maglev-for-dsh

DeepSeek Harness (dsh) delivers powerful agent loops, sandboxed tools, and multi-agent orchestration, yet the process semantics that matter most in real software development - requirement convergence, design stage gates, acceptance evidence, immutable specs - remain entirely absent. Maglev for DSH is a self-contained plugin built to close exactly that gap. Once installed, dsh gains a full pipeline from convergence through design to crystallization, where every iteration is mechanically verified, crystallized, and persisted, so the project no longer depends on one senior engineer or one particular AI stack to remember what has been done.

Three capabilities stand out. First, twenty-nine bundled Maglev skills cover the entire converge-design-verify-crystallize chain and are injected automatically via the bundled provider, so any project can discover and use them with zero local configuration. Second, the Maglev GUI surfaces Crystallization Cards and Reality Cards directly in the conversation flow, making visible what was captured each cycle and where the project currently stands in terms of capability domains, open themes, and contract status. Third, spec completeness checks plus a mechanical pre-crystallization gate turn verification from a soft, model-driven habit into a hard, mechanical constraint that cannot be bypassed.

It is aimed at teams and individual developers who build real software on dsh: people who want AI to write code but also need project knowledge to survive turnover, who value iteration speed yet require every cycle to be traceable and handoff-ready. The repository dogfoods its own methodology end to end, and that living proof is the best specification of the plugins value.