hezhongtang/dsh-capability-optimizer

README 摘要

dsh-capability-optimizer External-expert consultation for DeepSeek Harness: the agent headlessly invokes the Claude Code CLI with role personas — advisor, reviewer, designer, or your own — and weighs the subscription model's replies as reference answers. English 中文 Why this exists A single harness has one perspective. When the DSH agent hits a decision point, finishes risky work, or starts significant new code, a second opinion from a different model — Claude, on your existing Claude Code subscription — is cheap insurance and a genuine quality lift. This plugin makes that a first-class tool call instead of a copy-paste detour: the agent consults, Claude answers in role, and the reply is framed as advice to weigh, not obey (the same contract Oh My Pi's advisors use). Phase 1 speaks only to the Claude Code CLI. The settings schema (v2, one workspace per harness), the UI tab catalog, and the runner seam are already multi-backend: codex , zcode , kimi-code , pi , opencode , and omp each land later as a runner behind the same three tools. Features 🎭 Role personas Built-in advisor / reviewer / designer, or your own (name, prompt, dedicated model, dedicated fallback, dedicated thinking e…

分类