ai-business-skills: 144 bilingual marketing skills for Claude Code and other agents

The problem is prompt sprawl, not missing prompts

Most people doing marketing with an AI agent already have prompts. What they do not have is a consistent way to decide which prompt applies, what context it needs, and what a finished output looks like. ai-business-skills addresses that by shipping the prompts as files with a routing layer on top. The README describes the intent directly: it turns one marketing request into a routed workflow, context first, then the right skill cluster, then an agent or workflow when the job spans multiple steps. The intended user is narrow and identifiable: a Vietnamese SME or agency, a global SMB, a founder or coach building a personal brand, or someone running dropshipping. The repository is bilingual by design, with a skills/vi/ cluster and a skills/en/ cluster, and the marketing content carries regional benchmarks for the US, EU, SEA, LATAM and Vietnam across 2025 and 2026. This is a library for people who already have an agent open, not for people looking for a tool to install and click.

How routing works: context, cluster, agent, workflow

The system map in the README shows the data flow. A user request enters a foundation context layer stored as .agents/*.md files, which the README labels as product, audience, positioning, proof and objections. A mode router then sends the request to either the VN cluster or the Global cluster. From there, 6 universal agents handle role-based routing across strategy, content, design, performance, channels and personal brand. When a job spans multiple steps, one of the 19 workflows takes over, and the output is a plan, copy, brief, report, SOP or dashboard. The README also gives a second decision tree for entry points, mapping strategy questions to skills like 00 plan, 10 reverse KPI and 31 offer design, and content questions to 01 calendar, 04 video script, 05 ad copy and 30 design master. The interesting design choice is the context layer sitting in front of everything. A skill that writes ad copy is cheap to produce. A shared context file that every skill reads is the part that keeps output from drifting between sessions, and it is why the repository is organized as a system rather than a folder of prompts.

Installing it: two commands and a directory

The quick start is short. On macOS or Linux, clone the repository, change into it, and run bash install.sh --global, which the README says places 144 skills into ~/.claude/skills/marketing/. On Windows the equivalent is .\install.ps1 -Global. PowerShell is the primary language of the repository, which is consistent with the Windows-first install path. The README then suggests a test prompt in Claude Code, asking for a marketing plan for an AI course with a target of 100 students in 30 days, and states that the skill 00-ke-hoach-mkt activates on its own and produces a detailed .md file containing Vietnam 2025-2026 benchmarks, three KPI scenarios, budget allocation, a weekly timeline and a risk matrix. Treat that as the documented example rather than a verified output. The install flag is the only configuration surface shown in the material, and there is no config file, no environment variable and no per-project override documented in what is available here. If you need to pin a subset of skills, the material does not describe how.

The VN cluster is the part with no obvious substitute

Generic marketing prompt collections are abundant and mostly interchangeable. What is harder to find is a library that treats Vietnam as a first-class market rather than a translation target. The README states that the VN cluster contains 74 skills while the Global cluster contains 69, and the version note for v3.6.0 describes 33 Global mirror skills completing parity for 4 roles, following v3.5.0 which added 33 VN skills, a knowledge/ directory and a design-producer agent. So the two clusters are not symmetric. The VN side carries content the Global side does not, and the parity work described in v3.6.0 is explicitly about the four role SOP packs (content, design, performance, leader ops), not about every skill. If you market into Vietnam, that asymmetry works in your favor. If you market only into the US or EU, you are paying the maintenance cost of a bilingual repository to use roughly half of it.

Where it stops: no execution, no scheduling, no measurement

This is a file library for an agent, and it has the failure modes of one. Nothing in the material suggests a scheduler, a dashboard, a data connector or an API client. The output is described as plans, copy, briefs, reports, SOPs and dashboards, which in context means generated documents, not live reporting. If your actual problem is that campaign performance data never reaches the person who writes the next brief, this repository does not close that loop. The other limitation is verification. The README claims Vietnam 2025-2026 benchmarks and three KPI scenarios inside the plan output, but it does not say where those benchmark figures come from, how they were collected, or when they expire. A skill that emits a budget allocation and a KPI scenario is only as good as the numbers behind it, and a dated benchmark set ages. Before you put any of those figures in front of a client, check the source inside the skill file itself.

Compared with a general prompt library, the difference is the routing layer

The obvious alternative is a general-purpose prompt collection, or simply writing your own prompts into a project-level instructions file. The difference in approach is structural. A prompt collection gives you text you paste. ai-business-skills gives you a directory tree with a mode router, a context layer in .agents/*.md, role agents in agents/, multi-step sequences in workflows/ and a knowledge/ folder the README describes as foundation thinking to upload into role-based AI projects. That structure is what makes a request route to 00-ke-hoach-mkt instead of to whichever prompt you happen to remember. It is also what makes the repository heavier to adopt: you inherit someone else's taxonomy, numbering scheme and file layout. A hand-written project instruction file is faster to start and drifts less, but it will not give you 19 pre-built workflows or a VN and Global split. The honest comparison is that this repository trades flexibility for coverage, and coverage is only worth it if you recognize the taxonomy it imposes.

Maintenance, licensing and the version churn to expect

The repository is MIT licensed, which permits commercial use, modification and redistribution provided the copyright notice and permission notice are retained. That is a permissive license, and it is worth noting because the content is marketing material you may want to resell or embed in client work. This is not legal advice, and if you redistribute the skills inside a product you should read the LICENSE file rather than a summary. On maintenance, the release history shows a fast cadence: v3.0.0 in May 2026 as a rebrand with a dual-edition and three-module structure, v3.1.0 the next day adding Design Master bilingual and multi-platform onboarding, v3.2.0 two days later described as Anthropic pattern alignment and marked additive, and the README badge now at v3.6.0 with a note of zero breaking changes. Additive releases are cheap to absorb, but a rebrand at v3.0.0 means earlier installs may have used different paths. If you forked or edited skill files, plan on re-applying those edits at each minor version, because the material does not describe an override directory or a patch mechanism.

Editorial conclusion

Adopt it if you run marketing for a Vietnamese SME, agency or creator business and you already work inside Claude Code or a similar agent, because the VN cluster and the 2025-2026 Vietnam benchmarks are the part of this repository that is hardest to assemble yourself. Skip it if you want a marketing platform with a UI, scheduled execution or analytics, since nothing here runs on its own. Before installing, open skills/vi/ and skills/en/ and confirm that the two clusters cover the same jobs, because the README badge says 144 skills while the system map still labels the clusters 74 and 69, and that gap is the first thing you should resolve.

Official sources

1. License: MIT
2. minhnv0807/ai-business-skills on GitHub
3. Project website
4. README
5. Releases