CloudBase Development Guidelines

What It Solves

In CloudBase projects, failures are often not caused by a missing frontend snippet, but by routing the wrong scenario: Web login handled like a mini program flow, native apps routed through Web auth, AI calls debugged only at the model parameter, or management actions attempted before MCP availability is confirmed. This skill turns CloudBase development into a set of routing rules: identify the platform, auth domain, resource type, and call boundary before opening the matching capability docs.

How It Works

• Identify the scenario first, then read the matching skill: use stable identifiers such as auth-tool, auth-web, http-api, cloud-functions, cloudrun-development, and ai-model-web, rather than relying on repository paths or generated artifacts.
• MCP / mcporter first: before querying environments, deploying, or managing databases and functions, confirm CloudBase MCP is available and inspect tool schemas before execution.
• Explicit EnvId: code, configuration, and command examples should name the EnvId directly; aliases or short names should be resolved to the canonical full environment ID first.
• Respect platform boundaries: Web login uses auth-tool / auth-web, WeChat mini programs use miniprogram-development, native apps use http-api, and NoSQL vs. MySQL follow their own SDK or OpenAPI paths.
• AI calls need eligibility checks: createModel(...) accepts a GroupName, while the concrete model ID belongs in the model field of generateText / streamText; check resource-package and growth-plan status before code changes.
• Verify with evidence: avoid masking type issues with any, require static checks and builds, and add runtime verification when routing, forms, auth, or async UI is affected.

Boundaries and Caveats

Use this for development and troubleshooting across CloudBase Web, mini programs, native apps, cloud functions, CloudRun, AI agents, and AI model integration. It does not replace official SDK docs, and it is not a substitute for clarifying the platform, environment, or MCP setup when those are ambiguous. If the same implementation path fails two or three times, stop patching and re-check routing, auth domain, permission model, and SDK boundary.