One person, an entire company
There is a point in a one-person company where the constraint stops being ideas and becomes bandwidth. You have to understand the market while you build the product, build the product while you find customers, find customers while you write the content, and write the content while you watch the cash. Somewhere in between you are supposed to decide, well, what deserves your attention this week.
Nothing is broken when that happens. There is simply one of you and a whole company to run, and every part of it needs a different kind of thinking. The expensive thing is not any single task. It is rebuilding the method from scratch each time a different part of the company needs you: remembering how a pricing survey is supposed to be structured, what goes in a decision memo, which questions a discovery call has to answer, how a SAFE actually converts.
This pack is that method, written down forty-two times. Each skill is a small text file that tells Claude what to ask you for, which framework to apply, what the output has to look like and what to check before handing it back. Install them once and the thinking is there every time, in Claude Code, in Claude.ai or behind your own code.
Download the pack
42 skills, 25 categories, MIT licensed. A zip with a README, install steps and a folder per skill. Prefer to look first? Read the plain index of all 42 skills.
Skyrocket organic growth for your new product
Make AI recommend you, not competitors
Go viral on every social network
Optimise your website to rank on Google
What a skill is
An Agent Skill is a folder. It holds one required file, SKILL.md, and optionally a references directory for templates and worked examples and a scripts directory for code. Claude reads it the way a new hire reads an onboarding note: only when the task calls for it, and only as far as it needs to.
SKILL.md opens with a short block of YAML. Two fields are required. The name has to match the folder it lives in, in lowercase letters, digits and hyphens. The description is the part that does the work: it is the only thing Claude sees about a skill before deciding to use it, so it has to say both what the skill does and when to reach for it, in the words a founder would actually type. Here is one from the pack, verbatim.
---
name: runway-model
description: Builds a month-by-month cash runway model from current cash, burn and revenue assumptions, with base, bear and bull scenarios and a zero-cash date. Use when a founder asks how many months of runway they have, when they will run out of money, whether they are default alive, or what a hire or price change does to the cash position.
license: MIT
metadata:
author: Tau AI (get-tau.com)
version: "1.0"
category: Financial Planning
---Everything below that block is instructions, in plain Markdown. The pack keeps every skill to the same shape so you always know where to look, and so you can edit one without re-reading the other forty-one.
- When to use it, when not to
- Three or four concrete situations, then the cases that belong to a sibling skill, named. The point is to stop a runway question turning into a fundraising plan halfway through.
- Inputs to ask for
- What the skill needs from you, in what form, and what happens if you do not have it. No skill in the pack invents your numbers. It asks, or it states the default it used and tells you to replace it.
- Method
- Numbered steps that encode a real framework: the nine boxes of a Lean Canvas in the order they are meant to be filled, the four Van Westendorp questions and the four price points they produce, the five components of positioning, the RICE arithmetic. Where a calculation is fiddly, a script does it.
- Output format
- The headings and tables the answer must use, followed by a short worked example. The example is what keeps the output honest: it shows the level of specificity expected, not the vocabulary.
- Quality checks
- The list Claude runs before handing back: arithmetic, missing inputs, claims the evidence does not support, benchmarks quoted as targets.
- Related skills
- Which sibling to move to next, and why. This is what turns forty-two files into a system.
How Claude loads one is the part worth understanding, because it is why installing forty-two of them costs almost nothing.
At startup: the description
When triggered: the instructions
When needed: references and scripts
The format is Anthropic's Agent Skills format, published as an open specification. Anything written to it works in Claude Code, Claude.ai and the Claude API, and in the other agents that have adopted it. Nothing in this pack depends on Tau AI.
The 42 skills, by category
Twenty-five categories, because that is how the work presents itself to a founder: not as functions in an org chart, but as the question that arrives on a Tuesday. Every category has at least one skill. The ones a solo founder meets most often have two.
The list runs long on purpose. It is the index of the pack, and it is meant to be scanned for the question you have this week rather than read top to bottom. The category sits above each name; the one-liner says when to use it; the right column says what comes out.
Categories 1 to 3
Strategy, research and customers
Lean canvas
Use when: an idea is still a paragraph and you need to see all nine boxes at once, and know which assumption to test first
Output: A filled canvas in Ash Maurya's order and a ranked list of riskiest assumptions, each with its cheapest test
One-page strategy
Use when: you can describe what you are doing but not why it should work, or the plan has grown to twelve priorities
Output: Diagnosis, guiding policy, coherent actions, the where-to-play and how-to-win choices, and an explicit list of what you will not do
Market sizing
Use when: someone asks how big the market is and you want a number you can defend rather than one you found on a slide
Output: TAM, SAM and SOM done top-down and bottom-up, reconciled, with every assumption sourced and rated for confidence
Research synthesis
Use when: you have thirty tabs, a folder of PDFs and a page of notes, and need to know what you actually learned
Output: A findings memo: confirmed claims (two independent sources), contradictions logged, confidence per finding, what you still do not know
Customer interview guide
Use when: you are about to talk to customers and want to come back with evidence rather than compliments
Output: A jobs-to-be-done timeline script that asks about past behaviour, never pitches, and a synthesis grid to fill after each call
ICP definition
Use when: you are selling to everyone who replies and need to decide who to stop selling to
Output: An ideal customer profile across firmographics, triggers, pains, tech and buying process, with a weighted 0 to 100 scoring sheet and disqualifiers
Categories 4 to 8
Product, go-to-market, marketing, sales and growth
PRD writer
Use when: you are about to build something for more than a day and have not written down what done means
Output: A working-backwards press release and FAQ, then a light PRD: success metrics, scope in and out, user stories with acceptance criteria, non-goals
Feature prioritisation
Use when: the backlog is longer than the quarter and every item feels important
Output: A RICE-ranked table with reach, impact, confidence and effort shown per row, produced by scripts/rice.py from a CSV
Positioning statement
Use when: people nod at the demo and then ask what it is, or you are describing the product differently every week
Output: The five components in order, competitive alternatives through to market category, and a positioning statement you can put on a homepage
GTM launch plan
Use when: something is ready to ship and the plan for telling people is a tweet
Output: Bullseye channel selection from the nineteen traction channels, three tiers of launch, a before, day-of and week-after checklist, and dated launch metrics
Content calendar
Use when: you know you should publish and keep not publishing, or you publish whatever occurred to you that morning
Output: Content pillars derived from customer pains, a repurposing chain, and a four-week calendar sized for one person
Landing page copy
Use when: the page describes features and visitors leave without understanding the outcome
Output: Full page copy from a message hierarchy in the customer's own words, plus two alternate headlines and the first thing to test
Cold outreach sequence
Use when: you have a list and no replies, or you are writing every message from a blank page
Output: A four to six touch sequence built on a real trigger, one idea per message, under ninety words, ending with a breakup step
Discovery call script
Use when: calls end pleasantly with no next step and you cannot say afterwards whether the prospect has budget
Output: A SPICED script (situation, pain, impact, critical event, decision), a question bank, a next-step close and a matching notes template
Growth experiment backlog
Use when: you have ideas for growth and no way to tell which one to run this week or whether it worked
Output: Hypothesis cards, ICE scores, a minimum detectable effect and the sample size from scripts/sample_size.py, and a learnings log
Activation funnel audit
Use when: signups are fine and revenue is not, and you are not sure where people fall out
Output: A defined activation event, a step-by-step funnel table, the single biggest drop, and fixes ranked by expected lift over effort
Are you visible across search, AI and social? Run a free audit
Complete free diagnostics of you and your competitors
Categories 9 to 13
Operations, planning, time, decisions and problems
SOP writer
Use when: you have done the same task four times and are about to do it a fifth, or you want to hand it to someone
Output: A procedure with trigger, owner, inputs, numbered steps each with a done-criterion, exceptions, a checklist and a review date
Project plan
Use when: there is a date and a pile of work, and the pile has not been turned into an order yet
Output: Tasks of at most two days, milestones planned backward from the deadline, dependencies, the critical path, a risk register and a Now, Next, Later view
Weekly founder review
Use when: every Friday, for thirty minutes, whether or not the week went well
Output: Plan versus actual, a metrics snapshot, wins and misses with a cause each, decisions pending, and next week's three priorities with one needle-mover
Decision memo
Use when: a choice has been circling for two weeks and you keep re-deciding it in the shower
Output: A one-page memo: the decision, its deadline, one-way or two-way door, at least three options including do nothing, weighted criteria, what would change your mind
Pre-mortem
Use when: you are about to commit real money or months to a plan everyone already agrees with
Output: Causes of failure generated before they are clustered, rated for likelihood and impact, mitigations for the top five and dated tripwires
Root cause analysis
Use when: the same problem has come back for the third time after being fixed twice
Output: An is/is-not statement, five whys with a fact at every why, symptom separated from cause, a corrective action and a preventive one with a verification date
Categories 14 to 17
Numbers, money, pricing and fundraising
Metrics dashboard spec
Use when: you look at fourteen numbers each morning and none of them tells you what to do
Output: A north star metric, its input tree, and a precise definition (numerator, denominator, window, source, owner) for at most ten metrics
Cohort retention analysis
Use when: you want to know whether people who signed up in March are still here, and whether the curve ever flattens
Output: A monthly retention triangle and the average curve from scripts/cohort_retention.py, with a flattening test and the caveats small cohorts deserve
Runway model
Use when: you need the zero-cash date, or want to know what a hire or a price change does to it
Output: A month-by-month table for base, bear and bull scenarios from scripts/runway.py, the zero-cash month for each, and a default-alive check
Unit economics
Use when: you are about to spend on acquisition and do not know what a customer is worth
Output: Fully loaded CAC, ARPA, gross margin, churn, LTV, LTV to CAC, payback months and contribution margin from scripts/unit_economics.py, each with a one-line reading
Pricing research
Use when: the price was picked by looking at a competitor and you would like it to be picked by looking at customers
Output: A Van Westendorp survey, the four price points from scripts/van_westendorp.py, willingness-to-pay interview questions and a chosen value metric
Pricing tiers
Use when: you know roughly what to charge and now have to decide what goes in which plan
Output: Good, better, best packaging on one value metric, the fences between tiers, the free-versus-trial decision, and a pricing page outline
Investor update
Use when: on the same day every month, including the months you would rather skip
Output: A short update: three-line summary, metrics against plan, highlights, honest lowlights, specific asks, cash and runway
Round planner
Use when: you are deciding how much to raise and on what terms, and the phrase post-money cap still needs explaining
Output: A raise sized from milestones and burn, a plain-English SAFE glossary, and an ownership table from scripts/dilution.py across stacked SAFEs and a priced round
Categories 18 to 25
Legal, hiring, partners, support, competition and the rest
Contract review checklist
Use when: a contract has arrived and you want to know what to push back on before paying for an hour of advice
Output: A clause-by-clause table: what it says, the risk, what to ask for, and the questions to bring to a lawyer. A reading framework, not legal advice
Hiring scorecard
Use when: you are about to make a first hire and the job description is a list of adjectives
Output: A scorecard with a mission, three to eight measurable outcomes and competencies, then a structured interview plan and A, B, C grading
Contractor brief
Use when: you are handing work to a freelancer and want to avoid the third revision argument
Output: A scope of work with deliverables, acceptance criteria, checkpoints, payment schedule, IP terms and kill criteria, plus a short job post
Partnership proposal
Use when: another company keeps coming up in customer conversations and you want to approach them with something concrete
Output: A fit assessment, the shape that suits (co-marketing, integration, referral), pilot terms with an exit, a one-page proposal and the first note
Support macros
Use when: you are typing the same reply for the fifth time this week
Output: Ten to fifteen macros mined from real tickets with variables in braces, tone rules, escalation triggers and a review cadence
Churn save playbook
Use when: cancellations arrive and your only response is a refund
Output: A cancellation reason taxonomy, an exit survey, a save offer per reason with limits, when not to save, and a 30, 60, 90 day win-back sequence
Competitor teardown
Use when: you need to understand one competitor properly rather than glance at their pricing page again
Output: Their positioning in their words, their ICP, pricing, product, onboarding, channels and mined reviews, every fact dated, ending in implications for you
Sales battlecard
Use when: a prospect says they are also looking at a named competitor
Output: One page: where you win with proof, where they win honestly, landmine questions to ask, objection handling and when to walk away
Founder energy audit
Use when: the weeks are full and nothing important moved, or you are tired in a way a weekend does not fix
Output: A two-week log by category and energy, the givers and drainers, a delegate, automate, delete or keep matrix and weekly protection rules
OKR writer
Use when: a quarter is starting and you want three things to be true at the end of it
Output: One to three objectives, two to four key results each written from X to Y by date, initiatives kept separate, and a scoring rule
Framework picker
Use when: you have a question and are not sure which of the other forty-one skills it belongs to
Output: The skill and framework that answers it, the order for combined questions, or a plain statement that this is a small reversible choice and you should just decide
Use case library
Use when: you want to see the pack used end to end before you trust it with your own numbers
Output: Six worked scenarios, idea to first ten customers, changing prices, preparing to raise, first hire, stalled growth, a launch, each with the chain, inputs, outputs and a timeline
How the skills chain
The reason the pack is forty-two files rather than one long prompt is that founder work arrives in sequences. The answer to one question is the input to the next, and a skill that knows what comes after it can hand over cleanly instead of trying to do everything at once.
Every SKILL.md ends with a related-skills list, and framework-picker exists to route an open question to the right starting point. The chain below is the one most founders run first, from an idea to a launch, and it is the first scenario in the use case library.
From idea to launch, five hand-offs
1. lean-canvas
Starts with the idea as you would describe it to a friend and ends with the nine boxes filled and the riskiest assumption named. In almost every canvas that assumption is about the customer and the problem, not the product, which is why the next step is not building.
2. icp-definition
Takes the customer segment box and makes it specific enough to find people: the firmographics, the trigger that just happened to them, the pain in their words, and the disqualifiers. The output is a scoring sheet, which is what turns a list of names into a list of who to call first.
3. customer-interview-guide
Takes the ICP and the riskiest assumption and produces the script for testing it: a timeline interview about the last time they tried to solve this, the forces that pushed and pulled them, and what they used instead. Ten interviews later the synthesis grid says whether the assumption survived.
4. positioning-statement
Takes the synthesis grid, in particular the alternatives people actually used, and works through the five components in order: what they would do without you, what you have that those alternatives lack, what that enables, who cares most, and which market category makes it obvious. The statement at the end is written in their vocabulary because the grid supplied it.
5. gtm-launch-plan
Takes the positioning and the ICP and decides where to say it: the Bullseye pass over the nineteen channels, three cheap tests, one channel to focus on, and a launch in tiers with a checklist and a dated metric. From here the chain forks into landing-page-copy and cold-outreach-sequence, depending on which channel won.
Four other chains come up almost as often, and the README maps each of them.
- Understanding the numbers: metrics-dashboard-spec, then cohort-retention-analysis, then unit-economics, then runway-model
- Changing prices: customer-interview-guide, then pricing-research, then pricing-tiers, then decision-memo, then premortem
- Preparing to raise: one-page-strategy, then runway-model, then round-planner, then investor-update
- Making a first hire: founder-energy-audit, then sop-writer, then hiring-scorecard or contractor-brief, then contract-review-checklist
Installing the pack
Custom skills do not sync between surfaces. A skill installed in Claude Code is not on Claude.ai, and one uploaded to the API is on neither. Install on each surface you use; the folders are the same.
Claude Code
Unzip the pack
Output: A folder with README.md, INSTALL.md, LICENSE and a skills/ directory holding 42 folders
Copy the folders into your personal skills directory
Use when: you want the skills in every project on this machine. For one repository only, copy into .claude/skills/ at the repo root instead, and commit it if contractors should have the same skills
Output: ~/.claude/skills/runway-model/SKILL.md and 41 siblings. Not ~/.claude/skills/skills/…, which is the one mistake this step allows
Start a new session and check the list
Use when: skills are discovered at startup, so an open session will not see them until it restarts
Output: The 42 names in the list. Type /runway-model to call one directly, or describe the task in plain words and let the description pick it
Claude.ai
Turn on code execution
Use when: custom skills on Claude.ai need code execution and file creation enabled; on some plans the menu is labelled Features
Output: The toggle on, and a Skills entry visible under the same settings
Make one zip per skill
Use when: Claude.ai takes one skill per zip and wants the skill folder at the zip's root, so the pack zip itself will not upload. Run this from inside the skills/ directory after mkdir -p ../zips
Output: zips/runway-model.zip and 41 siblings, each containing <skill-name>/SKILL.md plus its references and scripts
Upload the ones you want and toggle them on
Use when: you do not have to upload all 42; start with the six or eight you will use this month
Output: Each skill listed under Skills with its own on/off switch. Uploaded skills are personal to your account, not shared with a team
The Claude API
Upload a skill to your workspace
Use when: you are calling Claude from your own code. The Python SDK's files_from_dir sends every file in the folder with the right path prefix; with curl, send each file as files[] with a filename that starts with the skill folder name
Output: A skill_id (skill_01…) and a latest_version_id. Custom skills are shared across the workspace, so upload once
Reference it in a Messages request
Use when: you also pass the code execution tool, type code_execution_20250825, because skills run inside its container. Up to 20 skills fit in one request
Output: A response in which Claude has read the SKILL.md, asked for or used the inputs, and run any script the skill names
Know what the container allows
Use when: the API container has no network access and cannot install packages
Output: Every script in this pack imports only the Python standard library, so all of them run there unchanged
Read a skill before you install it
Menu labels on Claude.ai have moved between Features, Capabilities and Customize over time. INSTALL.md inside the pack carries the current path and the one-line zip loop; if the labels have changed again, the skill upload is wherever code execution is turned on.
Where the skills stop
The point is not to have Claude run the company. It is to stop spending founder hours rebuilding the same thinking every time a different part of the company needs you. That line has consequences for what the skills will and will not do.
- They do not run the company. Every skill produces a document, a table or a plan, and a person decides what to do with it
- They need your real numbers. A skill that is missing an input asks for it or states the default it used. A runway model built on a guessed burn is worse than no model, because it looks like one
- They are frameworks, not professional advice. contract-review-checklist tells you what to ask a lawyer and is not the lawyer; round-planner explains how a SAFE converts and is not your counsel or your accountant
- The benchmarks inside them are hints, not targets. A funnel conversion quoted from a survey of other companies tells you where to look, not what to aim for
- They are text. Edit them. A skill rewritten with your own disqualifiers, your own tone rules and your own numbers is worth more than the general one
A skill is worth having when it makes the next version of a decision cheaper than the last one, and worth deleting when it stops
The alternative to installing a pack like this is not doing nothing; it is doing the same thinking by hand every time, and that has a cost of its own.
Rebuild the method each time
The same thinking, from scratch, on a different day
- The pricing survey gets structured differently every quarter, so the results do not compare
- The decision memo is written in your head and re-decided next week
- The discovery call misses the budget question because it was not on a list
Install the method once
The framework is there when the question arrives
- Claude asks for the inputs the method needs and refuses to invent them
- The output has the same shape every time, so this month compares with last month
- The related-skills list says what comes next, so one question does not become five
What this pack is not
Download the pack
42 skills, 25 categories, a README with the full index and INSTALL.md for all three surfaces. MIT licensed by Tau AI.
Sources and provenance
The collection began with Roland Charpentier's public post about building 42 practical Claude skills for the work a solo founder has to carry across a company. The pack published here is an original Tau implementation of that brief, not a copy of a private or gated pack.
- The source LinkedIn post by Roland Charpentier
- The open Agent Skills format specification
- Agent Skills authoring practices
- Intercom's original RICE method
- Y Combinator's current SAFE documents and user guide
Framework names identify their origin. The skill instructions and helper scripts are original adaptations, MIT licensed, and the pack's SOURCES.md records the research trail for maintainers.
Upload individual skills to Claude
Choose a skill below and download its ZIP. In Claude, open Customize → Skills, choose Create skill and then Upload a skill. Select the individual ZIP and enable it. Keep the complete collection for reading or local installation.
Open Claude Skills · Official upload instructions
Each download contains one skill folder with its supporting files. Upload happens in your Claude account; this page does not install anything automatically. Availability follows your account and workspace settings.
Activation funnel audit
Churn save playbook
Cohort retention analysis
Cold outreach sequence
Competitor teardown
Content calendar
Contract review checklist
Contractor brief
Customer interview guide
Decision memo
Discovery call script
Feature prioritisation
Founder energy audit
Framework picker
Growth experiment backlog
Gtm launch plan
Hiring scorecard
Icp definition
Investor update
Landing page copy
Lean canvas
Market sizing
Metrics dashboard spec
Okr writer
One page strategy
Partnership proposal
Positioning statement
Prd writer
Premortem
Pricing research
Pricing tiers
Project plan
Research synthesis
Root cause analysis
Round planner
Runway model
Sales battlecard
Sop writer
Support macros
Unit economics
Use case library
Weekly founder review
Andrew Persh
Founder, Tau AI