wbj0256/dsh-agent-handoff-deliberation

wbj0256★ 0TypeScript最后同步: 2026-08-19

在 GitHub 打开

Privacy-first cross-agent handoff and adversarial deliberation tools for the DeepSeek Harness ecosystem.

README 摘要

DSH Agent Handoff & Deliberation Privacy-first coordination primitives for DeepSeek Harness integrations. 1. Cross-agent task handoff : recipient-bound queued → accepted → completed blocked transitions with an audit event trail. 2. Multi-agent deliberation / disagreement : independent positions, direct rebuttals, and a decision gate that refuses false consensus. Independent ecosystem project; it is not an official DeepSeek product . It follows DeepSeek Harness guidance to publish ecosystem projects with the dsh-plugin topic. Scope Included: stateful handoffs, independent positions, rebuttals, decision gates, recursive credential redaction, unit tests, and CI. Explicitly excluded: mutual teaching , scheduling / cron / process spawning / autonomous dispatch , credential storage or transport, and provider-specific model calls. Install Usage Harness adapter apply(host) registers neutral tool definitions against a tiny host.tools.register() seam. This keeps the policy core testable and avoids coupling to unstable internal APIs. A deployment adapter can map these definitions to the installed DeepSeek Harness Cordis and defineTool version. Tools: handoff create , handoff accept , handoff …

在 GitHub 查看完整 README →
Agent/智能体agent-handoffdeepseek-harnessdeliberationdsh-pluginmulti-agentprivacyagent

分类