A skill library that runs my marketing work
AI gets you a fast first draft in a voice that isn't yours, against a strategy you haven't defined. I wanted the judgment encoded once, in files, instead of re-explained in every prompt.
- Claude
- Skill files
- Orchestration
Outcome
Seven skills covering strategy, production, distribution, and one family cookbook. One orchestrator that chains six of them from a single input.
What's in it
- ai-search-content-prioritizationSkill
Scores content ideas across six dimensions (click resilience, citation potential, brand mention potential, business value, proprietary advantage, expected effort) and returns a write-or-skip verdict.
Includes 13 content types worth prioritizing and 8 to deprioritize, mapped against real proof points rather than generic categories. Runs before anything reaches an outline.
Built onContent Prioritization in an AI Search Era by Aleyda SolisI kept her six scoring dimensions and rebuilt everything downstream: my own content-type taxonomy, a two-mode workflow for brainstorming versus validating, and scoring calibrated against my actual portfolio.
- linkedin-thought-leadershipSkill
Drafts LinkedIn posts in my voice across five formats, with a template library and a hard anti-pattern list.
The anti-patterns override the templates. Where a borrowed hook template collapses into a tidy three-beat arc, the skill flags it inline and rewrites the construction.
- messaging-depthSkill
Pushes surface-level B2B benefits through 3-5 rounds of 'so what?' until the messaging reaches emotional stakes and defensible positioning.
Two input modes. Give it pain points directly, or give it a URL and it analyzes the existing sitemap to extract what the current messaging is actually claiming.
- newsletter-skillSkill
Writes newsletter editions on a fixed five-part structure with per-section word budgets, plus a lighter weekly roundup mode.
500-800 words for a standard edition, 200-300 for a roundup. The word budgets are in the skill because that constraint is what keeps an edition from becoming a blog post in an email.
- recipe-converterSkill
Converts recipes from a URL, pasted text, or a photo of a handwritten card into structured markdown for the family cookbook site.
Not marketing work, which is the point. The same pattern that encodes editorial judgment handles ingredient grouping and course classification, and it reads a photo of a handwritten card as readily as a URL.
- viral-lead-magnet-playbookSkill
Designs a full LinkedIn lead magnet campaign end to end: freebie concept, title, launch post, comment-to-get distribution stack, and the nurture sequence after download.
Ten stages with a build-quality bar and a failure-mode checklist. Names the archetype it's using and why, so the campaign has a stated theory rather than a vibe.
Built ona workshop framework and lead magnet typology by Maria Ledentsova and Alex HormoziI took Ledentsova's 1 Person / 1 Problem / 1 Solution formula and Hormozi's lead magnet types, then rebuilt it for B2B SaaS and agency work: a 3-type gate on the solution row, an alternating virality-versus-expertise operating rule, and a distribution stack matched to tools I actually run.
- repurpose-allAgent
Orchestrates the other skills. Ingests a source once, confirms which formats to run, then dispatches six format skills in sequence against the same context and returns a summary table.
This is the piece that makes it a library rather than a folder. LinkedIn post, X thread, newsletter, video script, carousel, and SEO outline from one input, with shared voice rules applied across all six.
The problem
Every AI writing tool converges on the same voice, and every prompt starts from zero. You can fix a draft by hand. You can’t fix a workflow by hand, and you can’t paste your strategy into a chat window forty times a week and call that a system.
The usual answer is a doc full of rules. That fails for a boring reason: a doc is something you have to remember to open, and nobody opens it on draft nine.
What I built
Seven skills, with an orchestrator that chains six of them from a single source input.
The split matters. Strategy runs first and gates what gets made at all. Format rules live in the skill that owns the format. When I learn something about how newsletters specifically fail, that stays in the newsletter skill instead of leaking into everything else.
How it works
A piece of work moves through the library in a fixed order. Strategy first:
ai-search-content-prioritization scores the idea and returns write or skip,
so nothing gets produced just because it was suggested. Positioning next, where
messaging-depth interrogates the claim until it says something defensible.
Then production, where repurpose-all takes the source once and dispatches
LinkedIn, X, newsletter, video, carousel, and SEO outline against the same
context.
The orchestrator is the part I’d point at. Six formats from one input, with the same voice rules applied across all six, is the difference between a folder of prompts and something that compounds.
On borrowed frameworks
Two of these started as someone else’s work, and both are credited above with a note on what I changed.
I’m providing the context to ensure a) the author(s) who influenced the skill get proper credit, and b) to demonstrate that refining and reshaping skills to work for a specific workflow is part of the iteration and growth process. After seeing the original skill and/or best practices that influenced the final version I now use, I built everything around them: the content-type taxonomy, the two-mode workflow, the scoring calibration, the B2B adaptations, the gating logic, and the internal distribution stack.
Taking a good framework and adapting it is how the discipline works. Passing it off as invented is not.
What changed
Drafts arrive already scored and already in my voice. The editing I do now is about argument and evidence rather than deleting the word “delve” for the four hundredth time.
The compounding is the part that matters. Every time a pattern fails, it becomes a rule inside the skill that owns it, and the next run starts from there. The library gets better between projects rather than during them.
If you want this kind of workflow running inside your team, that's the work I do.