DeepSeek Harness Plugins · Workflow
Plugins that extend DeepSeek Harness (DSH) with vision, workflows, memory, web tools and more.
Manage AGENTS.md instruction files with global and workspace editing, custom path routing, and cross-tool source scanning with one-click default switching.
Engineering brakes for AI agents — enforced self-checks, course correction, evidence-driven debugging, and multi-agent handoff discipline in four zero-dependency, install-and-use skills.
A five-stage SDD workbench for DeepSeek Harness covering requirements, prototyping, system design, spec design, and dev-test, with project board, deliverable versioning, Git collaboration, and flexible Agent-driven iteration.
Import any ComfyUI API JSON as a persistent dynamic video workflow page with multi-server management, concurrent task queueing, and restart recovery.
A DeepSeek Harness productivity suite: one-command install of a model-routing plugin plus nine bundled agent skills for orchestration, task hierarchies, and concise dev workflows.
A bounded execution-autopilot bundle that maps declared project tasks, plans checks from changed files, delegates long work to Harness Jobs, and verifies output via version-2 receipts.
Zero-config per-turn git checkpoints for DeepSeek Harness: auto-stages and commits a tagged snapshot after each agent turn, with pause, resume, and status tools.
Queue chat messages and auto-dispatch them during DeepSeek off-peak valley windows to reduce API costs.
Global and project-level Markdown rule injection into dsh system prompts, with live file-watching and a visual management panel in settings.
A sidebar spec workspace for spec-driven development built on dsh-better-sidebar, providing OpenSpec-based authoring, change tracking, task checklist writing, and template creation in one view.
Organize long-term insights from DSH conversations into your Obsidian vault with a deterministic seven-step workflow, merge-first deduplication, and two-level validation.
Injects a language constraint at the top of the system prompt before every reasoning step, keeping the agent thinking in Chinese and preventing language drift.