AI Decision System
Replayable, offline-verifiable records for AI-assisted go/no-go decisions: experiment promotions, typed model answers and human-approved actions.
翻譯準備中
尚未偵測到更新。GitHub 發布版本後會顯示在這裡。
個人 GitHub 帳號下公開儲存庫的擁有者可以登入並驗證所有權,自行管理介紹,無需管理員核准。
登入並產生驗證檔案,在 30 分鐘內將其提交至此儲存庫的預設分支。其他分支或分叉儲存庫中的檔案無效。然後在下方檢查檔案,即可無需管理員核准直接管理專案。請保留檔案以維持管理權限。
如需登入,將前往登入頁面。登入後請回到此處,再次點擊所選的驗證按鈕。
在專案資料夾中執行 /nomorevibe verify,即可驗證所有權並自行管理介紹。
一分鐘完成驗證Discover and install DeepSeek Harness plugins, skills, MCP servers, agent presets, bundles, and profiles.
翻譯準備中
取自儲存庫 README
正在載入完整 README。
檢視完整 README ↗DeepSeek Harness is built on "everything is a plugin" but ships no registry — its
README asks authors to tag repositories with the dsh-plugin topic and leaves
discovery there. dsh.fish turns that topic into a searchable, multilingual
catalog with a shared, machine-executable install plan.
GET /api/v1/scoring, not buried in a blog post.rising sort that surfaces what is gaining stars this week.dsh-fish rate, hub_rate) where the rater actually used the artifact. The web renders the average, the distribution and every comment read-only, and republishes the aggregate as aggregateRating structured data.Browse the catalog at dsh.fish — search, filter by kind and category, read the install plan, copy the command.
Terminal — the CLI applies the plan for you:
npx @dsh-fish/cli add <artifact-id>
add / find / list / remove / update match the skills CLI
vocabulary and actually write skills, presets, bundles and profiles into
$DSH_HOME.
Inside the harness — add the hub plugin once:
dsh plugin --profile web add @dsh-fish/hub
It registers hub_search, hub_show, hub_install, hub_list, hub_remove,
hub_update and hub_account, so an agent can discover and install artifacts
without leaving the session. Signing in uses the OAuth device flow.
<profile> is whichever profile the harness boots. A desktop build has its own:
Local DSH runs local-dsh, so use its
bundled launcher and name that profile.
dsh plugin --profile local-dsh add @dsh-fish/hub
Publishing a plugin? Tag your repository with the dsh-plugin topic. The
hourly crawl inspects its package.json, SKILL.md or agent.cordis.yml and
classifies what the harness can actually load. See
docs/project/architecture.md
for the probe order.
| Kind | What it is | How it installs |
|---|---|---|
| Bundle | npm package declaring dsh.bundle.patch | dsh plugin --profile <p> add <spec> |
| Profile | ordered dsh.profile.bundles stack | one add per bundle, in order |
| Skill | SKILL.md bundle or flat Markdown | files written under $DSH_HOME/skills |
| Agent preset | directory holding one agent.cordis.yml | written to $DSH_HOME/.agent-presets/<id> |
backend/ Domain-Driven Design: domain, application, infrastructure, interfaces
frontend/ Feature-Sliced Design: app, pages, widgets, features, entities, shared
packages/
dsh-plugin-hub/ `@dsh-fish/hub` — the bundle users install into their harness
dsh-cli/ `@dsh-fish/cli` — `npx @dsh-fish/cli add <id>`
docs/ architecture, layer conventions, operations, ADRs
Both halves deploy as one Cloudflare Worker: Hono at /api/*, React Router
SSR everywhere else, D1 for the catalog, KV for sessions, and a Cron Trigger
that re-crawls every hour.
pnpm install
pnpm run dev # http://localhost:5173
Quality gates: pnpm run typecheck && pnpm run test && pnpm run test:e2e && pnpm run build.
Deployment, bindings and secrets: docs/operations/deployment.md.
Conventions and architecture: AGENTS.md, docs/project/architecture.md, docs/decisions/.