Bootstrap

Bootstrap is what you do after onboarding. Onboarding makes the workspace usable; bootstrapping gives it shape. You do it by talking to Nenji, the built-in guide, in two kinds of prompts: explainer questions when you want to understand or inspect, and workflow prompts when you want to design and build.

Nenji has two modes

The mode decides whether Nenji can change anything:

• Base mode (read-only) — Nenji explains, inspects, designs, reviews, and debugs. It will happily draft a plan, but it never mutates your workspace. Use it for explainer questions.
• #creator mode (write-enabled) — type #creator at the start of a message to let Nenji actually build. It routes the work to focused builder abilities (agents, abilities, domains, context blocks, Library knowledge, councils, routines, and project work), reads current state first, and verifies every change by reading it back. Use it for workflow prompts.

A good rhythm is: ask in base mode until you like the plan, then re-send with #creator to build it.

Explainer questions

Use these in base mode to understand Nenjo or inspect what you already have. No prompt prefix needed.

Understand a concept:

Explain the difference between an ability and a domain, and give me one concrete example of when I'd use each.

How do routines handle failure? Walk me through what happens when an agent step fails versus when a gate fails.

How does knowledge retrieval actually work for my agents — how do they find and read the right documents?

Decide what to build:

I need Nenjo to support this capability:

[describe the capability]

Should it be an agent, ability, domain, routine, council, or task? Explain the tradeoffs and recommend the simplest design.

Inspect and review your workspace:

Inspect my workspace and explain what each agent is responsible for, which tools and scopes it has, and where any look too broad.

Review my "[routine name]" routine. Explain what it does step by step, where it could fail, and whether it's missing any gates.

Workflow prompts

Use these with #creator to design and build. Nenji will usually propose the shape first, then build once you approve.

Build a routine from a process — the most common workflow prompt:

#creator turn this process into a routine:

[paste the process — the steps, who does what, and where approval is needed]

Design the trigger, steps, agents, gates, and success/failure paths first. Show me the graph, then build it once I approve.

Create a focused agent:

#creator create a focused [role] agent for this project. It should [describe its one responsibility]. Keep its scopes and tools minimal, and tell me exactly what you assigned.

Set up a council for collaborative work:

#creator set up a council that reviews [what] before it ships — a proposer, a critic, and a leader that synthesizes. Recommend the delegation strategy and explain why.

Scaffold project work:

#creator scaffold a project for [describe the work], with an initial set of tasks and the first agent to run them.

Create Library knowledge:

#creator create the initial Library knowledge for [domain] — a small set of linked documents my agents can search and cite. Propose the documents and their relationships before creating them.

Bootstrap the whole workspace:

#creator help me bootstrap this workspace for [team or domain].

Ask me the minimum questions you need first. Then propose the first project, the first agents, a starter routine, and the Library knowledge to seed — and build it once I approve the shape.

Tips for better prompts

• Describe the work, not the resource — let Nenji choose the right primitive.
• Name your constraints: required tools, models, repositories, approval steps, or compliance rules.
• Say whether Nenji should ask questions first or propose a draft immediately.
• Ask for the smallest useful version — you can always grow a resource later.

What to expect

In base mode Nenji answers, inspects, or hands you a design. In #creator mode it makes the smallest correct set of changes, uses the narrowest builder ability for each resource, and reports back the stable slugs and refs it created so you can find and revise them later. For larger bootstraps it will start with a plan and build only once you approve the shape.