dsh-code-scan

Asking a DeepSeek Harness agent to review code used to mean leaving the chat, opening a terminal, running semgrep by hand, and pasting the raw JSON back into the conversation. dsh-code-scan folds that entire loop into a single natural-language instruction: point the agent at any directory and it handles the scan, parses the results, and returns a ready-to-read report without you ever switching windows.

The report groups findings by file, tags each entry with a line number and a severity level (ERROR, WARNING, or INFO), sorts severity-first, and caps output at 200 entries by default. This keeps the most critical issues floating to the top while preventing oversized result sets from overwhelming the agent's context window. If semgrep is missing from the system PATH, the plugin returns a clear Chinese-language hint instead of crashing, keeping the debugging experience smooth.

Built for developers who already rely on DSH for daily coding, pull-request review, or lightweight security audits. There is no new toolchain or workflow to adopt; simply ask the agent to scan a directory and the security check slots cleanly into the same reasoning chain you are already working in.