DeepSeek Harness Plugins · Workflow
Plugins that extend DeepSeek Harness (DSH) with vision, workflows, memory, web tools and more.
Replaces DSH Web's Plan with a Grok-style hard gate: after /plan the model can only edit plan.md, and the approval card cannot be bypassed by auto-approve.
A Git worktree plugin for DeepSeek Harness that creates task branches, registers isolated Workspaces, and opens dedicated sessions.
Configure project-scoped custom Skill roots for DeepSeek Harness via a per-project YAML file.
Manage MCP server connections and local skill toggles from the Web GUI with zero DSH source changes; all edits persist to the profile.
A read-only DSH plugin that surfaces the live state of Claude Code, Codex, and Cursor sessions and their scheduled tasks in one unified four-state vocabulary.
Resolves unique kebab-case skill substrings in DeepSeek Harness slash tokens, e.g. /oneshot maps to game-package-oneshot when unambiguous.
Say a trigger phrase and your agent takes over the full video editing pipeline — analysis, rough cut, subtitles, voiceover, and QC — entirely on your local machine.
An Agent Society plugin built on the Yogacara eight-vijna architecture, bundling a dual-loop runtime, memory consolidation, MCP tool bridge, and six seed skills into a single installable DSH package.
Bridges .claude/.agents/.trae skills and workspace MCP servers with GUI toggles and persistent state.
Agent-driven kanban boards for DeepSeek Harness with a fixed five-column workflow, per-workspace boards, task-session narration, and human review gates.
A frontier-math research plugin for DeepSeek Harness that scores problem selection, audits prior work, tracks claim dependencies, and verifies independent certificates to improve novel, reproducible results.
Holds background subagent settled notices and reports while the parent session is busy, delivering them as new conversation turns only after it becomes fully idle.