DeepSeek Harness Plugins · Workflow
Plugins that extend DeepSeek Harness (DSH) with vision, workflows, memory, web tools and more.
Benchmark-driven self-evolution plugin for Agent Profiles: changes are accepted only when strictly better than the current version, with precise rollback otherwise.
Watches agent execution traces, mines recurring workflows, and JIT-compiles them into reusable skills or flows with hot-reload into dsh.
Cross-session horizontal messaging for DSH: any two local sessions can discover and message each other, daemon-free.
Bridges Pi Agent skills into DeepSeek Harness with zero migration, auto-injecting them as available skills.
Run tests in a single tool call for the agent: auto-detect the framework, execute tests, and return a structured pass/fail summary to save tokens.
Bridges OpenAI Codex skills, instructions, and configuration into DeepSeek Harness with zero migration.
Bidirectional conversation migration between DSH and Codex, Claude Code, Pi, Reasonix, or OpenCode with safe dry-run previews and batch import/export.
Adds a scheduled polling workspace to DeepSeek Harness, automating recurring tasks like checking folders, fetching data, and generating reports in isolated sessions without interrupting active conversations.
A global MCP service manager for DSH that lets you add, toggle, and remove MCP servers from a settings panel, with JSON batch import and real-time connection status display.
Spawns a read-only verifier subagent before each turn closes, collects a structured verdict, and steers non-pass gaps back into the driving agent.
Provides auditable, bounded source-retrieval workflows for DeepSeek Harness agents, making source scope, retained evidence, deterministic JSON selection, and unresolved gaps fully visible.
Multi-agent task orchestration for DSH: dependency-based wave planning, parallel git-worktree-isolated lanes, automatic review and merge, durable state, and a real-time dashboard.