Model or dataset
haowjy/creative-writing-skills avatar
haowjy/creative-writing-skills

Creative Writing Skills: a set of Claude agents for writing fiction

claude skills focused on creative writing

526 stars71 forksPythonApache-2.0

At a glance

What is it?
This repository packages creative writing as a set of agents and skills for Claude Code, Claude.ai and the Meridian harness, where one agent plans and drafts, another criticises a single craft dimension at a time, a third behaves like a reader, and the knowledge of the story is kept in a growing knowledge base rather than in a chat window.
Who is it for?
Creative Writing Skills fits a writer who already has prose and wants the model to match their voice rather than produce generic prose, and it fits a writer who works in long form and needs continuity tracked across sessions, because the knowledge base is a project artefact rather than a chat history. It does not fit someone who wants a single prompt, because the design assumes a harness that can dispatch several agents and hold shared project files.
Can I use it commercially?
Yes. Apache-2.0 is a permissive licence: you can use, modify and sell software built on it, as long as you keep its copyright and licence notices.
Is it still maintained?
Yes. The repository last received commits 5 days ago.
What is it written in?
Mainly Python, according to GitHub's language statistics.

Answers come from the project's GitHub data, last synced on October 20, 2026, and from our analysis. They are not legal advice.

Editorial analysis

The shape of the thing: shared skills, specialised agents

The premise is that writing fiction is not one task. The repository describes a system for writing novels, short stories and serial fiction with AI that maintains your voice, tracks your continuity and gets better the more you use it, and the architecture follows from that. From first brainstorm to polished draft, specialised agents handle each mode of work, meaning writing, critiquing, revising and exploring, while shared skills carry the craft methodology. That separation is the interesting design decision. The skills are the knowledge, the prose and scene craft, the style and voice references, the genre and page-level technique, and they are written once and consulted by whichever agent happens to need them. The agents are the workers, and each has one job. A reader of the repository structure sees the same split on disk, with an agents/ directory, a skills/ directory, a bootstrap/ directory and a drafts/ directory.

Installing it under Meridian, Claude Code and Cowork

There are several installation paths and they are not equally capable. The Meridian path is two commands plus a bootstrap.

bash
meridian mars add haowjy/creative-writing-skills
meridian mars sync
meridian bootstrap

After that you run meridian to start a session with muse, and the first-time setup runs the bootstrap automatically. Claude Code and Cowork use the same plugin format, and for Claude Code the installation is a slash command that adds the marketplace and then installs from it.

bash
/plugin marketplace add haowjy/creative-writing-skills
/plugin install creative-writing-skills@cw

Cowork goes through the sidebar instead, Customize, then Personal plugins, then the plus button, then Create plugins, then Add marketplace, then Add from repository, where you enter haowjy/creative-writing-skills and install the creative-writing-skills plugin, and the readme notes that Cowork follows the same flow as Claude.ai but also runs the agents. Once either is installed, you start a session with muse as the agent.

bash
claude --agent creative-writing-skills:muse

Then you run the one-time project setup, which is a slash command rather than a shell one, and it creates your AGENTS.md file and your kb/ directory structure.

code
/creative-writing-skills:project-setup

Claude.ai, where the muse does the talking

The Claude.ai route is different in kind, because there is no agent dispatch there. creative-writing-muse turns an ordinary Claude chat into a creative writing partner: you turn it on, tell it what you are working on, and it carries you from first idea to finished draft in one conversation, talking through ideas, drafting scenes in your voice, flagging what is not working and revising. The other skills, prose, scene craft, critique and structure, are the craft knowledge it draws on, and you do not have to invoke them yourself. That last point is the difference between the two setups in one sentence. On Claude Code you invoke specialised agents; on Claude.ai the muse invokes the knowledge itself.

Getting the skills onto Claude.ai takes one of two routes. The easier one is the marketplace: in the sidebar, Customize, then Personal plugins, then plus, then Create plugins, then Add marketplace, then Add from repository, enter haowjy/creative-writing-skills and click Sync. The other is to download the .skill files from the latest release and upload them one at a time through Customize, then Skills, then the plus button, then Upload skill. Then you start a chat, turn on creative-writing-muse and describe what you want to write. If you would rather build the files yourself, the readme says to clone the repo and run the packaging script.

bash
python scripts/create_skill_zips.py

That regenerates the archives in the zips/ directory.

The three phases, and the flow between them

The diagram in the readme is worth reading as prose because it explains the workflow. Everything routes through muse, and muse sits above three areas. The Explore and Plan area fans out brainstormers for creative variety, spawns character-sims to discover voices, and uses outliners to shape structure once a direction is chosen. The Draft and Revise area is where prose work happens, with the writer producing fresh drafts, revisions, bridges, alternate takes and polish, and critics evaluating focused craft dimensions while an editor gives a whole-book editor pass across structure, voice, line quality, copy consistency and proofing. The Knowledge area holds the kb-lead and the style-creator.

The arrows between the areas are the part that makes it a system rather than a pile of prompts. Explore feeds Draft once a direction is confirmed, Draft feeds Knowledge with decisions and facts, and Knowledge feeds back into the next session's context. The two dotted arrows inside the draft area are the checks that run in parallel rather than in sequence, with muse routing key scenes to the reader-sim and voice checks to the continuity-checker.

The agents, one job each

There are ten agents and the table gives each a single role. muse is the author-facing creative partner for all story work, from planning through production handoff. writer produces production prose from briefs, critique notes and style references, using progressive mode guidance for drafts, revisions, bridges, alternate takes and line polish. critic gives deep adversarial critique of a draft, one focus area at a time, which is the phrase that matters, because it means the critique is not a general impression but a pass over a single dimension.

The editor is described as a third-party book editor, covering structure, voice, line quality, copy consistency and proofing priority, which sets it apart from the critic by being the one that looks at the whole thing rather than at a facet. reader-sim gives experiential reader response to a draft, moment by moment, and character-sim does in-character conversation for voice discovery and relationship testing. continuity-checker cross-references content against established canon for contradictions. brainstormer does creative option generation for a scoped question or angle, and outliner sequences confirmed direction into arc, chapter and beat-level outlines. style-creator analyses prose samples to produce style reference files for the project's voice.

The knowledge base, and why it is not an agent

One design decision is stated explicitly and it is the one that makes long-form work possible. Knowledge extraction into the knowledge base is not a bundled agent: muse routes it to meridian-base's kb-lead loaded with the story-memory skill, or performs it directly where no kb-lead subagent exists. Fact extraction into the knowledge base flows through the story-memory skill, and style-creator captures voice patterns from prose samples, so the knowledge base grows as the project evolves and gives every future agent accurate context.

The difference between the two cases is worth noting because it is a portability decision rather than an oversight. On Meridian there is a kb-lead agent to dispatch the work to, and on a harness without one the same task has to be done in the main session. Either way the output is the same artefact, a set of facts and decisions written to the project directory, which is what makes the knowledge survive the end of a conversation. The top-level repository entries show the same structure on disk, with a kb/ style project directory created by the setup command, a drafts/ directory for prose, an AGENTS.md at the root, and a skills/ directory holding the shared craft material.

The skills, and the promise of getting better with use

The skills table is where the craft knowledge lives, and the readme lists creative-writing-modes for the pen-on-paper prose modes of fresh draft, revision, bridge, alternate take and line polish, and creative-writing-craft for craft references on prose, scenes, style, voice and genre or page-level technique. writing-principles is described as covering reader reward channels and AI failure modes, which is the most interesting entry, because it is the part of the system that is about the model rather than about fiction. The remaining skills named elsewhere in the readme are prose, scene craft, critique and structure, plus story-memory for the knowledge base.

The claim that the system gets better the more you use it rests on the style files. You create style files from your existing prose, and then draft new scenes that match, which means the voice of the project is an artefact that survives between sessions rather than a setting you re-describe each time. The rest of the feature list describes the surrounding workflow. You can brainstorm without committing, exploring plot options, character arcs and world mechanics with multiple AI perspectives before deciding anything. You can catch your own mistakes with structured critique, continuity checks and simulated reader reactions. And you can keep everything in sync, because the knowledge base updates as your story evolves. The repository is licensed under Apache 2.0, the primary language is Python for the packaging script, and the release history shows a series of v0.5.x tags in October 2026, the newest being v0.5.14.

Editorial conclusion

Creative Writing Skills fits a writer who already has prose and wants the model to match their voice rather than produce generic prose, and it fits a writer who works in long form and needs continuity tracked across sessions, because the knowledge base is a project artefact rather than a chat history. It does not fit someone who wants a single prompt, because the design assumes a harness that can dispatch several agents and hold shared project files. It also does not fit a writer who does not want the model reading their drafts closely, since the critique and continuity agents exist precisely to point at problems. Before you start, set up the project structure with the project-setup command so the AGENTS.md and kb/ directory exist, and decide whether you want muse's single guided conversation on Claude.ai or the full agent set under Claude Code, since the second is where the specialised agents actually run.

Frequently asked questions

How do I install the Creative Writing Skills?

On Meridian, run meridian mars add haowjy/creative-writing-skills, then meridian mars sync, then meridian bootstrap. On Claude Code, run /plugin marketplace add haowjy/creative-writing-skills and then /plugin install creative-writing-skills@cw, and start a session with claude --agent creative-writing-skills:muse. On Claude.ai, add the marketplace from the sidebar or upload the .skill files from the latest release.

What is the difference between muse and the other agents?

muse is the author-facing partner for all story work, from planning through production handoff, and it routes work to the other agents. The writer produces prose, the critic gives adversarial critique on one focus area at a time, the editor gives a whole-book editor pass, reader-sim gives experiential reader response, and character-sim, brainstormer and outliner handle exploration.

How does the system keep track of my story between sessions?

Through a knowledge base in the project directory, created by the project-setup command together with an AGENTS.md file. Fact extraction into it flows through the story-memory skill, dispatched to meridian-base's kb-lead agent where one exists and applied directly where it does not, and the knowledge base grows as the project evolves so every future agent has accurate context.

Can the AI write in my own voice?

That is what style-creator is for. It analyses prose samples to produce style reference files for the project's voice, and writer drafts new scenes from those references. The style files live in the project directory, so the voice is carried across sessions rather than re-described each time.

Official sources

  1. haowjy/creative-writing-skills on GitHub
  2. Issues
  3. License: Apache-2.0
  4. README
  5. Releases
Add this badge to your README

If you maintain this project, the badge below links readers to this analysis and shows its maintenance status from the daily GitHub snapshot. Paste the markdown into your README; add ?metric=license or ?metric=stars to the image URL for a different field.

Add this badge to your README

markdown
[![Hysen Labs](https://hysenlabs.com/badge/haowjy-creative-writing-skills.svg)](https://hysenlabs.com/projects/haowjy-creative-writing-skills)

Was this useful?