floka-marketplace
Claude Code plugins for people building real things in Norwegian and English.
Install the marketplace
/plugin marketplace add Floka-as/floka-marketplace
Then install whichever plugins you want:
/plugin install humanizer@floka-marketplace
/plugin install board@floka-marketplace
/plugin install no-nb@floka-marketplace
/plugin install forge@floka-marketplace
What's inside
Strip AI-writing tells from text. Universal patterns from Wikipedia: Signs of AI writing — em-dash chains, the rule of three, sycophantic openers, knowledge-cutoff hedges, AI vocabulary, copula avoidance, signposting, fragmented headers — plus a deep Norwegian (bokmål/nynorsk) section covering anglicisms, særskriving, sin/sitt/sine reflexive possessives, capitalization, quotation marks, number/date formatting, V2 word-order violations, and form-consistency issues that LLMs leak through.
Activates when you ask Claude to humanize, edit, or review text. Optional voice calibration from a writing sample.
Try it. Ask Claude to humanize: "Our cutting-edge platform serves as a testament to the transformative potential of AI." You'll get back something a human would actually say.
Build, audit, and consult a personal advisory board of operator personas. No advisors are bundled — you compose your own roster through a 5-minute interactive interview. The skill suggests archetypes and concrete candidate operators based on your situation, declared biases, and decision-making style. You pick. The result is a markdown file you edit, version, and own.
Five commands: /board:build (compose), /board:audit (evaluate composition for gaps and bias-counterweight balance), /board:invoke (single advisor), /board:meeting (3–5 advisors), /board:pipeline (full canonical strategic sprint, all 8 stages, structured brief out).
Why this and not a chatbot persona? Pre-curated advisor sets ship someone else's taste. The operator who's right for an indie SaaS founder isn't right for a Norwegian B2B services firm. Your board should reflect your situation. This plugin treats the board as a first-class artifact: built deliberately, audited regularly, versioned alongside the rest of your work.
Translate English content into idiomatic Norwegian (bokmål by default, nynorsk on request) — by intent, not word-for-word. Auto-activates on translation requests. Handles markdown, HTML, plain text, emails, marketing copy. Applies Norwegian conventions for numbers (3,14), dates (8. mai 2026), quote marks («...»), capitalization (lowercase months, languages, job titles, headings), and pronouns (du, never De).
Pairs with humanizer — translate first, then run the Norwegian audit pass on high-stakes content.
Why no-nb? Språkrådet tested ChatGPT in 2024 and found ~2.6 errors per page in bokmål and ~8 per page in nynorsk. The errors that remain after that report are mostly anglicisms and English sentence shapes wearing Norwegian costume. This plugin pushes back on those by default.
Spec-driven autonomous development for Claude Code. Seven commands implementing the workflow: /forge:contract writes a thorough spec through code exploration, library research, and gray-area discussion. /forge:auto loops build→evaluate→fix until ACCEPT. /forge:build, /forge:evaluate, /forge:parallel, /forge:pulse, /forge:status complete the toolkit.
The pattern: scout the code → research the libraries (DeepWiki) → discuss gray areas → write a contract → build with evidence-before-checkbox discipline → evaluate skeptically with a fresh-context sub-agent.
Why this matters. Most "let Claude build it" workflows fail the same way: a thin task description becomes a thick pile of code that mostly works but missed the intent. The fix isn't more babysitting — it's a richer artifact before the build. Forge treats the spec as the highest-leverage moment; the autonomous-build loop only works because the contract is rich enough to ground it.
The thesis
Most "Norwegian" output from LLMs is English in translation. Most AI-persona plugins are toy chatbots. Most AI-writing tools chase generic edge cases. Most "let Claude build it" workflows produce code that mostly works but missed the intent.
