DeepSeek Harness Plugins · Workflow
Plugins that extend DeepSeek Harness (DSH) with vision, workflows, memory, web tools and more.
An agent preset for scaffolding projects from scratch with architect persona, six tech-stack templates, and a strict initialization workflow.
One /octo command orchestrates multiple AI models to research, build, and debate, with a 75% consensus gate before delivery.
Shows a Chinese key-info panel in the DSH web settings page, summarizing runtime env, profiles, plugin status, sessions, models, and errors, with a JSON endpoint.
Automatically supplements grep/glob searches with QMD semantic retrieval and asynchronously injects results into the next-step model context, zero dependencies.
A question-driven skill for initializing project documentation and AI-collaboration workflows, with a core focus on remediating existing projects and also supporting new-project scaffolding, cross-platform compatible.
Restore DSH conversations from /export zip or jsonl files with structure and SHA-256 validation, config sync preview, and transactional rollback.
Unlocks image input in DeepSeek Harness by textifying image blocks so text-only opencode-go upstreams can still process images via the vision skill.
A DeepSeek Harness compatibility layer that onboards legacy Claude Code / Codex plugins as unified lifecycle units, preserving original Skill, MCP, and Hook semantics, manageable via web UI or ctx.compat API.
Export DSH settings, credentials, plugin patches, and agent presets as portable bundles or a single AES-256-GCM encrypted file, then restore on a new machine.
Ports Baoyu's 21 daily-efficiency AI skills (illustrations, comics, translation, publishing, and more) into DeepSeek Harness; the agent auto-invokes them on demand.
Two selectable agent presets (General and Coding Pro) for DeepSeek Harness covering research, planning, file handling, and repository-level coding workflows.
Parses Word/PDF question files into printable exam papers with writing space, and features a smart engine for clustering, semantic search, and diverse selection to reduce LLM token costs.