dsh-patch-edit-plus

When models need to edit files inside DeepSeek Harness, the native edit tool can be verbose for small changes in large files, and there is no atomic way to batch changes across many files. dsh-patch-edit-plus fills that gap with a single apply_patch tool that accepts git/unified diff by default and optional Codex apply_patch syntax. Every patch is verified against current file content first; if any hunk mismatches the entire patch is aborted with zero bytes written, eliminating the partial-state headaches of sequential edits.

The plugin handles file creation, multi-hunk updates, deletion, and rename operations. Writes follow DSH's official intent dance end-to-end, while delete and move run through a sandbox-aware shell. CRLF line endings and missing trailing newlines are preserved faithfully. On hunk failure the error carries file name, hunk index, expected-line preview with visible whitespace, and the actual file excerpt so the model can retry with full context. Name collisions with other apply_patch providers are resolved by automatic suffixing (apply_patch_1, _2, ...) so DSH never fails to start.

Built for developers who make frequent cross-file batch edits in large codebases, and for teams driving DSH sessions with GPT-family models that naturally prefer Codex patch syntax. The plugin registers exactly one tool, never overrides a native tool or global service, and removes every trace of itself on unload, leaving the existing workflow completely untouched.