Skip to content

Skills

A Skill is a reusable workflow card: it tells an agent how a certain class of task should be done.

It is not a tool and not a script. An agent sees which Skills are available in the SkillCatalog and opens the matching SKILL.md with read when it needs the procedure — so a Skill is always Markdown you can open and read, never an invisible black box.

Built-in Skills

Writing Flow (use in order)

SkillWhat it does
novel-guideThe overview roadmap. Read it first when you are not sure which skill applies now
novel-idea-explorationIdea exploration: genre, rough plot, giving shape to a vague world
novel-setupThe four project setup phases: project init → worldbook skeleton → character design → World Engine init
novel-writingThe writing loop controller: story design → sign off and commit → prose / review / revision. The golden opening three chapters are its first-round specialization
novel-writer-executionThe detailed execution manual for a writer that has received an assignment

Always Available

SkillWhat it does
novel-genre-researchGenre and competitor analysis, wired to ranking data
novel-technique-character-card-workshopHeavyweight character building and character sheet curation
llmlintProse style checking and repair, see llmlint
stop-slopStrip the AI voice while you write

Import Tools

SkillWhat it does
novel-import-silly-tavern-cardSillyTavern character card / worldbook / preset import
novel-import-tomato-referenceImport external novels from epub or downloader output

For Developers

SkillWhat it does
profile-system-guideA system tour of the Harness, TSX profiles and ProfileTurnPlan
tsx-profile-editingEditing TSX profiles, see Write a Profile from Scratch
skill-creator / skill-creator-zhCreating and maintaining Skills themselves

Legacy

RP模式 serves the retired RP entry points. It is preserved but currently unusable.

Where Skills Live

Like workflows, Skills use three-layer overrides, addressed by directory name:

LayerLocation
System built-inShips with NeuroBook
User layer.nbook/agent/skills/<key>/SKILL.md under the Workspace Root
Project layer.nbook/agent/skills/<key>/SKILL.md under the current project root

SKILL.md declares name and description in frontmatter — the description is what makes an agent think of using it, so spell out when it applies instead of just naming the feature.

Runnable Skills

Some Skills ship their own CLI; llmlint is one. These are called runnable skills and have a formal package contract: the version in package.json is the SemVer source of truth, and the catalog reports the version and the root path. Their node_modules does not enter managed assets, and first use goes through a dependency install gate.

Visible ≠ Usable

A profile can declare an allowlist with skills.include. Seeing it in the catalog does not mean it can be used — the allowlist is filtered uniformly in the prepare layer. For example, the allowlist of leader.assets holds only four asset-editing skills.

Skill vs. Workflow vs. Tool

What it providesWhen to use it
ToolOne atomic capabilityRead a file, write a slice, call a child agent
SkillKnowledge and method"What steps this kind of task takes and what to watch for"
WorkflowFixed execution orchestrationSteps are settled and you need concurrency, loops or a state chart

Do not build a workflow just to package a block of instructions. That is a Skill's job.

Keep Reading