Creative Writing Skills is an AI-powered toolkit designed to assist authors in novel and fiction writing. It addresses the challenges of maintaining authorial voice, tracking continuity, and managing the iterative process from initial brainstorming to a polished draft. The system targets writers seeking sophisticated AI co-pilots that learn and adapt to their specific project and style.

How It Works

This project employs a modular architecture of specialized AI agents orchestrated by a central muse agent. Key agents include writer for prose generation, critic for evaluating drafts across multiple reward channels (transportation, aesthetic, social simulation, flow), and revision-writer for targeted edits. A persistent knowledge base (kb/) stores project-specific information like character profiles, world lore, and style guides, generated by agents like style-creator and chronicler. This approach ensures continuity and allows AI-generated content to consistently match the author's established voice and narrative canon, facilitating a collaborative writing loop.

Quick Start & Requirements

Installation can be done via the Meridian CLI (meridian mars add haowjy/creative-writing-skills, meridian mars sync, meridian bootstrap) or within Claude Code/Cowork environments by adding the marketplace (/plugin marketplace add haowjy/creative-writing-skills) or uploading .skill files from releases. First-time setup typically involves a project-setup command. Requires the mars CLI and access to Claude Code, Cowork, or Claude.ai.

Highlighted Details

• Agent Specialization: Utilizes distinct agents for brainstorming (brainstormer), character exploration (character-sim), drafting (writer), critique (critic), revision (revision-writer), continuity checking (continuity-checker), and reader simulation (reader-sim).
• Persistent Knowledge Base: A structured kb/ directory stores evolving project state (styles, characters, world, timeline, canon, issues), ensuring context and consistency across sessions.
• Voice Mimicry: The style-creator agent analyzes existing prose to generate style reference files, enabling new drafts to match the author's unique voice.
• Iterative Refinement Loop: The core workflow involves writing, critiquing, and revising, with agents designed to handle specific aspects of this cycle, including simulated reader feedback.

Maintenance & Community

The README does not specify community channels (e.g., Discord, Slack) or notable contributors/sponsorships. Development is indicated by commands for version bumping (mars version patch).

Licensing & Compatibility

The project is licensed under the Apache License 2.0, which generally permits commercial use and integration into closed-source projects. Compatibility is noted across Claude Code, Cowork, and Meridian environments, with some functional differences, particularly regarding agent orchestration depth.

Limitations & Caveats

Direct agent interaction is limited in standard claude.ai chats, where only skills are accessible. The Meridian version offers deeper, hierarchical orchestration compared to the flatter structure in Claude Code/Cowork. Some features, like the use of rhetorical questions in skill prompts, are noted as current experiments.