B — Some risk signals Graded October 16, 2026Minor concerns worth a quick check.Based on public adoption and transparency signals as of October 16, 2026. Our opinion, not a code audit or security guarantee.
AI agent skill · Developer tools, Web & browsing
oh-story-claudecode
Claude Code / Codex / OpenCode agent skills for writing Chinese web novels (网文): 扫榜、拆文、写作、去AI味、封面全流程，长篇短篇都支持 | 13 skills for novel writing and long-form fiction with file-based continuity tracking. MIT.
Bottom line: Signal-based grade (4/7 points): very widely starred on GitHub, public source code. This reflects adoption and transparency, not a code audit.
- Popularity score
- 44/100
- Rank
- #225 of 500
- GitHub stars
- 7.7k
What it does
oh-story-claudecode provides 13 specialized AI agent skills for Chinese web novel (网文) writing, including ranking analysis, story structure decomposition, content generation, AI-style removal, and cover art creation—supporting both short and long-form fiction with file-based continuity tracking. It integrates with Claude via MCP to streamline the full novel production pipeline from planning through publication.
Who it’s for
Chinese web novel authors and publishers seeking to accelerate drafting and editorial workflows. Best suited for creators comfortable with AI-assisted writing who want structured, end-to-end tooling rather than generic LLM prompts.
Copy-paste setup
Check these steps against the project’s own instructions before running anything.
Verify against the repo README at https://github.com/zenstory-ai/oh-story-claudecode (opens external site) for the latest Claude Desktop integration steps. The project is MIT-licensed; typical MCP setup involves adding server configuration to Claude Desktop's mcpServers JSON, but exact syntax and authentication requirements should be confirmed in official documentation.
Risks
- Data persistence & privacy: File-based continuity tracking stores novel drafts locally or remotely; verify where intermediate/completed work is retained and whether sensitive plot details are encrypted.
- Prompt injection via content: Long-form fiction input could contain adversarial text designed to manipulate the AI's behavior or bypass editorial guidelines.
- AI-generated content attribution: Use in commercial publishing may create legal ambiguity around authorship, copyright ownership, and disclosure obligations to readers/platforms.
- Model-specific dependency: Tight integration with Claude APIs means feature changes, rate limits, or pricing shifts directly impact workflow continuity.
- Supply-chain trust: 7.6k GitHub stars suggest adoption, but limited npm download signals and no visible funding/org backing; monitor for abandoned maintenance.
How to stay secure
Least-privilege permissions
- Restrict file access: Configure the skill to read/write only to folders containing novel drafts; block access to system files, credentials, or other projects.
- API key scoping: Use Claude API keys with usage limits and disable unused features (e.g., if you only need writing, turn off image generation); store keys in a password manager, never in code or shared documents.
- User role limits: Assign team members read-only access unless they need to edit or publish; separate "writer," "editor," and "publish" roles if the platform supports it.
Network isolation
- Encrypt file storage: If drafts are stored remotely (cloud), enable encryption-at-rest; verify the host (GitHub, AWS, etc.) meets your data protection requirements.
- HTTPS-only connections: Ensure all communication with Claude's servers and file storage uses encrypted channels; disable plain-text or unencrypted fallbacks.
- Audit data transfers: Document where novel files flow (local device → API → storage); regularly check logs to confirm no unexpected copies are being sent elsewhere.
Monitoring
- Log all generations: Record when content is created, revised, or exported; flag unusually large requests or repeated failed attempts that suggest prompt injection attacks.
- Watch API usage: Set spending alerts and quota thresholds; sudden spikes may indicate unauthorized use or a breach.
- Review version history: Check that file-based tracking captures all edits; look for unexplained changes or deletions, especially to sensitive story elements before publication.
Red flags to watch for
- Unexpected behavior changes: If the skill suddenly produces off-brand writing, ignores your guidelines, or refuses tasks it previously handled, suspect prompt injection in a novel excerpt.
- Unattributed or AI-detected content: Before publishing commercially, use plagiarism/AI-detection tools to verify the output; confirm you have legal rights and plan disclosure of AI use to readers/platforms.
- Abandoned tool warnings: Monitor the GitHub repository for inactive maintenance, unanswered security issues, or lack of updates for >6 months; consider switching to actively maintained alternatives.
Alternatives
- Sudowrite — Purpose-built AI writing assistant for long-form fiction with genre-specific templates; proprietary, closed-source.
- NovelAI — Subscription model for story generation and character development; independent infrastructure, not MCP-based.
- Atticus or Vellum — General content automation platforms with some novel/publishing use cases; broader scope, less specialized.
- Native Claude API + custom prompts — Lower-level, no framework overhead; requires manual workflow orchestration but full transparency.