Preface¶

Once a long manuscript reaches tens of thousands of words, common issues are rarely about writing style, but rather contradictory settings, mismatches between the main text and character profiles, becoming increasingly disorganized after revisions, and it is difficult to catch all problems in a single full read-through. Relying solely on “letting the AI read once,” quality and cost are uncontrollable.

Below introduces sarsgy-wenjian/dsh-wenjian-team. It breaks a multi-person collaborative verification process into 10 reusable skills, using the sequence “Read-only Check → Three-tier Suggestions → Author Selection → Targeted Replacement → Word-by-word Review,” and does not automatically rewrite the manuscript. It is suitable for DSH users who need to repeatedly verify long manuscripts and setting collections.

What is it¶

A DSH plugin, source code hosted on GitHub, maintained by sarsgy-wenjian, license MIT. It deploys 10 skills in DSH (installed to ~/.dsh/skills) and a wenjian_team tool, used to perform structured validation on long novels, plot outlines, and setting collections.

It addresses three types of problems: contradictory settings, conflicts between content and baselines (setting overview, character profiles, project rule packs), and the lack of confirmation lists and second-level verification during the revision process.

Core Features¶

The plugin comes with 10 skills, each corresponding to a role:

The main process is six steps: Create Package → Validate (Content + Structure) → Three-tier Suggestions → Author checks items one by one → Targeted Replacement → Review. Each step stops to wait for confirmation. In the final review sheet, blocking must be zeroed out, and unchecked items will not be executed.

It also has two clear boundaries:

• The validation and suggestion roles are read-only; revisions only execute based on the author’s checked item list; saving to disk requires separate authorization;
• The plugin only contains skills and templates and does not carry any manuscripts, settings, or scoring answers.

Installation & Enablement¶

Install from GitHub:

dsh plugin --profile desktop add github:<sarsgy-wenjian>/dsh-wenjian-team

For local development, you can use the link method to point to the repository directory:

dsh plugin --profile desktop add link:/path/to/dsh-wenjian-team

After installation, you need to restart DSH (bundles are loaded when the process starts), then open a new session, and ask the AI to call the wenjian_team tool to confirm status. The new session should return that all 10 skills are “already up to date” and prompt to load wenjian-team-lead.

The uninstall command is:

dsh plugin --profile desktop remove dsh-wenjian-team

You can also ask the AI to call uninstall on wenjian_team in a new session; this method only deletes skills that haven’t been modified by the user, while skills modified by the user will be retained.

Typical Usage¶

First prepare two things: the manuscript to be verified (outline/main text/setting collection) and the baseline (setting overview, character profiles, project rule packs). If there is no rule pack, you can first let wenjian-rules-compiler guide the creation of a pack.

Then in a new session say:

Load the wenjian-team-lead skill, form a team according to the 00 Team Formation instructions, and use this workflow to verify “My Manuscript”. The manuscript is at …, the baseline is at ....

The Team Lead will execute according to the six steps, stopping at each step to wait for confirmation. Before finalizing, check the review sheet: blocking must be zeroed, and unchecked items will not be executed.

Note two points: The plugin will not automatically form a team; the Team Lead needs to spawn teammates according to the team formation commands in wenjian-team-lead, and it depends on DSH Agent Teams; the plugin does not contain any project data; rule packs and task package templates need to be prepared by the user or generated according to templates.

Suitable Scenarios & Notes¶

Suitable for scenarios that require multiple rounds of verification for long manuscripts and setting collections, and want a confirmation list and verification record for every change; also suitable for users who want to run “content validation” and “structure validation” separately.

Before use, confirm a few points: The plugin runs with the permissions of the current DSH process; before installation, check lib/index.js, each skill’s source code, and LICENSE to ensure they meet your security requirements; you need DSH Agent Teams to complete the process fully; after installation, DSH must be restarted for the bundle to load.

The README provides internal testing costs. The cost-reduction version v2.3.2 was ¥2.559 (-32.3%) on a roughly 9k-word outline project, with output tokens -18.9%. These numbers come from records of the same project and are only for reference of magnitude; different manuscripts, baselines, and model routing may have differences.

Summary¶

The value of this plugin lies in transforming long manuscript verification from a “single read-through” into a workflow with division of labor, confirmation lists, and verification: validation roles are read-only, revisions only execute based on checks, and saving requires separate authorization. The project directory page is at https://www.skillhub.cn/plugins/sarsgy-wenjian/dsh-wenjian-team, and the source code is at https://github.com/sarsgy-wenjian/dsh-wenjian-team.