DeepSeek Harness Plugins · Workflow
Plugins that extend DeepSeek Harness (DSH) with vision, workflows, memory, web tools and more.
Automatically parses bid documents to extract rejection clauses, checks for risks, typos, and logical inconsistencies, then generates a report in PDF, HTML, or Markdown.
A DeepSeek Harness plugin for Terraform IaC management, supporting plan/apply gating, state inspection, output queries, and resource validation with read-only-first safety.
A lawyer-style contract-drafting workflow for DeepSeek Harness, covering deal analysis, clause gap discovery, five-way parallel review, iterative revision, and delivery of a complete auditable contract package for mainland China practice.
Windows desktop automation MCP plugin for DSH, bundling a Computer Use runtime with window observation, accessibility-based actions, result verification, and high-risk action confirmation.
Interactive task-level execution graph for DSH, aggregating Agent, tool, code-edit, and test nodes with snake, DAG, and timeline layouts, live status tracking, and bidirectional Trajectory linking.
Generic per-turn context injection for DSH Web profiles: manage prompt lists in settings and inject each enabled prompt into every conversation round as a visible notice line, reminding without executing.
Compare the workspace instruction sources that DeepSeek Harness, Codex, and Claude Code each load for the same repository.
Integrates Jira and GitLab boards into DeepSeek Harness so the agent can drive kanban operations via tools, with GitLab branch switching and in-session analysis.
Automated presentation generation for DeepSeek Harness that turns outlines and illustrations into editable .pptx, offline HTML slides, and a structured manifest with ten built-in themes and template support.
Natively reimplements the Hermes adaptive intelligence layer for DeepSeek Harness, providing persistent memory, skill management, and RL trajectory collection with compression in pure JavaScript.
Highlights the taskbar icon, switches to the right session, and pops the window when an agent asks a question, requests approval, or a task completes.
Forces subagent model and reasoning effort through a configurable per-route allowlist, and steers a running child at its next step instead of after its full turn.