bencium-marketplace: 17 skill directories and the names the tables miss

Seventeen skills, seventeen directories, and a name in each list the other lacks

Three numbers describe this marketplace and they agree only on the total. The opening line calls it a Claude Code plugin marketplace with 17 skills and tools. The four category tables below that line hold 6 design rows, 1 marketing row, 6 productivity rows, and 4 development rows, which adds to 17. The root of the repository holds 17 skill directories. The total is therefore right three times, yet the sets behind it are not identical. `eu-ai-act-reviewer/` exists as a directory and appears in no table. `bencium-loop` has a row in the productivity table and no directory here, because that row says the skill installs from its own repository. So the figure a reader counts from the prose is stable while the contents it stands for are not. A reader browsing the tree finds one skill the documentation never names, and a reader installing by name finds one skill the tree never ships.

Four install routes, two of which are commands and none of which pin a version

The install section offers four routes and only two of them are commands. The first is the recommended path through skills.sh:

# Browse all available skills
npx skills add bencium/bencium-marketplace --list

# Install all skills globally
npx skills add bencium/bencium-marketplace -g --all

# Install a specific skill
npx skills add bencium/bencium-marketplace -g --skill typography
npx skills add bencium/bencium-marketplace -g --skill design-audit

The second route runs inside the Claude Code CLI, where the two lines are typed into the session rather than into a shell:

# Add marketplace
/plugin marketplace add bencium/bencium-marketplace

# Install any plugin
/plugin install bencium-controlled-ux-designer@bencium-marketplace

The third route has no command at all. For the Claude Cowork app you copy the `skills/` directory from any plugin into the project's `.claude/skills/` folder. The fourth route claims compatibility with 40+ AI coding tools that support markdown-based skill files, names OpenAI Codex, Gemini CLI, Cursor, and Windsurf among them, and gives another manual copy, this time into whatever prompt or context directory the target tool uses. Every command above names a repository rather than a version, so each one fetches the default branch, which is main. There are no GitHub releases to pin, and the last push on the repository is dated 2026-08-30, which means a global install takes whatever main held at the moment the command ran.

The build loop row in the productivity table points at another repository

One row of the productivity table describes a tool this repository does not contain. `bencium-loop` is listed as a lightweight AI-assisted build loop with a scaffold step, a living roadmap, a plan and spec gate, a verify gate backed by an independent evaluator, and deploy and rollback. The row claims nine commands and no infrastructure, and it claims a memory that survives `/clear`. The closing sentence of that row is the operative one: the skill installs from its own repository, bencium/bencium-loop. Nothing in the root entry list corresponds to it, so `npx skills add bencium/bencium-marketplace -g --all` does not install the loop, and the deploy and rollback pair has no counterpart in the per-plugin layout that the skill structure section defines. Anyone planning around this set has to treat the marketplace and the loop as two separate fetches. The verify gate is the part most likely to matter in practice, since it is the capability that decides what a session is allowed to call finished, and it sits outside this tree.

One productivity entry is a hook and a statusline rather than a skill file

The productivity section labels one of its entries as a hook plus a statusline rather than a SKILL.md skill, and that label changes what the marketplace can install. `emotion-statusline` shows the model's classified emotional state in the statusline and raises a bold-red warning when the desperate state is detected. The row ties that state to reward hacking, described as a stressed model quietly cutting corners while sounding composed. Because the entry is not a skill file, the skills.sh routes that install skills do not reach it, and the Cowork instruction to copy a `skills/` directory does not reach it either. What remains is the standard plugin layout, which places a plugin.json inside a `.claude-plugin/` directory and a SKILL.md inside a `skills/` subdirectory, and neither of those locations describes a statusline. Of the seventeen directories at the root, sixteen are skill plugins and one holds a hook and a statusline. The opening line folds both kinds into a single count without separating them.

The spinner verb list is a root file you merge into settings.json by hand

The spinner verbs take a third shape entirely. `spinner-verbs.json` sits at the root as a bare file, with no `.claude-plugin/plugin.json` above it and no `skills/` directory beside it, so neither marketplace command touches it. The documented step is to copy its contents into `~/.claude/settings.json`, either replacing the default list or appending to it:

{
  "spinnerVerbs": {
    "mode": "replace",
    "verbs": ["..."]
  }
}

The file holds 64 satirical tech-bro verbs that stand in for the default "Thinking..." message, with entries such as "Enshittifying platforms" and "Overcompensating with rocket size". It is the only asset here that changes the interface rather than adding a capability, and the only one whose install is a two-mode file merge instead of a plugin fetch. The layout section leaves a smaller gap too: it requires a `.claude-plugin/plugin.json` inside every plugin directory, and a `.claude-plugin/` directory also sits at the root, but the README does not say what that root directory declares. GitHub labels the primary language Shell, and the root entry list holds no shell script to account for it.

Two design rows cancel each other and a third refuses to look at behavior

The design table asks the reader to choose between incompatible mandates. `bencium-controlled-ux-designer` is positioned for enterprise and regulated work, with WCAG 2.1 AA, mathematical scales, and an always-ask-first protocol. `bencium-innovative-ux-designer` is positioned for landing pages and campaigns, and commits to shadows, gradients, and experimental typography. Both are offered as production choices, and nothing in the marketplace ranks them or flags a conflict when both end up in the same project. A third design entry narrows its own scope instead of broadening it: `design-audit` produces phased, implementation-ready design plans and is described as purely visual, explicitly not touching functionality. The sixth design row, `typography`, is the only one with an enforcement rule, since it applies automatically to generated HTML, CSS, and React code rather than waiting to be invoked. WCAG 2.1 AA is the single external standard named anywhere in the four tables, and it carries its version number.

Marketing is one row, and it is the only skill with a web page in the tree

Marketing gets a single row, and it is the only skill in the repository that ships a page you can open. `insurgent-campaign` targets startups, NGOs, and solo brands that are being outspent by incumbents. Its stated procedure is to audit spend asymmetry, assemble what it calls an insurgent channel stack, refuse to fabricate competitor names, and output a lift-test plan instead of vanity metrics. That refusal is a data integrity rule written into the table itself, and it matters more than the campaign framing, because a tool that will not invent names cannot be used to manufacture a comparison. The linked page lives in the tree as `insurgent-campaign/index.html`, so this row resolves to a file a reader can inspect before installing anything. Every other row resolves to a SKILL.md and nothing else. That asymmetry is not a defect, but it does mean the depth of the remaining sixteen entries is knowable only after they are installed.

The framework rows carry the marketplace's only internal counts

Seven rows describe ways of thinking rather than artifacts, and the concrete numbers live in one of them. `hungarian-humanizer` is the outlier: it targets Hungarian text only, and it works from a fixed inventory of 26 patterns, split into 12 Hungarian-specific and 14 universal ones, plus 4 style markers. Those three figures reconcile with each other, and no other row carries a count. `negentropy-lens` evaluates systems along entropy and negentropy to surface tacit knowledge gaps, and `adaptive-communication` sorts the user into high-context relational or low-context transactional and adapts its replies to match. In the development table, `vanity-engineering-review` reviews codebases, architectures, PRs, and technical plans for code built for ego rather than user value, while `bencium-aeo` optimizes content for citations in ChatGPT, Claude, Gemini, and AI Overviews. `renaissance-architecture` and `human-architect-mindset` add first-principles design, domain modeling, and AI-aware problem decomposition. Only the Hungarian row tells you exactly how much is inside it.

Editorial conclusion

Treat bencium-marketplace as a directory to read before an install rather than a bundle to adopt in one step. The design, typography, and audit skills are self-contained and arrive through any documented route, but bencium-loop needs a second fetch, emotion-statusline and spinner-verbs.json need manual work, and eu-ai-act-reviewer ships with no description of what it does. Before adopting the set wholesale, check what your own pipeline already enforces, because the typography skill auto-applies to generated HTML, CSS, and React code and the two UX skills hand out opposite instructions.

Frequently asked questions

Official sources

1. bencium/bencium-marketplace on GitHub
2. Issues
3. License: MIT
4. Project website
5. README