dsh-seed-society

Building a multi-agent collaboration pipeline in DSH usually means manually wiring together the memory layer, an MCP tool bridge, and scattered seed skills. Parameter alignment across layers is fragile, and a single misconfiguration in mnemes autoDream loop can trigger repeated consolidation failures. dsh-seed-society uses the Yogacara eight-vijna framework as its architectural backbone, bundling the auditable dual-loop runtime from seed-society, dsh-mneme memory consolidation tuning (including the reasoningEffort root-cause fix), the mcp__society__* tool bridge, and six seed skills into one configuration-composition package. After installation every layer activates automatically, eliminating the need for per-layer reconciliation. This plugin is intended for teams or individual developers who want structured, auditable multi-agent workflows within DSH. It strictly limits itself to configuration composition and bridging:it does not alter acceptance criteria, does not expand model permissions, and does not bypass approval workflows. The server side holds no secrets and executes no shell commands, while mnemes LLM arbitration and CAS snapshot-hash auditing remain semantically intact.