Loading ToolBench
Zero-deploy, Git-native content management CLI for Markdown stories and knowledge bases. Built-in MCP Server for AI integration, EPUB export, and structure-aware data pipelines for LLM fine-tuning.
Four tools with complete schemas and descriptions in Chinese. Naming follows verb_noun pattern (scan_, read_, write_, edit_). Descriptions are moderately detailed (100-200 chars) and explain WHAT and WHEN. However, parameter descriptions lack depth on constraints, valid ranges, and error recovery guidance. Output schemas are not documented. Error handling is minimal, no guidance on what to do if a folder doesn't exist or validation fails. Token-optimization hints (verbose, tailLength) are present but not formalized as constraints. No pagination support despite potential for large result sets.
更新故事 config.json 的元数据字段。可编辑：summary/status/series/seriesOrder/volume/links/author/originalWork/originalAuthor/cover/language/wordCount。字段值传 null 表示移除该可选字段。title/type/created/isMultiChapter 为身份或审计字段，禁止修改。写入前经仓库级 schema 校验，校验失败不写盘。修改后请运行 build 更新 README
读取指定故事的章节内容（支持按章节索引按需加载）。tailLength 可只返回章节末尾 N 字符用于续写衔接，节省 Token
列出所有故事及元数据（标题/类型/状态/字数/系列）。默认返回精简列表节省 Token；传 verbose=true 获取完整详情
将正文写入指定故事（原子写入 text.md）。validate=true 时写入后立即执行仓库合规检查并返回结果（不阻断写入）
Output schemas not documented. LLMs cannot predict response structure (fields, types, nesting) for scan_stories, read_chapter, write_chapter, edit_config. Agents must guess what fields to extract for downstream calls.
Parameter constraints missing. 'chapterIndex' (read_chapter) lacks min/max bounds; 'tailLength' has no range guidance; 'fields' (edit_config) has no schema for allowed keys or value types. LLMs will pass invalid values without validation hints.
Error handling lacks recovery guidance. Handler code checks 'if (!folder)' and returns error, but description does not explain what causes this error or how to recover. No guidance on validation failures (edit_config) or missing chapters.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 60 | 2026-07-28+ | v2 |
No pagination for scan_stories. If repo contains 100+ stories, returning all in one response wastes tokens and risks context window exhaustion. Description should mention result limits and offer offset/limit parameters.
edit_config 'fields' parameter is untyped object. Description lists allowed keys (summary, status, series, etc.) but does not formalize them as an enum or schema. LLMs may pass arbitrary keys; validation happens server-side with no pre-call guidance.