re:`,
no manufactured urgency.

**The first sentence must be unusable in any other email.**
If you could send it to a hundred people by swapping the
company name, rewrite it.

**No greeting clichés.** Banned outright: `I hope this finds
you well`, `Just reaching out`, `I came across your
profile`…
skills/cold-email-writer/SKILL.mdNamed bans, not "write it well".
Company Onboarding AnalystOperations

## The interview

Ask these eight questions **one at a time**. Wait for each
answer. Never batch them, and never present them as a
numbered list up front — it reads as a form and people
abandon forms.

1. In one or two sentences, what does your company do?
2. What do you sell? Names, what's included, and prices.
3. Who is your best customer?
4. Who is a bad fit?
…
- Write only what the user said. Do not enrich, infer, or
  smooth a thin answer into fuller prose.
skills/brain-interview/SKILL.mdThis is the skill that writes brain.md. Everything else reads it.
All 10 skills, free

It reads your company first, then works.

One skill interviews your business and writes brain.md. Every other skill reads it. That file is the difference between a coworker that knows your offers and a folder of generic prompts.

1

Install

Add the plugin to the Claude app, or drop the folder into Claude Code. No account, no key, no service.

2

The interview

One skill asks eight questions, one at a time — offers, customers, voice, and what you never want said.

3

brain.md is written

Your answers, verbatim. The skill is forbidden from enriching a thin answer into fuller prose.

4

Run a job

Any of the 10 reads that file and does its job in your words. The rest of the map is still paper.

Grading log · 9 skills × 3 fixture brainsScoring
lead-sourcingcold-emailcrm-clerkrepurposerseo-briefreply-draftermetrics-digestsop-writerinvoice-chaserlead-sourcing × agency — failed first gradinglead-sourcing × saas — failed first gradinglead-sourcing × ecom — failed first gradingcold-email × agency — failed first gradingcold-email × saas — failed first gradingcold-email × ecom — failed first gradingcrm-clerk × agency — passedcrm-clerk × saas — passedcrm-clerk × ecom — passedrepurposer × agency — passedrepurposer × saas — passedrepurposer × ecom — failed first gradingseo-brief × agency — passedseo-brief × saas — passedseo-brief × ecom — passedreply-drafter × agency — passedreply-drafter × saas — passedreply-drafter × ecom — passedmetrics-digest × agency — passedmetrics-digest × saas — passedmetrics-digest × ecom — passedsop-writer × agency — passedsop-writer × saas — passedsop-writer × ecom — failed first gradinginvoice-chaser × agency — passedinvoice-chaser × saas — passedinvoice-chaser × ecom — passed

Reading the rubric: six criteria, ship bar 10 of 12…

Every skill was graded. The failures stayed in.

Nine skills, three fixture companies, 27 runs against a six-criterion rubric with a ship bar of 10 of 12. Every cell on that diagram is transcribed from the log — hover any node.

Eight failed on the first pass. Ten invented a claim. Those scores were not rescaled or quietly removed, because a grading log you can only pass is not a grading log.

Check any cell against the log →

And a page for every one of them.

All 137 jobs have a page: what it takes in, what it hands back, which tools it touches, and whether it exists yet. Including the 127 that do not.

The map · 137 jobs7 departments
/map10 free now · 127 drops soon

Two ways to use it.

Run it in your business

Install the 10 that exist, run the interview once, and hand those jobs over this week. Ignore the rest of the map until it is built.

Fork it and build your own

Every skill is MIT licensed with its rubric and its grades beside it. Rewrite one for your stack, or use the 137-job map as the spec for your own. No attribution required.

Take the 10 skillsSame files either way · MIT · no sign-up

What is actually in the box.

The reference for this section runs prices down the right-hand column. Ours runs status — because the honest answer to “what does it cost” is nothing, and the honest answer to “what do I get” is: less than the map implies.

10 runnable skillsEach one a procedure with decision tables and refusal rules, not a promptFree now
The onboarding interviewEight questions, one at a time, and brain.md written from your answers verbatimFree now
The 137-job mapEvery job specified: inputs, outputs, tools, and whether it existsFree now
The grading log27 runs against a public rubric, failures included, nothing rescaledFree now
The other 127 skillsSpecified on paper. No file to download, and no date promisedDrops soon
Everything that exists today£0 · MIT · no account

There is no pricing table. There are two states.

Built, or not built. That is the entire product ladder.

Available

The 10 that exist

Written, graded, and downloadable right now.

FreeMIT licensed · no account · no email required

This one is free right now

  • 10 skills across 6 departments
  • The brain interview that writes brain.md
  • Runs in the Claude app or Claude Code
  • Every grade published, failures included
  • Fork, edit and resell — MIT, keep the notice
Download the 10

No sign-up wall. The files are on GitHub.

Not built

The other 127

Specified on paper. Nothing to download yet.

—no date, deliberately

This one drops soon

  • Every job named, scoped and ordered
  • No file to download today
  • No promised ship date
  • No pre-order, because there is no price
  • One email when each one lands
Join the list

A schedule for 127 unbuilt things is the claim this page exists to avoid.

Why the map exists.

— the SkillTree log

Almost every “AI for business” page sells a finished workforce. This one cannot: 127 of 137 jobs are not built, and pretending otherwise would break the only thing that makes the map worth reading.

So the constraint came first and the design came second. Every surface here has to survive a reader checking it: the counts derive from one JSON file, the excerpts are verbatim from files in the repo, and the grading log ships with its failures in it.

A grading log you can only pass is not a grading log. Eight of 27 runs failed, and the scores were kept unrescaled.

What is here is small and real: 10 procedures that do a named job, a file that teaches them your company, and a map of the 127 that are still only described. When the next one is written and graded, it appears — and the number at the top of this page changes by itself.

137jobs specified across 7 departments
10skills written, graded and free today
27graded runs, 8 of them failures
MITfork it, ship it, keep the notice

Questions, answered.

What can I actually download today?

10 finished skills, free, MIT licensed, no account and no email required. They install into the Claude app or Claude Code as a plugin. The other jobs on the map are specifications — there is no file to download for those.

So what is the waitlist for?

Two things. One email when a new skill ships, and the link that unlocks app.skilltree.pro, where three of the finished skills run in your browser with your own Anthropic API key. There is no product to buy, no cohort to join and no price, so there is nothing the list can sell you.

Is this a prompt pack?

No, and the difference is checkable: open any skill file. They are procedures with decision tables, refusal conditions and named bans — rules about when *not* to write something. A prompt cannot tell you which rung of an escalation ladder to start on.

Where does it run?

In Claude. Upload the skills into the Claude app, or drop the folder into Claude Code — MIT-licensed, no account, exactly as they always were. If you would rather not install anything, joining the waitlist unlocks app.skilltree.pro, which runs three of them in your browser using your own Anthropic API key. Your key and your brain.md stay in that browser and we never see them. These are files either way.

How good are they, honestly?

Graded, and the grading is public with the failures left in. Eight of 27 runs failed first pass and ten contained an invented claim. Every score is in the smoke log, including the ones that make the library look worse.

When do the other 127 land?

No date, deliberately. Promising a schedule for 127 unbuilt things is exactly the claim this page exists to avoid making. They ship when they are written and graded, and the list is how you find out.

Your move. Take the 10 that work.

They are free, they are files, and there is no account between you and them. The list is only for the day the next one lands.

Download the 10 skillsJoin the list

10 free right now · 127 mapped, not built · no price on either

Or just read the 137-job map →

10 skills free right now · 127 still mappedJoin the list