# startup-skill: four skills, six install paths, one subscription hint

> A repository of markdown agent skills covering startup design, competitor research, positioning and pitching, with six documented ways to install them and a page that recommends a specific paid plan because the skills run multiple research agents. The strongest part is the research skill's own scope limit: social and community sources count only when a person hands them over.

**ferdinandobons/startup-skill** — AI agent skills for startup validation, competitive intelligence, and planning

- Repository: https://github.com/ferdinandobons/startup-skill
- Website: https://startupskill.me
- Stars: 1,159 · Forks: 156
- Language: Unknown
- License: MIT
- Published: 2026-09-10 · Updated: 2026-09-10 · Language: en
- Canonical page: https://hysenlabs.com/projects/ferdinandobons-startup-skill

## Four skills, and the deliverable set is enumerated

The repository is four directories, and each one is described by what it produces rather than what it does.

The design skill covers market research, customer discovery, competitive analysis, brand, product definition, financial projections and validation experiments, organized as eight phases plus a pre-flight check and a customer interview gate, and it claims more than thirty structured deliverables. The competitor skill produces battle cards for every competitor, a pricing landscape, a feature matrix and a strategic report. The positioning skill applies a named consultant's framework and produces a positioning document, a competitive alternatives map, a market category analysis and messaging implications. The pitch skill produces five spoken and written formats plus a slide-by-slide deck outline, with a scoring rubric, question-and-answer preparation and investor roleplay practice.

Two numbers are worth noting because they are the only quantities on the page. The competitor skill profiles five to eight or more competitors across three research waves. And the positioning skill is described as working through five plus one components of the framework, a count the page does not enumerate.

The framing on the front page sets expectations in a sentence: what a ten-thousand-dollar strategy consultant would deliver, and if your idea should die, it will tell you. That is a claim about the output's willingness to be negative, and the validation-first wording in the release history suggests it is a direction the project has been moving in rather than a settled property.

## The same skill has two modes and the trigger is the mode

The usage section does something more interesting than list commands. It gives five sample requests and shows which skill each one routes to, and one of those five routes to the same skill as another in a different mode.

The full request is a question about whether a product idea is worth building, and it runs the design skill's whole process, the pre-flight check, the eight phases, and the customer discovery gate. The short request is the words fast track mode, and it runs a compressed go-or-no-go analysis from the same skill.

So the routing key is not just which skill but which depth, and the depth is expressed in the request. That is a design choice with a cost: two very different amounts of work sit behind one skill name, and a user who asks a short question does not necessarily realize they asked for a fast answer rather than the full one.

The other three routes are unambiguous. Asking who your competitors are in a market, with battle cards and a pricing comparison wanted, goes to the competitor skill. Asking how to position a product against named incumbents goes to the positioning skill. Asking to prepare a raise, with a round size and growth numbers in the sentence, goes to the pitch skill.

Direct invocation is also available, using a namespaced form with the plugin and the skill name joined, so you can bypass the trigger entirely when you know which one you want.

## The plugin has a documented update failure and a four-step remedy

The recommended install is two commands that add a marketplace source and then install a plugin from it, and the plugin has its own name, which is not the repository name.

```bash
claude plugin marketplace add ferdinandobons/startup-skill
claude plugin install startup@startup-skill
```

The update section is where the maintenance story gets honest. There is an update command for the plugin, and then a caveat: if the update does not pick up new changes, refresh the marketplace source and reinstall. The remedy is a four-command sequence that uninstalls the plugin, removes the marketplace source, adds it again, and installs again.

That is a real failure mode in a two-file release process where the marketplace caches what it knows about, and it is documented rather than left to be discovered. A user on the update path who sees stale content now has a documented procedure instead of a guess.

The other three paths behave differently, and the differences are worth knowing. The command-line and SkillKit installs are overwrite operations, so re-running them is the whole update. The submodule needs an explicit remote flag, because a plain submodule update would not see the new commits. And the web app path is manual in both directions: download the packaged files from the releases page again and upload them in settings.

## One install method is four hand-written copy commands

Six install methods are documented for a repository that contains no code, and the contrast between them is instructive.

Two are automated: adding the repository through a command-line installer, and installing through a separate toolkit that the page says also works with Cursor and Copilot. Two are git-native: adding the repository as a submodule into an agent skills directory, and cloning it. One is the plugin pair shown earlier. And one is the web app, where you download packaged skill files from the releases page and upload them in the agent's settings.

Here is the block that covers four of them, and it is worth reading as one unit:

```bash
# CLI Install
npx skills add ferdinandobons/startup-skill

# Clone and copy
git clone https://github.com/ferdinandobons/startup-skill.git
cp -r startup-skill/startup-design .agents/skills/
cp -r startup-skill/startup-competitors .agents/skills/
cp -r startup-skill/startup-positioning .agents/skills/
cp -r startup-skill/startup-pitch .agents/skills/

# Git submodule
git submodule add https://github.com/ferdinandobons/startup-skill.git .agents/startup-skill

# SkillKit (works with Claude Code, Cursor, Copilot, etc.)
npx skillkit install ferdinandobons/startup-skill
```

The clone method is the one that will age. It is written as four separate copy commands, one per skill directory, each landing in the same destination, so the count of skills is encoded in the number of lines in the documentation. Add a fifth skill and this section needs a fifth line, and anyone who followed it in April has four directories in their project that no update command will refresh.

The submodule path avoids that by keeping the whole repository as one reference, which is the cleaner answer for a set of directories that arrive and leave together.

## The competitor skill draws its line at human-supplied social sources

One phrase in the competitor skill's description is the most careful sentence on the page, and it is easy to read past.

The skill builds its output from real reviews, forums, web data, and reviewed social or community source packets when provided. The last clause is a scope limit stated in advance: social and community material counts as evidence when a person supplies it as a packet that gets reviewed, and not otherwise.

That is a defensible line to draw. Search results and published reviews are things an agent can find and cite, while a forum thread or a private community is not, and a skill that quietly invented the second category would be producing exactly the kind of unfalsifiable competitive analysis its users cannot check. Declining to search there and asking for material instead puts the sourcing decision where a human can see it.

The cost is coverage. A community where a product's reception is discussed will not appear in the battle cards unless someone brings it in, and the three research waves are described without saying whether the packet substitutes for a wave or is read alongside them.

## The pitch skill scores the deck with a rubric it wrote

The pitch skill is the one with the most moving parts, and one of them deserves a second look.

Its outputs are five spoken and written formats, a ten-minute pitch, a five-minute pitch, a two-minute pitch, a one-minute elevator version and an email, plus a slide-by-slide deck outline. Alongside those it produces a scoring rubric, question-and-answer preparation, and roleplay practice where the agent plays the investor.

The scoring rubric is the interesting piece, because it is both the tool and the target. A skill that generates a deck and then grades that deck against criteria it chose in the same session is rehearsing the argument rather than testing it, and the number that comes out is a measure of internal consistency, not of investor appeal.

That is still useful. A deck that cannot be summarized in one minute is a real finding, and a set of questions the agent anticipates from the rubric is a better starting point than none. But the score itself should be read as a self-check.

The release history suggests the pitch surface has been changing: one release in June was titled for a validation-first flow and a pitch deck outline, so the deck outline is a recent addition rather than a founding feature.

## The repository is prose, with a plugin manifest and two agent instruction files

The tree tells you what kind of project this is. There is a plugin manifest directory, a workflows directory, two agent instruction files, a changelog, a contributing guide, an assets directory, and the four skill directories. There is no source directory, no test directory, and no examples directory.

The language field is reported as unknown, which is consistent with a repository whose product is markdown and JSON rather than code. That has one practical consequence worth naming: there is nothing to unit test here, so the quality of the output is entirely a property of the instructions and of whatever model is running them.

The release titles are more informative than the version numbers. One mentions deep-reasoning cues for new models, which tells you a release was about getting better behavior out of a newer model rather than adding capability. Another mentions a validation-first flow alongside the pitch deck outline. The most recent release, on the first of July, carries no descriptive title at all, and the last commit on the default branch carries the same date.

So the project is three months past its last commit as of writing, with four skills, an MIT license, and a page that is candid in one place, about token cost, and careful in another, about what counts as evidence.

## Conclusion

startup-skill is a set of instructions rather than software, so the useful question is what the instructions actually do and what they decline to do. They produce a documented deliverable set, they route by intent, and the competitor skill draws its line at human-supplied social sources, which is a more defensible position than most. Three things to know before you spend a session on it. The cost is real and the page says so, recommending a specific paid plan because the work fans out into multiple research agents, with resumption handled by asking the agent to continue rather than by a checkpoint file. The plugin update path has a documented failure that requires uninstalling and reinstalling. And the pitch skill grades the deck with a rubric it also generated, so treat that score as a rehearsal rather than an evaluation.

## FAQ

### Which skills does startup-skill provide?

Four. startup-design covers market research, customer discovery, competitive analysis, brand, product definition, financial projections and validation experiments in eight phases plus a pre-flight check and a customer interview gate. startup-competitors produces battle cards, a pricing landscape, a feature matrix and a strategic report. startup-positioning applies a named consultant's framework. startup-pitch produces five pitch formats plus a deck outline, a scoring rubric, question and answer prep, and roleplay practice.

### How do I install startup-skill in Claude Code?

Add the marketplace source with claude plugin marketplace add ferdinandobons/startup-skill, then install with claude plugin install startup@startup-skill. Other documented paths are a command-line skills installer, a separate toolkit that also works with Cursor and Copilot, a git submodule, cloning the repository and copying the four skill directories into .agents/skills/, or downloading the packaged skill files from the releases page and uploading them in the web app's settings.

### How do I update startup-skill?

The plugin path has an update command, and the page documents its failure mode: if changes do not appear, uninstall the plugin, remove the marketplace source, add it again and reinstall. The command-line and toolkit paths just re-run the install, which overwrites the previous version, and the submodule path uses a remote update flag. The web app path means downloading the latest files and re-uploading them.

### How many tokens does running startup-skill take?

The page warns that these skills run multiple research agents and can consume a large number of tokens, recommends a specific paid plan for the best experience, and says that if a session hits the limit you can ask the agent to resume from where it left off and it will pick up the process. There is no documented checkpoint file, so resumption goes through the conversation.

### What sources does startup-competitors research?

Real reviews, forums, web data, and reviewed social or community source packets when a person provides them. It profiles five to eight or more competitors across three research waves and produces battle cards, a pricing landscape, a feature matrix and a strategic report.

## Sources

- [ferdinandobons/startup-skill on GitHub](https://github.com/ferdinandobons/startup-skill)
- [License: MIT](https://github.com/ferdinandobons/startup-skill/blob/main/LICENSE)
- [Project website](https://startupskill.me)
- [README](https://github.com/ferdinandobons/startup-skill/blob/main/README.md)
- [Releases](https://github.com/ferdinandobons/startup-skill/releases)

---

Hysen Labs editorial analysis, written from the project's own repository and release notes. Cite the canonical page: https://hysenlabs.com/projects/ferdinandobons-startup-skill