When developing agents in the DeepSeek Harness (DSH) environment, code reviews often lack a systematic process. Models can easily miss edge cases or specific security risks. The dsh-plugin-code-review plugin solves this problem by introducing a structured review workflow.

This is a structured code review skill plugin for DeepSeek Harness, maintained by YYTbit. It helps agents complete high-quality code reviews by teaching them systematic inspection steps.

Core Features

The plugin provides agents with a set of inspection criteria covering five dimensions of code quality:

  • Correctness: logic errors, edge cases, error handling.
  • Security: input validation, secret management, injection risks.
  • Performance: unnecessary loops, N+1 queries, blocking operations.
  • Style: naming conventions, complexity, dead code.
  • Testing: coverage, edge cases.

Installation and Enablement

Use the official installation command to add the plugin to the specified DSH configuration file:

dsh plugin --profile your-profile add dsh-plugin-code-review

After installation, the skill runs together with the DSH process.

Typical Usage

After installation, directly issue code review instructions to the agent in the conversation. The plugin automatically applies the inspection criteria above.

  • “Review this diff”
  • “Check this PR for security issues”
  • “Review src/auth.ts”

Output Format

The agent outputs review results in a predefined log format, usually including severity labels and specific locations:

[CRITICAL] src/auth.ts:42 - Missing input validation on user email
[WARNING] src/api.ts:18 - Potential SQL injection via string concatenation
[SUGGESTION] src/utils.ts:7 - Function too long, consider splitting

Notes

  • The plugin runs with the permissions of the current DSH process.
  • Before installation, it is recommended to check the source code and license (MIT) to ensure compliance with usage requirements.

Summary

The plugin improves the agent’s code review capability through a structured process, helping developers quickly identify potential issues. The plugin directory and source code are available at: