tenki public-claude-skills

A small, opinionated marketplace of Claude Code skills from tenki. Each one packages a discipline that tends to get skipped under deadline pressure.

This repo is structured as a Claude Code plugin marketplace. Install the bundle in one command, or browse individual skills below.

Install

In any Claude Code session:

/plugin marketplace add tenki-labs/public-claude-skills
/plugin install tenki-public-skills

That registers five slash commands: /zk-llm-design, /ml-or-not, /design-ai-experiment, /bokmaal-proof, /reflect.

To update later:

/plugin marketplace update tenki-public-claude-skills

What's in the bundle

Skill	Trigger	What it does
zk-llm-design	/zk-llm-design <system>	Walks through threat model, what is being proved, scheme selection (Halo2 / Plonk / STARKs / FHE / TEE), and feasibility for a given LLM size. Produces a one-page design doc with a smallest-viable-v0.
ml-or-not	/ml-or-not <problem>	Runs a 7-gate pre-check (rules, stability, data, volume, tolerance, drift, economics) and returns a defended verdict: yes / no / not yet. If no, proposes the cheaper alternative concretely.
design-ai-experiment	/design-ai-experiment <change>	Turns a vague AI change into a pre-registered experiment with a falsifiable hypothesis, control condition, sample-size estimate, kill criterion, and a paste-ready pre-registration block.
bokmaal-proof	/bokmaal-proof <text>	Six-pass Norwegian bokmål proofreading: anglicisms, sentence rhythm, register drift, pronoun and tone, Språkrådet recommendations, and særskriving. The pass most Norwegian language tools miss.
reflect	/reflect <decision, belief, action>	Multi-method structured reflection: first principles, steelmanning, pre-mortem, inversion, stakeholder views, second-order thinking. Surfaces the tensions and lands on a position (HOLDS, CHANGES, or DEFERRED with a resolution path).

Click a skill name above for the full SKILL.md.

/zk-llm-design | /ml-or-not | /design-ai-experiment | /bokmaal-proof | /reflect

Why these five?

Each one packages a discipline that gets skipped because doing it well takes 30 minutes of structured thinking that no one budgets for. The skills make the structured version cheaper than the unstructured version.

• /zk-llm-design stops teams from picking a ZK scheme before they have defined what they are proving. The threat model picks the proof; the proof picks the scheme; the model size picks whether the scheme is feasible. Reverse that order and you waste months.
• /ml-or-not stops teams from reaching for ML because it is the default. Most "ML problems" fail at gate 1 (rules a domain expert could write down) or gate 7 (lifetime cost exceeds lifetime value). The skill catches that before sprint planning.
• /design-ai-experiment turns "let's try this and see" into a hypothesis, a control, a sample size, and a pre-committed kill criterion. Past-you with no sunk cost is wiser than future-you under deadline.
• /bokmaal-proof is for Norwegian bokmål writers. LLMs and most language tools handle bokmål grammar fine, but they miss anglicisms, register drift, and særskriving. This skill catches what a Norwegian reader would notice.
• /reflect stops "let me think about this" from producing one perspective dressed up as several. Runs structurally different methods (first principles, steelmanning, pre-mortem, stakeholder views) against the same input, then surfaces where they disagree. The disagreement is where the real insight lives.

Repo structure

public-claude-skills/
  .claude-plugin/
    marketplace.json          # marketplace registry (Claude Code reads this)
  plugins/
    tenki-public-skills/
      .claude-plugin/
        plugin.json           # plugin manifest
      commands/
        zk-llm-design.md      # slash command entrypoints
        ml-or-not.md
        design-ai-experiment.md
        bokmaal-proof.md
        reflect.md
      skills/
        zk-llm-design/SKILL.md
        ml-or-not/SKILL.md
        design-ai-experiment/SKILL.md
        bokmaal-proof/SKILL.md
        reflect/SKILL.md
  LICENSE
  README.md

Forking and customising

The skills are plain Markdown with no scripts and no dependencies. To adapt them:

• Edit the SKILL.md for any skill. Changes take effect on the next /plugin marketplace update.
• Add your own skill: drop a folder under plugins/tenki-public-skills/skills/ with a SKILL.md, and a corresponding commands/<name>.md with the slash-command entrypoint.
• Tighten or loosen the gates in /ml-or-not, swap the scheme table in /zk-llm-design, or add domain-specific anglicisms to /bokmaal-proof. The skills are short on purpose so they are easy to read end-to-end.

PRs welcome if you ship something broadly useful.

Versioning

This repo follows semantic versioning. Skills are content, not code, so version bumps reflect changes in advice or process, not API breaks:

• Patch (0.1.x): typo fixes, small examples, table updates.
• Minor (0.x.0): new skill, new section in an existing skill, meaningful update to recommended approach.
• Major (x.0.0): change in the core process of an existing skill that would surprise existing users.

License

MIT. See LICENSE.

---

Built by tenki.