// the find
aaron-he-zhu/aaron-marketing-skills
69 marketing skills + 5 commands for Claude Code & AI agents across 4 disciplines — SEO/GEO, influencer (IMPACT), paid ads (ROAS) & email (SEND) — on one shared contract. 5 quality frameworks with auditor gates: CORE-EEAT · CITE · C³ · ROAS · SEND.
A library of 69 Claude Code skills and slash commands that turns an AI agent into a marketing operator across SEO/GEO, influencer, paid ads, and email. The whole thing is Markdown files with a thin Bash hook runner and some zero-dependency Python helpers — no framework, no build step. For marketing practitioners who already use Claude Code daily and want structured workflows rather than ad-hoc prompting.
The 'keyless by default' principle is genuinely useful — every skill has a free/own-data path, so you're not gated behind paid API keys to get started. The shared skill contract (seven sections, consistent handoff schema) means skills actually compose instead of being isolated prompts. The quality gate system (CORE-EEAT, ROAS, SEND, etc.) gives structured verdicts with explicit veto conditions rather than vague scores. The HOT/WARM/COLD memory tier with a 80-line/25KB cap is a practical solution to the context bloat problem in multi-step agent workflows.
The benchmark frameworks (CORE-EEAT 80 items, CITE 40 items) are internally defined with no external validation — they're confident-sounding rubrics that the author invented, not industry standards. At 2300 stars for a specialized Claude Code skills library released in 2025, the star count smells inflated relative to the likely actual user base. The 'no pip' constraint means the Python helpers stay simple but also means any meaningful data processing is off the table. The install path is fragile — it depends on `/plugin marketplace add` and `/plugin install` Claude Code commands that aren't stable or widely documented, so the 'just works' install story may break as Claude Code evolves.