dsh-bridge

Switching from Claude Code, Codex CLI, OpenCode, or Jcode to DeepSeek Harness leaves a familiar gap: the slash commands you lean on every day simply are not there. dsh-bridge ports seventeen of those high-frequency commands—/model, /resume, /memory, /mcp, /review, and others—onto DSH-native seams so they behave exactly the way your previous harness taught you. The entire command surface is English-first, smoothing over a plugin ecosystem that otherwise skews heavily toward non-English locales.

Trust is the second problem the plugin tackles head-on. Every community plugin recommendation ships with an adversarial review report card that cites file-and-line evidence you can re-check yourself. The audit is performed by a model that did not write the artifact under review, grading it on a five-band scale from A to F. Twenty-four plugins have been graded so far, and you can filter the verified catalog with /browse, inspect any report card via /trust, and install with /install only after explicitly acknowledging the risk tier. No plugin is ever suggested before it has cleared that pipeline.

This is built for developers migrating from another AI coding tool who want zero-friction onboarding, teams that treat plugin supply-chain security as a first-class concern and demand verifiable evidence behind every safety claim, and anyone who prefers an English command interface in a predominantly non-English catalog. The plugin does not bundle a model; bring your own provider endpoint and API key, and the rest boots cleanly.