realkimbarrett/advertising-skills: a SKILL.md library for direct-response ad work
Advertising Skills for Open Claw, Claude Code & AI agents. Direct response, paid ads, funnels, and copy systems.
At a glance
- What is it?
- This repository packages paid-acquisition and copywriting thinking as Agent Skills rather than prompts. The README is explicit about the mechanism and the target user, and equally explicit that no benchmark, release or test evidence is published alongside it.
- Who is it for?
- Adopt this if you already run paid traffic and want an agent to hold structure across avatar, offer, awareness, mechanism, angles and funnel steps instead of producing one-off copy. Do not adopt it if you need a maintained platform integration, a versioned release, or a library with published evaluation data, because none of those appear in the supplied material.
- Can I use it commercially?
- Not without permission. GitHub finds no licence file in the repository, and without a licence all rights are reserved by default: you may read the code but not reuse it. Check the README, or ask the authors, before using it.
- Is it still maintained?
- Yes. The repository last received commits 174 days ago.
- What is it written in?
- GitHub does not report a main language for this repository.
Answers come from the project's GitHub data, last synced on September 15, 2026, and from our analysis. They are not legal advice.
DEEP OPEN-SOURCE ANALYSIS
The problem is agent output that reads the same every time
The README opens with a before-and-after pair. The before line is "Scale your business with proven strategies"; the after line is "You're not bad at ads. You've just never been shown how to structure them properly." The claim being made is not that AI writes badly. It is that AI marketing output is "Generic. Polished. Forgettable." and that the cause is missing structure rather than missing vocabulary.
The intended user is named plainly: media buyers, founders running paid ads, agencies wanting consistency across teams, and operators building AI-driven marketing systems. The README also states what the repository is not, listing generic content marketing templates, "write me 10 captions" prompts, and surface-level frameworks as out of scope. That is a narrow audience by design. Someone who wants a caption generator will find the skill set heavier than the task.
Skills are folders with a SKILL.md, not prompt strings
The layout is one skill per folder at skills/<category>/<skill-name>/SKILL.md. According to the README, each file declares when to use the skill, what inputs it needs, how it thinks, what it outputs, and what it should not do. That last item is the part most prompt libraries omit, and it is the part that constrains an agent from drifting into generic advice.
The README splits skills into five groups. Foundations covers avatar-extraction, voice-of-customer-miner, offer-extraction, proof-harvester and positioning-core. Copy Chief holds the Eugene Schwartz layer: schwartz-awareness-mapper, market-sophistication-analyzer, dominant-desire-extractor, mechanism-builder, headline-matrix, objection-crusher and close-architect. Operator OS holds execution skills such as scroll-stopping-creative, ad-angle-multiplier, conversion-path-builder, landing-page-architect, email-sequence-architect, performance-diagnosis and testing-roadmap. QA holds generic-language-killer, claim-checker, consistency-auditor and final-polish. Orchestrators are pre-built multi-skill flows: full-funnel-campaign-orchestrator, booked-call-campaign-orchestrator and campaign-diagnosis-orchestrator.
The README frames the split as "The Copy Chief Brain" plus "The Operator Brain", and argues most systems carry one and not both. That is the design thesis, and it is testable by reading the folders rather than by trusting the description.
Chaining is the actual data flow
There is no runtime, no server and no API in the supplied material. The mechanism is sequential: the agent reads one SKILL.md, produces an artifact, and that artifact becomes the input for the next skill. The README gives a booked-call campaign chain in order: avatar-extraction, offer-extraction, schwartz-awareness-mapper, mechanism-builder, ad-angle-multiplier, scroll-stopping-creative, conversion-path-builder, objection-crusher, generic-language-killer.
Read that order and the argument becomes visible. Audience definition precedes offer extraction. Awareness mapping precedes mechanism construction. Creative production happens after angles exist, not before. Quality control runs at the end, on the finished messaging rather than on a draft. The README states the chain produces clear positioning, strong angles, ads, a clean funnel and conversion-ready messaging, and points to /examples/booked-call-campaign-ai-ads.md for a worked output.
Two things follow from this. First, the library is only as good as the handoffs: if avatar-extraction returns something thin, every downstream skill inherits the weakness, and nothing in the README describes validation between steps. Second, the chain is long. Nine skills for one campaign means nine rounds of agent work, which is a real cost in tokens and in the operator's attention.
Getting it running: the install path is the whole setup
Installation is a directory placement. The README says to install into .agents/skills/ and states compatibility with Claude Code, Cursor, Windsurf, Codex, custom agent frameworks and Open Claw. The repository description repeats the same targets. No package manager, installer script or CLI command appears in the material, so there is nothing to run beyond placing the skills folder where your agent looks for it.
That means the first practical check is path resolution, not configuration. If your agent reads skills from a different directory, or expects a manifest file that this repository does not describe, the skills will sit on disk unused. The README does not document a config key, an environment variable or a registration step, so treat the folder path as the integration surface and verify it against your agent's own skill-loading rules before assuming anything is wired up.
The second practical check is the chained filenames. The README names skills in prose and in the example flow, and those names map to paths such as skills/copy-chief/schwartz-awareness-mapper/SKILL.md. Confirming that each named skill exists in the tree, and that the orchestrator files reference the same names, is a five-minute read that tells you whether the library is internally consistent.
Where this library will let you down
The README is a positioning document, not documentation. It gives no worked example inline, no input schema for any skill, and no statement about what a skill returns beyond the phrase "what it outputs". The only artifact it points to is /examples/booked-call-campaign-ai-ads.md, and the supplied material does not include its contents. Until you read that file, the output format is unverified.
There is also no release history in the supplied material. No tags, no changelog, no versioning scheme. The last push date is recorded, but nothing tells you whether the skill set is stable or mid-rewrite. The roadmap lists expanded Copy Chief material, advanced funnel orchestration, creative testing systems and platform-specific layers for Meta, TikTok and YouTube, which reads as work still to come rather than shipped capability. The README lists paid acquisition across Meta, TikTok and YouTube under what the library covers, and the roadmap lists platform-specific layers as future work. Those two statements sit awkwardly together, and the difference matters if you were expecting platform-level handling today.
Finally, the copy is written in the register of an ad. Sentences like "If you use this properly, it won't just improve your ads. It will change how you think about them." tell you about the author's sales instincts and nothing about the skills. That is not disqualifying, but it means the README cannot be used as evidence of quality. The evidence is in the SKILL.md files.
How it differs from a prompt pack or a marketing template library
The obvious alternative is a collection of marketing prompts: a document of reusable instructions you paste into a chat window. The difference is scope and state. A prompt pack gives the model a task; this gives the model a role, a set of declared inputs, a stated output and an explicit list of things it should not do, and it expects artifacts to pass between skills. A prompt pack has no equivalent of the QA group, and no equivalent of generic-language-killer running as a separate pass over finished copy.
The second alternative is a marketing template library, the kind that ships funnel diagrams, email skeletons and ad formats. Those give you the artifact directly. This gives you the reasoning that produces the artifact, which is slower and depends on the agent doing the reasoning well. If your bottleneck is having nothing to start from, templates win. If your bottleneck is that everything you produce sounds like everything else, the structure-first approach is the one aimed at your problem.
The README's own framing is the cleanest statement of the difference: it says most AI marketing libraries optimise for output, and this one optimises for thinking. Whether that holds depends on the SKILL.md contents, which the README does not reproduce.
Licence, maintenance and what adoption actually costs
The README states the licence is MIT, and the author is listed as Kim Barrett. MIT is permissive, so reuse inside commercial client work is broadly permitted, but the supplied material does not include a LICENSE file, a copyright line or a year. Confirm the file exists in the repository before you rely on the grant. Nothing here is legal advice; if you are embedding the skills in a product you sell, have someone check the actual licence text rather than the README line.
Maintenance cost has two parts. The first is upstream: with no releases and no changelog, updates arrive as commits to main, and you either track the branch or pin a commit hash. Pinning is the safer default for client work, because a changed SKILL.md can shift the output of every downstream skill in a chain. The second part is yours: if you edit a skill to match your own offer or voice, you now own a fork, and pulling upstream changes means reconciling those edits. The README gives no guidance on extending skills, so plan on reading the existing files before writing new ones.
Against that, the marginal cost of trying it is low. Copy the skills folder into .agents/skills/, run one chain end to end, and read the output against the promises in the README. The library either produces messaging you would not have written yourself or it does not.
Editorial conclusion
Adopt this if you already run paid traffic and want an agent to hold structure across avatar, offer, awareness, mechanism, angles and funnel steps instead of producing one-off copy. Do not adopt it if you need a maintained platform integration, a versioned release, or a library with published evaluation data, because none of those appear in the supplied material. Before wiring it into a client workflow, open skills/qa/claim-checker/SKILL.md and skills/copy-chief/schwartz-awareness-mapper/SKILL.md, confirm the inputs each one expects, and check that your agent actually resolves .agents/skills/ at the path you install to.
Community notes