lenny-skills counts 86 skills in its description and 76 in its readme

The description says 86 and the readme says 76

The repository's own summary line says eighty-six product management skills.

The readme headline says seventy-six product management and engineering skills.

Both are in the metadata a reader sees, one above the other, and they are not reconciled anywhere in the document.

There is a plausible explanation. The readme's second release restructured the collection into a product operating model, grouping skills into process areas and adding vertical playbooks and a career track. A regrouping can change a headline count without changing how many folders exist, particularly if the two numbers count different things, such as folders against published skills, or episode-derived skills against all of them. The document does not say.

The consequence is small and annoying in the same way. The count is the first thing anyone reading decides whether to download, and there is no way to settle it from the document. Counting the folders would settle it, but that means cloning first.

Everything else numeric in the readme is specific and checkable: the source corpus is described as five hundred and ninety-seven episodes and posts, the insight count as four thousand and nineteen, and the newsletter addition as three hundred and forty-nine posts. So the precision is there everywhere except on the one number that appears in the summary metadata above the readme.

The readme's own badge points at a branch you do not get by default

Near the top of the document, the badge links to the skills directory on a branch whose name is a version number rather than the word main.

The default branch of the repository is main. A clone gets you main.

So the readme's primary navigation link sends you to a different tree from the one the installation instructions operate on, and the installation instructions themselves begin with a plain clone of the default branch.

That split is easy to miss because both are presented as normal. There is no note saying the badge is showing a preview, no note saying the default branch lags, and no note at all about which one the seventy-six or eighty-six count refers to.

It matters for anyone evaluating the collection without installing it, which is the audience the linked site is aimed at. If you follow the badge you are reading a branch that may have a different skill set than the one you would install, and the readme gives you no way to tell which is which.

The related detail is the versioning. There are two releases. The first was tagged in late January and described as an initial release. The second was tagged in mid-July, on the same day as the last push to the repository, and named as a second version of a database rather than of a package. A skills repository has no version to speak of except its content, so a version tag on the database rather than the folders is a deliberate distinction that the readme does not make explicit.

Source counts per skill run from 6 to 47, and that is the only quality signal

Every skill in the catalogue tables has three columns: a name, a sentence about what it helps you do, and a number.

The number is how many sources the skill draws on. It is the only indication in the table of how much material is behind each skill.

The spread is wide. In the strategy group the counts run from 15 to 47, with the strategy definition skill at 47 the highest visible in the document and the north star metrics skill at 15 the lowest in its group. In planning, roadmap prioritisation has 43 while planning cadence has 13. In discovery, product taste has 36 and product experimentation has 9, with customer interviews at 11. In building and shipping, the product stack skill has 6, which is the lowest number anywhere in the visible tables.

So a reader choosing between two skills has one quantitative signal and it is the depth of the underlying material, not the quality of the writing.

That is more useful than nothing, and it is more useful than an undifferentiated list. It also means the collection is uneven by construction, which is normal for something distilled from an archive rather than written to a template.

The sentence in the second column is the other signal, and it is written in a consistent voice. Every one of them is an imperative about a capability, which makes the table read as a catalogue of outcomes rather than of topics.

Portability is claimed in a sentence, and one directory is documented

The readme explains what a skill is in two sentences. They are markdown files that give an agent specialised knowledge and workflows, and you add them to your project so that the agent applies the right frameworks when the task matches.

It names two kinds of consumer in that sentence. Claude Code, and any agent that reads the skill file format.

Then the installation section gives exactly one destination, a dot-prefixed directory belonging to one specific tool.

git clone https://github.com/RefoundAI/lenny-skills.git
cp -R lenny-skills/skills/writing-prds .claude/skills/

So the format is portable and the documented path is not. An agent that reads the same file needs the folder somewhere else, and the readme does not say where, does not say whether a settings file has to point at it, and does not mention a project-level versus user-level distinction.

The alternative offered is not a second path. It is a link to the catalogue site, which has per-skill download links. That solves getting one file rather than the repository, and it still leaves the question of where to put it.

The copy command does reveal one convention worth keeping. You copy a single skill folder, not the whole directory, and the folder name is the kebab-case name used everywhere else in the tables.

Every skill ships a references file, which is where the templates live

The second release added something that is easy to miss in a list of seventy-six rows, because it is a file path rather than a skill.

Each skill carries a references file for its artifacts, and that is where the named frameworks, checklists, and templates live. The stated principle is that these are the things the guests actually used, not ones written afterwards for the skill.

That distinction changes what a skill is. Without it, a skill folder is a prompt. With it, a skill folder is a prompt plus the source material it is summarising, and the two can be checked against each other.

The same paragraph is where the verification claim sits: every quote is checked verbatim against the source transcript or post, and the readme adds that there is no paraphrase drift.

What a quote claim gives you and what it does not is worth being precise about. It says the text of a quotation matches the source, which rules out invented quotes and misquotation. It does not say where any individual quote lives, and the sources column is a count rather than a list, so a reader cannot check the attribution of one line from the repository alone.

The corpus numbers go with it: five hundred and ninety-seven episodes and posts, four thousand and nineteen sourced insights, and for the newsletter portion, three hundred and forty-nine posts including a recurring series on how specific companies build product.

The first release was audio only, and the second quietly doubled the corpus

The two versions differ in a way that changes what the skills are made of rather than just adding to it.

The first release was podcast only. The second adds newsletter content, described as three hundred and forty-nine posts of frameworks and templates, including a named recurring series.

So the second version is not the first with more episodes. It is a second kind of source. A podcast episode is a conversation, and what comes out of it is framing and opinion. A newsletter post is usually an argument with a structure, and what comes out of it is a framework with steps.

That is consistent with where the templates land. The second release's headline addition is per-skill templates and frameworks, and the newsletter is where frameworks come from.

The other structural change is the regrouping. The first version was presumably a flat list of topics, and the second organises skills into the process areas a product organisation actually runs, with strategy, planning, discovery, building, launch, growth, team, and an operating cadence group, plus vertical playbooks and a career track.

Which means the taxonomy is the product. The skill count is a byproduct of it, and two different numbers for the count in one repository is what a taxonomy in transition looks like.

Five entries at the root, one of them the entire project

The repository has five things in it: a gitignore, a contributing guide, a licence, the readme, and a directory called skills.

There is no source directory, no build configuration, no test suite, no continuous integration directory, and no scripts. The primary language is reported as unknown, which is the honest answer for a repository of markdown.

That is worth stating plainly because it tells you what kind of artefact this is. It is a data set with a readme, published under a permissive licence, and nothing about it executes.

The consequence for a reader is that there is no way to check the collection's integrity beyond reading it. No test asserts that a skill file has the front matter an agent needs, no lint passes over the markdown, and nothing verifies the quote claim. The verification described in the readme is a manual editorial process performed by the curator.

The contributing guide is the only document about how to add to it, and the catalogue site is where the per-skill downloads live, so the repository is the bulk distribution and the site is the browsable front end.

Curation is credited to a company rather than an individual, which is worth noting for a collection whose selling point is that each quote was checked against a transcript. There is one curator, and the verification standard is theirs.

Editorial conclusion

lenny-skills fits someone who wants written product practice available to an agent without subscribing to anything, and who is willing to read a skill before adopting it. Four things to check first. How many skills you actually get, since two different counts appear and the catalogue is cut off partway through. Which skills are worth your time, since the source counts per skill vary by roughly eight times and that column is the only quality signal in the table. Whether the install path suits your agent, because portability is claimed in a sentence and only one vendor's directory is documented. And whether the quoted material is what you need it to be, since every quote is claimed verbatim against a transcript rather than linked to a citation you can inspect.

Frequently asked questions

Official sources

1. Issues
2. License: MIT
3. README
4. RefoundAI/lenny-skills on GitHub
5. Releases