Introduction¶

One of the core design principles of DeepSeek Harness (DSH) is a plugin architecture. The dsh-ditto plugin addresses the pain point of manual review in batch tasks. When processing 30 similar files, letting the AI execute tasks directly can easily accumulate errors and lacks a human review step. This plugin introduces a “review first, then apply” mechanism, establishing a standard by validating a small number of samples, then performing full preview and execution.

What It Is¶

dsh-ditto is a DeepSeek Harness (DSH) plugin authored by darrien1998. It is a Review-first batch automation tool. Its core workflow is: first have the AI extract and process three samples for human review and correction to establish a standard; then preview all remaining files based on that standard; finally, after human confirmation, safely write files in bulk.

Installation and Activation¶

The plugin is installed using DSH’s plugin command. Before installation, ensure that pnpm is available for the target profile.

dsh plugin --profile web add dsh-ditto

After installation, use the doctor command to check the environment and permissions:

dsh plugin --profile web exec dsh-ditto doctor

*Note: The --profile parameter must be replaced according to your actual environment (for example, headless, acp, or a custom name).

Core Features¶

The plugin provides the following capabilities:

• Review-first batch automation: Review first, then process the full batch.
• Evidence-backed Code to Spec: Generate code-to-spec documents with line-number evidence.
• Safe bulk file organization with manifests: Safe bulk file organization, supporting the generation of CSV, SQL, Markdown, and other manifests.
• Deterministic ZIP delivery: Deterministic ZIP delivery, supporting verification manifests and sidebar files.
• Human-in-the-loop approval: Human approval is required before the full preview.

Usage Examples¶

Code → Spec¶

Generate a unified Markdown specification document from a codebase:

1. Discover: Scan directories and report the scope and exclusions.
2. Review: Extract three structurally different samples (such as the one with the most exports, the one with the most imports, and the smallest one), generate a draft with line-number evidence, and edit it manually until satisfied.
3. Approve: Confirm the samples as the standard.
4. Preview: Based on the sample standard, preview the documentation for all remaining modules. Content that cannot be supported by evidence is marked as “Needs confirmation”.
5. Apply: After approval, write the Markdown files to the output folder (source files are not modified).

File organization¶

Batch rename and migrate PDF files:

1. Define Rule: Use regular expressions to match file names and define source and target paths.
2. Review: Preview all proposed copies and deliverables.
3. Manifests: Before approval, you can export mapping manifests in CSV, JSON, or Markdown format, or generate SQL Server INSERT statements (text generation only; no execution).
4. Apply: After confirming there are no issues, perform the copy operation. A ZIP archive is generated.

Notes¶

• No model calls: The plugin itself does not call LLM APIs and does not store API keys. Model calls are handled by the DSH agent; the plugin only receives structured evidence and drafts.
• Data safety: Original files are never moved or modified; existing files are never overwritten. All operations are based on previews, and files are written only after confirmation.
• Environment requirements: Node.js 22 or higher and DSH 0.1.5-rc.1 or higher are required.
• SQL safety: Although the plugin can generate SQL statements, it never connects to a database or executes SQL.

Conclusion¶

dsh-ditto provides DSH with a solution for maintaining safety and controllability in batch automation scenarios. By keeping decision-making authority with humans and delegating execution to deterministic plugin logic, it reduces the risk of errors in batch tasks.

• GitHub Repository
• DSH Plugin Directory