DeepSeek Harness Plugins · Model Inference
Plugins that extend DeepSeek Harness (DSH) with vision, workflows, memory, web tools and more.
Drive your local Photoshop from AI: batch cutout, batch processing, and arbitrary script execution — zero dependencies, no network.
A lean DSH-native spec-driven development plugin that guides the full workflow from specification to verified implementation in a single-model session, with no external CLI or Git mutation.
Replaces the oldest base64 images with deterministic placeholders when Charm Hyper requests exceed the 10 MiB gateway body cap, with a recovery retry on rejection.
Manage local GGUF models in DeepSeek Harness with auto launch on first message, idle unload after 5 min, and named parameter presets.
A DSH plugin that independently controls AI reply and thinking languages via system prompt injection with flexible configuration.
A TIC-80 fantasy computer plugin for DeepSeek Harness that lets LLMs conversationally build retro games with live hot-reload, sprite editing, chiptune music, and multi-format export.
A DeepSeek Harness host bundle for h3-hip, providing MiniMax-H3 semantic generation, prompt skills, and v1alpha protocol checks for local loopback inference.
Preserves BOM and multi-byte encodings (GBK, Shift-JIS, and 19 others) so DSH agents can read and edit files without corrupting them.
One-click enhancement that rewrites a short draft into a goal-driven, detail-complete prompt using your current session model, with five styles and instant undo.
A distribution layer for DeepSeek Harness that adds free-first model routing, automatic fallback, task-aware scheduling, safe core updates with rollback, multilingual UI, and one-click install on Windows and Linux.
An auditable tool-result context compression selector for DeepSeek Harness, adding an orthogonal code-skeleton compression gate with multiple strategies and exact token measurement.
Adds a request-privacy toggle to DeepSeek Harness that strips user, session, and compaction markers from chat request headers.