Перейти к основному содержимому

Перевод: оболочка страницы на русском; тело навыка (SKILL.md) пока на английском — это исходные инструкции агента.

Cursor Sdk Bridge

Map Cursor SDK agents to VibeOS API and ACP.

Метаданные навыка​

ИсточникВстроенный (установлен по умолчанию)
Путьskills/software-development/cursor-sdk-bridge
Версия1.0.0
АвторVibeOS
ЛицензияMIT
Платформыlinux, macos, windows
Тегиcursor, sdk, api, acp, automation, openai
Связанные навыкиvibeos-api-automation, cursor-capabilities-bridge, analytical-ui, open-webui-vibeos

Справка: полный SKILL.md​

к сведению

Ниже полное определение навыка, которое VibeOS загружает при активации. Это инструкции, которые видит агент.

Cursor SDK Bridge

Cursor’s Agent SDK (@cursor/sdk / cursor-sdk) runs agents from scripts or CI (local cwd or Cursor cloud VM). VibeOS does not clone Cursor’s cloud VM. Map the same jobs onto the OpenAI-compatible API server + ACP.

When to Use​

  • User asks for Cursor SDK / Agent.create / programmatic agents in VibeOS
  • Scripts, bots, or Open WebUI need to drive an agent over HTTP
  • Choosing between API (remote-friendly) and ACP (IDE-embedded)

Prerequisites​

# ~/.vibeos/.env  (secrets only)
API_SERVER_ENABLED=true
API_SERVER_KEY=change-me-local-dev

Gateway running (vibeos gateway / vibeos gateway). Docs: website/docs/user-guide/features/api-server.md. Skill vibeos-api-automation has the full CLI cheat sheet.

How to Run​

Local / scripted agent (≈ Cursor SDK local runtime)​

vibeos api health
vibeos api chat "Summarize AGENTS.md in 3 bullets"
vibeos api run "Investigate flaky test X" --wait
RUN=$(vibeos api run "…" | jq -r .run_id)
vibeos api stream "$RUN"
vibeos api stop "$RUN"

HTTP (OpenAI-compatible):

curl -s http://127.0.0.1:8642/v1/chat/completions \
-H "Authorization: Bearer $API_SERVER_KEY" \
-H "Content-Type: application/json" \
-d '{"model":"default","messages":[{"role":"user","content":"hi"}]}'

Long runs: POST /v1/runs + events stream (see API docs). Summary UI: vibeos api runs --summary / skill analytical-ui.

IDE-embedded agent (≈ Cursor in-editor agent)​

./scripts/run-coding-acp.sh --status
./scripts/run-coding-acp.sh # VIBEOS_HOME=…/profiles/coding

ACP is the IDE bridge — not a second chat product. Use coding profile policy.

What we deliberately do not clone​

Cursor SDKVibeOS
Cloud VM + auto-clone repoNo — use your VPS/Timeweb + local/git cwd
Agent.create cloud runtimeAPI server on your machine/host
Marketplace Bugbot binaryskill cursor-review-bridge
Cursor-hosted secrets vault~/.vibeos/.env only

Quick Reference​

Cursor SDK ideaVibeOS
Agent.prompt / syncvibeos api chat or /v1/chat/completions
Agent + streamvibeos api stream / SSE on runs
Resume sessionvibeos api session-create + prompt --session
Cancelvibeos api stop <run_id>
List runsvibeos api runs / --summary
IDE agentACP + run-coding-acp.sh
Open WebUI front-endskill open-webui-vibeos

Procedure​

  1. Enable API server + key in .env; start gateway.
  2. Smoke vibeos api health.
  3. Pick sync (chat) vs async (run + stream).
  4. For IDE work, use ACP coding profile instead of spawning a second UI.
  5. New HTTP clients: read /v1/capabilities before assuming endpoints.

Pitfalls​

  • Expecting Cursor cloud VMs — not shipped; host your own runner
  • Leaving API_SERVER_KEY empty or committed to git
  • Mixing ACP session and API session IDs — different stores
  • Enabling GitHub Actions as the only ship path — Timeweb is the online path

Verification​

vibeos api health
vibeos api chat "ping"
./scripts/run-coding-acp.sh --check