Skip to main content
skill CX Codex

kb-management — Claude Code skill: what it does, how to install

Organizes story knowledge bases in kb/ with rules for canon, vocab, and characters. It helps keep lore and continuity consistent across writing agents.

Install into Claude Code · Skills CLI (Vercel) All methods and platforms ↓
npx skills add haowjy/creative-writing-skills --skill kb-management -a claude-code
Open on GitHub

What kb-management does

The skill turns kb/ into durable project memory across writing agents. It organizes story facts into distinct layers: chapter canon in kb/canon/, reference articles in kb/world/, character profiles in kb/characters/, voice samples in kb/styles/, and vocabulary files. Brainstorms, critiques, and drafts stay in work/ until facts are finalized.

Each document covers a single concept and uses descriptive naming like fire-magic.md instead of session dates. When a page exceeds ~200 lines or addresses unrelated topics, the guidelines require splitting it and linking entries with relative paths. For terminology, vocab entries record canonical forms, definitions, aliases, and sources to resolve naming conflicts early.

Check AGENTS.md before editing, because project-specific settings can override this layout. Routine fact extraction from finished chapters belongs in /story-memory, keeping direct edits in kb/ for confirmed author decisions.

What it helps with

  • organize story lore and character profiles in kb/
  • track invented vocabulary with aliases and definitions
  • separate in-progress drafts in work/ from canonical story facts
  • split oversized wiki pages exceeding 200 lines into focused docs

Example prompts for the agent

  • Create a new character sheet in kb/characters/ based on the finalized chapter.
  • Add an in-world term to kb/world/vocab.md with its definition and aliases.
  • Split this lore note into focused files in kb/world/ because it exceeded 200 lines.

How to install kb-management

1Skills CLI (Vercel)

npx skills add haowjy/creative-writing-skills --skill kb-management -a claude-code

2Manually: put the files in place

~/.claude/skills/kb-management/SKILL.md

Copy the whole skill folder; for one project use .claude/skills/kb-management/

Step by step, with commands for each platform: How to install skills in Claude Code: five ways

Questions about kb-management

How do I install kb-management in Claude Code?

Run npx skills add haowjy/creative-writing-skills --skill kb-management -a claude-code.

What does kb-management do?

It provides conventions for maintaining a story wiki in kb/. The skill defines folder structures for canon, characters, vocabularies, and voice styles while keeping rough drafts in work/.

How does kb-management differ from story-memory?

The /story-memory tool handles routine extraction of facts from finished drafts. In contrast, kb-management guides overall folder layout, vocab tracking, and direct edits made by the author.

Source text

SKILL.md · Apache-2.0
---
name: kb-management
description: >
  Maintaining the story knowledge base: creating, updating, and organizing
  wiki-style reference pages in kb/. Use when capturing finalized story
  knowledge, updating character profiles, documenting world mechanics, or
  restructuring the kb.
---

# KB Management

The knowledge base (`kb/`) is the project's durable memory. Every agent reads
from it for context. This skill covers how to maintain it well.

## Layers

**Canon**: established facts the story has committed to. Once a chapter is
published/finalized, the facts it establishes are canon. Contradicting canon
breaks reader trust.

**Wiki**: synthesized reference pages. How the magic system works, character
relationships, faction politics. Living documents that evolve as the story
develops.

**Styles**: voice reference files derived from prose samples. The writer
and critic agents depend on these for voice consistency.

**Vocab**: canonical story terms, aliases, and exclusions. Project-wide terms
live in `kb/vocab.md`; domain terms live beside the domain they govern, such as
`kb/world/vocab.md`.

**Issues**: tracked writing problems that span multiple chapters (recurring
tics, pacing patterns, continuity errors). See the story-memory skill.

## Page Conventions

### One Concept Per Document

Each doc covers one coherent topic: one character, one location, one system.
When a doc covers two unrelated topics, split it. When two docs explain the
same thing from different angles, merge or cross-reference.

Name files by what they describe (`fire-magic.md`, `protagonist.md`), not
when they were written (`session-3-notes.md`).

### Organization

```text
kb/
  characters/
    <name>.md              # one file per character
  vocab.md                 # project-wide canonical terms
  world/
    vocab.md               # worldbuilding terms when needed
    <topic>.md             # locations, factions, systems
    <domain>/
      vocab.md             # subdomain terms when needed
      <topic>.md           # nest when a domain has many pages
  timeline/
    <arc-or-period>.md     # chronological entries
  canon/
    <chapter-or-arc>.md    # hard facts per chapter/arc
  styles/
    <style-name>.md        # voice reference files
  issues/
    <issue-name>.md        # tracked writing problems
```

The project's `AGENTS.md` may customize this. Read it first.

### Linking

Link to related pages with relative paths. Cross-reference instead of
duplicating: one source of truth per concept. A character page links to
the location page for their home, the timeline entry for their arc, etc.

### Readability

Write pages that work in isolation:

- **Self-contained**: enough context that a reader doesn't need three
  other pages first
- **Scannable**: headers, bullets, tables. Bold key terms on first use.
- **Concrete**: specific quotes, chapter references, scene citations
- **Current**: update when the story invalidates or extends what's here

## Vocab Pages

Use vocab pages when terms matter across agents: magic names, faction labels,
place names, titles, relationship labels, invented words, recurring in-world
phrases, and genre terms with project-specific meanings.

Each entry should include:

- **Canonical name**: the form agents should use
- **Definition**: one to three sentences, including what the term is not when ambiguity is likely
- **Aliases**: names the author, characters, drafts, or older kb pages actually use
- **Source**: where the usage was established or decided

Resolve conflicts early. If two terms seem to name the same thing, pick the
canonical form with the author or flag it in the report instead of carrying
both forward silently.

## When to Create vs Update

**Create** a new page when a concept is finalized enough to be referenced
by other agents. Don't create pages for things still in story-planning.

**Update** an existing page when new chapters establish facts about it,
when the author makes decisions that change it, or when a page has become
stale.

**Split** when a page grows past ~200 lines or covers multiple unrelated
concepts.

## What Belongs in KB vs Work

- Finalized knowledge → `kb/`
- Draft iterations, brainstorm captures, critique reports → `work/`
- Promoted facts after a draft completes → `kb/canon/` or relevant wiki page

Use `/story-memory` for routine fact extraction from completed chapters.
Direct kb edits are for the author or muse when capturing decisions
interactively.

© haowjy · Apache-2.0 · haowjy/creative-writing-skills

Similar skills

All skills for Agent setup: memory, context, orchestration →

From the same set22

Project sponsors

Companies that keep this analytics open
Gold sponsor

Impressions, clicks and code copies — open statistics. We keep the numbers public for the same reason NeBlask itself is public. Become a sponsor