DeepSeek Harness Plugins · Workflow
Plugins that extend DeepSeek Harness (DSH) with vision, workflows, memory, web tools and more.
Adds mapped skills, custom commands, and quick buttons to the / menu by scanning SKILL.md directories; pure incremental with no official plugins replaced.
A concurrent browser slot-pool plugin for DSH mcp-client: fixed CDP ports, atomic locking, crash self-healing, and strict session isolation.
A DSH port of obra/superpowers: 14 agentic skills that enforce a structured coding workflow from brainstorming and planning through TDD, systematic debugging, and code review.
Replaces repetitive manual confirmations with an independent reviewer model while retaining sandbox and human approval, letting the agent propose safer alternatives when denied.
Exposes the DSH Web server to the local network via transparent TCP forwarding on 0.0.0.0:3081 by default and registers LAN addresses in DSH trustedHosts.
Reproduces Codex Guardian and Grok escalation approval policies inside DSH, automatically reviewing tool calls via LLM adapters without unattended human prompts.
An evidence-backed delivery governance layer for AI coding agents, covering stage gates, risk tracking, review separation, and release readiness across multiple agent platforms.
A temporary DSH adapter that bridges image input to text-only models by transcribing images via Codex Luna before routing to configurable downstream providers.
A zero-dependency prompt plugin that routes image recognition to a mimo-v2.5 subagent, falling back to read_image on failure.
Enhances DSH Web approval by adding an annotated-reject that steers the model and a hard stop on plain reject, with zero dependencies.
A structured code review workflow for DeepSeek Harness, covering correctness, security, performance, style, and testing.
Registers the native Playwright CLI as a DSH passthrough tool so agents can invoke local playwright commands directly without subcommand拆分 or argument remapping.