Skip to content

cloudbase-session.md

Latest commit

 

History

History
37 lines (25 loc) · 2.99 KB

cloudbase-session.md

File metadata and controls

37 lines (25 loc) · 2.99 KB
Raw

CloudBase Plugin Session Context

Use CloudBase guidance only when the current repo, prompt, or tool call makes it relevant.

Core principles

  • MCP-first, CLI-fallback: Prefer CloudBase MCP tools (queryEnv, manageFunctions, manageHosting) when they are available in this session. If MCP is missing or not yet loaded (first session / post-install before restart), configure MCP for the next session and use tcb CLI now (tcb login, domain commands such as tcb fn deploy / tcb hosting deploy — not tcb deploy). See skills/cloudbase/references/tooling-fallback.md. Do not stall waiting for restart; do not invent APIs from memory.
  • Skills on demand: The full catalog stays in skills/ beside this plugin. Hooks load topic-sized chunks via prompt analysis. Prefer local skill files under skills/<name>/SKILL.md. Use searchKnowledgeBase(mode=skill, skillName="<name>") only as a CloudBase docs/knowledge lookup — never HTTP-fetch remote skill markdown into the agent context.
  • Verify, don't trust memory: CloudBase APIs change frequently. Always check current docs via searchKnowledgeBase before implementing.

Scenario routing

CloudBase has 4 scenarios — the detected scenario should guide your primary approach:

  • Web (React/Vue/Vite) → web-development + auth-web-cloudbase + cloudbase-document-database-web-sdk + cloud-storage-web
  • Mini Program (WeChat) → miniprogram-development + auth-wechat-miniprogram + cloudbase-document-database-in-wechat-miniprogram
  • CloudRun (container/Docker) → cloudrun-development + cloud-functions
  • Database (schema design) → data-model-creation + relational-database-mcp-cloudbase or postgresql-development-cloudbase

Mandatory first step

When CloudBase MCP tools are available in this session, resolve the session in this order — every management tool fails with AUTH_REQUIRED / ENV_REQUIRED until steps 1 and 2 are done:

  1. auth({ action: "status" }) — confirm the login; if not logged in, complete auth({ action: "start_auth" }) first.
  2. auth({ action: "set_env", envId: "<EnvId>" }) — bind the target environment when none is bound yet.
  3. queryEnv({ action: "info" }) — then resolve:
    • envId — use in all subsequent config files and code
    • RuntimeMode — "postgresql" or "nosql" (determines database skill routing)
    • RuntimeModeHints.RecommendedSkills — backend-recommended skills for this environment

If MCP tools are not available yet, configure MCP for the next session and resolve envId via tcb CLI (tcb login → tcb env list / tcb env use) per tooling-fallback.md. Skip these steps only when envId is already known from this session.

Platform auth (critical — never mix)

  • Web: CloudBase Web SDK auth.toDefaultLoginPage() — never use OPENID directly
  • Mini Program: wxContext.OPENID (免登录) — never use Web SDK auth
  • CloudRun: Server-side via @cloudbase/node-sdk — verify tokens, never trust client claims blindly