Advertising Skills names 26 skills, claims MIT, and gives no install command

The documented layout is two levels deep and the install target is one

The library states its own path convention plainly: each skill lives in its own folder, `skills/<category>/<skill-name>/SKILL.md`. The installation section names one destination and nothing else. It says the library is compatible with any agent using the Agent Skills pattern, to install into `.agents/skills/`, and it works with Claude Code, Cursor, Windsurf, Codex, custom agent frameworks and Open Claw.

Those two statements do not describe the same shape. The shipped tree nests a skill two directories deep under a category, while the install target is a single flat directory. Nothing on the page says whether `foundations/avatar-extraction/` becomes `.agents/skills/avatar-extraction/` or `.agents/skills/foundations/avatar-extraction/`, and the category names are written as display labels with spaces and capitals, which are not usable as directory names as written. So the one step between cloning the repository and an agent finding the skills is unspecified.

There is also no command anywhere on the page. No clone line, no installer script, no package manager entry, no dependency list. The repository root is five entries: `AGENTS.md`, `README.md`, `VERSIONS.md`, `examples/` and `skills/`. Copying a directory by hand is the implied install, and any agent that needs a specific category nesting will only find out at discovery time.

Twenty-six named skills, and the page never says how many

Counting the five category lists gives 26 distinct skill names. Foundations holds five: avatar-extraction, voice-of-customer-miner, offer-extraction, proof-harvester and positioning-core. Copy Chief holds seven, including schwartz-awareness-mapper, market-sophistication-analyzer, dominant-desire-extractor, mechanism-builder, headline-matrix, objection-crusher and close-architect. Operator OS holds seven more, from scroll-stopping-creative through landing-page-architect, email-sequence-architect, performance-diagnosis and testing-roadmap. QA holds four, and Orchestrators holds three.

No count appears anywhere. The library says each skill is focused, opinionated, reusable and designed to work together, and the repository description names the subject areas, but a reader who wants to know the size of the thing has to add the lists up themselves. The worked example uses nine of the 26 in sequence, from avatar-extraction and offer-extraction through to generic-language-killer at the end.

That chain overlaps with a skill. The example is titled as building a booked-call campaign, and Orchestrators contains booked-call-campaign-orchestrator, which is described as a pre-built multi-skill flow. So the same outcome can be reached by chaining nine skills by hand or by invoking one orchestrator, and the page does not say which route it prefers or whether the orchestrator is exactly that nine-step chain.

One structural note from the chain itself: the last step is a QA skill, not a creative one, so filtering is positioned as the terminal stage of the pipeline rather than a parallel review.

Meta, TikTok and YouTube appear as current coverage and as roadmap items

The What This Is section says the library is built around paid acquisition across Meta, TikTok and YouTube, plus creative testing and scaling. Five sections later, the Roadmap lists platform-specific skill layers for Meta, TikTok and YouTube, along with creative testing systems and advanced funnel orchestration, as work still to come.

So the same three platforms are sold as covered and scheduled. Either the front-page bullets describe the intended scope rather than the current files, or the roadmap is stale. There is no way to tell from the page, and it matters because platform-specific layers are where the operational detail lives: a Meta layer and a TikTok layer are not the same set of formats, placements or reporting conventions, and a library that names three platforms without shipping platform layers is giving you a general method and a promise.

The same tension shows up in the Copy Chief category. Expanded Copy Chief library, covering mechanisms, big ideas and proof systems, is the first roadmap item, while mechanisms and proof are already listed as current skill names. Read as a whole, the front page describes a finished system and the roadmap describes how much of it is still being written.

The License section says MIT and the root has no license file

The page ends with a License heading and one word under it: MIT. The repository root does not contain a LICENSE file. The five entries are `AGENTS.md`, `README.md`, `VERSIONS.md`, `examples/` and `skills/`, so there is nothing named LICENSE, nothing named COPYING and nothing in a subdirectory at the top level that would carry one either.

That combination, a stated license in the prose and no license text in the tree, is the kind of gap worth resolving before you copy 26 files into your own agent configuration. A single line of text in a README is not the same artifact as a license file, and the difference is what a downstream reader can point at. The repository's own metadata also records no recognized license identifier and no primary language, which is consistent with a repository that is mostly markdown and has no license file for tooling to detect.

For a library whose stated purpose is reuse, that is the one omission to fix first, and it is a small fix.

Three root documents, two of them never mentioned, and no releases

The repository is three markdown files and two directories. The README body references one of them: it points at `/examples/booked-call-campaign-ai-ads.md` under Example Output, and that file does exist. The second example, `examples/why-this-library-exists.md`, is never referenced from the page.

`AGENTS.md` and `VERSIONS.md` are also unreferenced. The convention of shipping an `AGENTS.md` is common in agent tooling repositories, and a versions file usually tracks release history, but neither is linked or described, so a reader cannot tell from the front page whether the versioning is per-skill, per-category or repository-wide. The recent release list is empty, so there is nothing published to compare against either.

The declared homepage, `advertisingskills.io`, is likewise not linked anywhere in the page text. The repository was last pushed on 2026-03-26. That is the state to plan around: a skills library with no releases, no version history anyone is told about, and a single commit date to judge freshness by.

None of this makes the skills less usable. It does mean the file you would consult first, the versions document, is the one the front page never points you to.

Every skill's contract includes a line about what it should not do

The skill contract has five parts: when to use it, what inputs it needs, how it thinks, what it outputs, and what it should not do. Four are the usual shape for a reusable instruction file. The fifth is the one worth noticing, because a prohibition travels with the skill rather than living in a separate policy document that an agent might not load.

The QA category is where that shows up most concretely. It holds generic-language-killer, claim-checker, consistency-auditor and final-polish, described as what makes the output actually good. A claim-checker and a consistency-auditor are both verification steps with an implicit veto: their output is a judgement that something should not ship. The category description for these four is a quality bar rather than a creative role, which is a different kind of skill from ad-angle-multiplier or headline-matrix.

The library also states that skills are designed to be chained, and the example flow makes that concrete by ordering nine of them. Designation matters when you build a pipeline: the filter steps sit at the end of that chain, after creative and conversion work, so the last thing to run decides what reaches the customer.

One named copywriter underwrites the theory, and Open Claw is never defined

The method is attributed. The library is built around Eugene Schwartz-style copywriting, one of its two mental models is labelled The Copy Chief Brain with awareness levels attributed to Schwartz, and two skills carry the name directly, schwartz-awareness-mapper and market-sophistication-analyzer, alongside dominant-desire-extractor and mechanism-builder, which are also concepts from that body of work. The other model is called The Operator Brain and covers ad creative and angles, funnel paths and conversion logic, qualification and filtering, and testing and performance diagnosis.

The page's claim is that most systems have one of these two and this has both. Read as a dependency rather than a claim, it means a large share of the library encodes one author's framework, and a reader who does not share it inherits its vocabulary through the skill names alone. The roadmap's first item, an expanded Copy Chief library covering mechanisms, big ideas and proof systems, is an admission that the flagship category is the least built.

Then there is the audience line. The repository description leads with Open Claw ahead of Claude Code, and Open Claw appears last in the compatibility list alongside Claude Code, Cursor, Windsurf, Codex and custom agent frameworks. Nothing on the page says what Open Claw is or why it is named first, which for a repository whose only interface is a folder convention is a strange omission.

Editorial conclusion

Use this if you want a written direct-response method you can read and edit, and you already know how Agent Skills get discovered, because the library is explicit about categories and vague about installation. Verify three things first. The on-disk layout is `skills/<category>/<skill-name>/SKILL.md` while the install target is `.agents/skills/`, and nothing says whether the category level survives the copy, so check that before pointing an agent at the directory. The License section says MIT but no license file is in the root, so treat the terms as unstated until you find one. And note the last push is dated 2026-03-26 with no releases published, which means the platform-specific layers for Meta, TikTok and YouTube promised on the front page are the same items the roadmap lists as future work. If you want a working command rather than a folder convention, this repository does not have one.

Frequently asked questions

Official sources

1. Issues
2. Project website
3. README
4. realkimbarrett/advertising-skills on GitHub