claude-skills-founder is a Claude Code plugin for startup strategy, not for code

The positioning is the whole product

The tagline is startup strategy from your terminal, not another dev tool, and the README makes the contrast explicit.

The note at the top says that most Claude Code skill packs are developer tools, covering code review, git workflows and security audits, while this one is for the founder sitting in a repo at 2am trying to figure out if their idea is worth building.

That sentence tells you the audience precisely. Someone with a repository open, a half-formed idea and no time, who wants the questions asked in the right order rather than another linting tool.

The provenance is stated as well. The skills were built at Emotix while running the authors' own startup, and they encode the frameworks, questions and patterns the authors wished they had on day one. That is a different claim from a list of prompt templates, and it explains why the skills are structured as workflows rather than as single prompts.

The project is Shell as its primary language, MIT licensed, with the skills directory alongside commands/, shared/, evals/, examples/ and a .claude-plugin/ directory that carries the plugin manifest. The last push was on 2026-10-17 and the recent tags are v2.1.4, v2.1.1 and v2.1.0, all from 2026-10-07, so the versioning is active and the patch numbers are moving quickly.

Thirteen skills in three groups

The skill catalogue is organised into three groups, and the grouping tells you how the author thinks about the order of work.

Validation and research holds five. `/validate-idea` stress-tests with a 7-dimension scorecard. `/product-brief` produces a structured brief from one sentence. `/competitor-matrix` gives a feature comparison and positioning gaps. `/persona-gen` creates 3 personas with a priority matrix. And `/user-interviews` produces a Mom Test script and an analysis framework.

Strategy and growth holds five more. `/mvp-scope` is ruthless feature triage. `/pricing-strategy` covers tiers, unit economics and psychology. `/go-to-market` runs from pre-launch to a 90-day growth plan. `/landing-page` writes conversion copy section by section. And `/email-sequence` produces onboarding emails that are ready to send.

Fundraising and metrics holds the last three. `/pitch-deck` gives a 12-slide outline with per-slide content. `/fundraise-prep` does a readiness assessment, investor targeting and a 12-week timeline. And `/metrics-dashboard` covers the 5 metrics that actually matter at your stage.

Two details in the wording are worth noticing because they are commitments rather than descriptions. The Mom Test reference in `/user-interviews` names a specific method rather than saying user research. The metrics skill claims 5 metrics that matter at your stage, which implies the skill branches on stage, and the stage is something you supply.

Every skill takes plain-language input and returns scorecards, tables and checklists you can act on, and the project makes the point that running a skill with no input makes it ask you for what it needs.

Skills that build on each other through a folder

This is the part of the design that makes the pack more than thirteen separate prompts.

Each skill saves its output to a `founder/` folder in your project and reads what earlier skills saved. The example given is that running `/founder:competitor-matrix` once means `/founder:pricing-strategy` and `/founder:pitch-deck` reuse its sourced competitor prices instead of starting over.

The folder layout shows what lives there:

founder/
├── facts.md               # what you told the skills: stage, metrics, team, the raise
├── validate-idea.md
├── competitor-matrix.md
├── pricing-strategy.md
└── ...

`facts.md` is the important entry. It is where the facts you tell the skills once are recorded, and the tip at the end of the usage section says that facts you give once are saved to founder/facts.md so you do not have to repeat them. That is the mechanism that makes the composition work: the skills agree on a shared state file rather than each keeping its own context.

The folder is yours to keep or not. The README says to commit it to keep a history of how your thinking changed, or add it to `.gitignore` to keep it private. That choice matters because the files contain your metrics, your team and your raise, so the default for a founder running this against a real company is probably the second option, and the fact that both are offered rather than one being the default suggests the author has seen both needs.

A history of how your thinking changed is also a real artefact. A pitch deck that evolves across a fundraising process is something a founder can look back at, and the commit history of that folder is the record.

The one rule is that it does not make things up

The rule about sourcing is the most important thing in the README, and it is worth reading precisely.

Facts about the outside world, such as competitor prices, funding rounds, market sizes and benchmarks, come with a source link. When a skill cannot find one, it says Estimate or Not found instead of guessing. And your own metrics, customer quotes and testimonials are never invented, with the missing ones showing up as placeholders for you to fill in.

That is a two-part rule and the two halves solve different problems. The sourcing requirement addresses the outside world, where a plausible-looking competitor price or market size is exactly the kind of thing a model will produce fluently and wrongly. Saying Estimate or Not found instead gives you something you can act on, since a stated estimate can be sanity-checked against your own knowledge and a wrong one cannot be distinguished from a made-up one.

The placeholder rule addresses your own data, and it is the harder half to enforce, because a model asked to write a landing page will happily produce three customer testimonials. Refusing to invent them and leaving a placeholder is the only version of this that is useful.

The project pairs this with an explicit request for bug reports, framed as if a failure of the rule were a bug rather than an expected limitation. If an install step fails or a skill invents a number, that is a bug and they want to hear about it, with the reasoning given that a broken install is silent otherwise, since people try it once and leave.

Installing as a plugin, and the v1 to v2 migration trap

There are two install routes, and the plugin route is the recommended one.

Inside Claude Code, the plugin route is two commands:

/plugin marketplace add emotixco/claude-skills-founder
/plugin install founder@emotix

After that the skills are available as `/founder:validate-idea`, `/founder:pricing-strategy` and so on, and updates come with `/plugin marketplace update emotix`.

The token cost is stated, which is unusual and useful. The skill descriptions add about 1,100 tokens to each session, and a skill's full instructions load only when you run it. So the pack is designed around the progressive disclosure model, with the descriptions acting as an index that the model reads to decide whether to pull in more.

The clone route is a single git clone into the skills directory:

git clone https://github.com/emotixco/claude-skills-founder.git ~/.claude/skills/founder

Claude Code loads the folder as a plugin on the next session with the same `/founder:` names, and the update command is a git pull against that directory:

git -C ~/.claude/skills/founder pull

The upgrade note is the one to read if you used v1. Version 1 installed flat files into `.claude/commands/founder/`, and you have to delete that folder, or the symlink if you used the symlink option, before installing v2, or you will see every skill twice. The location is either the one in your home directory or the one inside the project where you installed it. The v2 layout is a plugin, which is why the old flat commands would otherwise linger and be discovered alongside the new names.

Running a skill, and the examples you should read first

Usage is a slash command with your context after it, and the examples in the README are concrete.

> /founder:validate-idea An AI tool that analyzes pitch decks and gives investor-perspective feedback

> /founder:competitor-matrix Project management tools for solo founders (not teams)

> /founder:pricing-strategy B2B SaaS for restaurant inventory, targeting independent restaurants with 1-3 locations

> /founder:pitch-deck Pre-seed raise, $500K, AI competitor analysis for founders, 200 beta users, $2K MRR

The last one is the most informative, because it shows the kind of input the skills expect. A stage, a number, a positioning, a user count and a revenue figure. The tip in the usage section says to be as specific as possible and to include your stage, metrics, constraints and what you have already tried.

Run a skill with no input and it asks for what it needs, which is the right behaviour for a tool that has to know your stage before it can advise.

Before installing, the README points you at the examples. The restaurant-inventory directory has unedited output from three skills run in sequence on one idea, including the founder files they wrote, and you are told to read it before installing to see what you get. Unedited is the operative word, and that is what makes it worth reading, because the difference between a good output and a plausible one is mostly visible in what it declines to say.

The evals directory tests the plugin against plain Claude with `claude plugin eval`, and the README is unusually honest about the result: on Opus 5, 8 of 10 cases score the same with the plugin and without it. That is a valuable thing for a project to publish about itself, because it says the skills mostly provide structure and prompts rather than a capability the model lacks.

Editorial conclusion

Adopt claude-skills-founder if you are running a startup from a terminal and want structured prompts for the strategy work that usually has no tooling, since the skills return scorecards, tables and checklists rather than prose. Skip it if you want an advisor rather than a template, because the output quality depends on the model you run it with and the evals are candid about that, and if you need facts about your market that the tool cannot source, since anything without a source link comes back marked as an estimate or not found. Verify first that the Claude Code version you have supports plugins and skills, that you install the v2 layout rather than the v1 flat commands, and that you read the example output before installing, because the examples directory shows unedited output from three skills run in sequence.

Frequently asked questions

Official sources

1. emotixco/claude-skills-founder on GitHub
2. Issues
3. License: MIT
4. README
5. Releases