dsh-imagedit

In day-to-day Agent workflows, local image editing is often scattered across ad-hoc scripts, while AI image generators such as gpt-image-2 frequently produce broken contours, cracks, and fragmented artifacts that force costly re-runs. dsh-imagedit bundles deterministic local image processing and smoother generation into a single plugin, letting an Agent complete a full pipeline—from cutout to export—in one call, with no cloud API or manual command stitching required.

The feature set covers nearly every routine image operation: two cutout modes (dependency-free flood-fill for quick results, rembg AI for precision); trim, horizontal/vertical flip, rotation, padding, and canvas normalization (scale + center); brightness, contrast, and saturation adjustments; blur and sharpen filters; rounded corners, custom borders, and EXIF auto-orient; JSON-manifest-driven batch processing with recursive folder scanning, plus automatic sprite-sheet assembly and sprites.json generation. On the export side, PNG output is auto-compressed via oxipng or pngquant when available, alongside JPEG and quality-controlled WebP. For generation, the generate_image tool automatically appends a smoothness modifier to the prompt, addressing gpt-image-2's tendency toward fragmented rendering at the source.

The plugin ships in two complementary forms: a cordis plugin that registers image_edit and generate_image as Agent tools, and a Python CLI skill with identical functionality. The Agent prefers the tool when available and falls back to the CLI otherwise. It is well suited for workflow developers who need local asset trimming, bulk image export, or sprite-sheet production, as well as DSH-ecosystem users who value reproducibility and want to stay entirely offline.