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)
| Skill | What it does |
|---|---|
novel-guide | The overview roadmap. Read it first when you are not sure which skill applies now |
novel-idea-exploration | Idea exploration: genre, rough plot, giving shape to a vague world |
novel-setup | The four project setup phases: project init → worldbook skeleton → character design → World Engine init |
novel-writing | The writing loop controller: story design → sign off and commit → prose / review / revision. The golden opening three chapters are its first-round specialization |
novel-writer-execution | The detailed execution manual for a writer that has received an assignment |
Always Available
| Skill | What it does |
|---|---|
novel-genre-research | Genre and competitor analysis, wired to ranking data |
novel-data | Queries locally cached novel rankings and book details for genre analysis |
novel-technique-character-card-workshop | Heavyweight character building and character sheet curation |
llmlint | Prose style checking and repair, see llmlint |
stop-slop | Strip the AI voice while you write |
Import Tools
| Skill | What it does |
|---|---|
novel-import-silly-tavern-card | SillyTavern character card / worldbook / preset import |
novel-import-tomato-reference | Import external novels from epub or downloader output |
For Developers
| Skill | What it does |
|---|---|
profile-system-guide | A system tour of the Harness, TSX profiles and ProfileTurnPlan |
tsx-profile-editing | Editing TSX profiles, see Write a Profile from Scratch |
skill-creator / skill-creator-zh | Creating and maintaining Skills themselves |
Legacy
RP模式 serves the retired RP entry points. It is preserved but currently unusable.
Where Skills Live
Skills are addressed by directory name, and that directory name is the skill id:
| Layer | Location |
|---|---|
| System built-in | Ships with NeuroBook |
| User layer | .nbook/agent/skills/<id>/SKILL.md under the Workspace Root |
A user-layer directory replaces the built-in skill of the same name wholesale; the two are never merged file by file.
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. To show a Chinese label in the interface, keep name as the lowercase English id and put the Chinese name in metadata.displayName.
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 provides | When to use it | |
|---|---|---|
| Tool | One atomic capability | Read a file, write a slice, call a child agent |
| Skill | Knowledge and method | "What steps this kind of task takes and what to watch for" |
| Workflow | Fixed execution orchestration | Steps 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.