Add a proactive "verify against official docs" step before code generation

[debbyliu0206]

What is your feature suggestion?

Add a proactive "verify against official docs" step in the workflow and phase-specific skills (especially google-agents-cli-workflow, google-agents-cli-deploy, and google-agents-cli-adk-code) that instructs the coding agent to fetch and read relevant official documentation before writing code or deployment configs, rather than relying on its training data for version-sensitive details.

Currently, the only doc-lookup guidance lives in the Troubleshooting section of google-agents-cli-workflow Principle 2 ("Use WebFetch on URLs from the ADK docs index curl https://adk.dev/llms.txt for deep dives"), which is reactive — it tells the agent what to do after something breaks.

Proposed change: Add a doc-verification gate at two levels:

1. In google-agents-cli-workflow Phase 3 (Build), before writing or modifying agent code:

   Before writing any code in this phase, fetch the relevant official doc page (e.g., from https://adk.dev/) for any API, model name, SDK method, or deployment target you plan to use. Verify that the syntax, parameter names, and specs match the current documentation. Do not rely on training data for version-sensitive details.

2. At the top of google-agents-cli-deploy and google-agents-cli-adk-code, a similar instruction:

   Before writing any code or configuration in this phase, fetch and read the relevant official documentation to confirm that the APIs, model names, and deployment specs you plan to use are current.

This is consistent with the existing pattern where the workflow skill already says "Re-read the relevant skill before each phase — not after you've already started and hit a problem." The same principle should extend to official docs.

What will this enable you to do?

This would significantly reduce the debug cycle time when building and deploying ADK agents.

Current failure pattern I'm repeatedly hitting:

1. Coding agent generates code or deploy config using outdated syntax from its training data (e.g., deprecated model names, changed API parameters, old Terraform resource attributes)
2. Code fails at runtime or deployment
3. During debugging, the agent itself identifies the root cause: "the API spec has been updated" or "this syntax was changed in a newer version"
4. Agent then fetches the official docs, finds the correct current usage, and fixes the code

Steps 3–4 happen reactively and repeatedly. The debug loop with these skills is slow, and these errors are entirely preventable if the agent checks docs upfront.

Since every coding agent (Gemini CLI, Claude Code, Codex, Antigravity) has the same stale-training-data problem, the skills layer is the ideal place to mitigate it — it's agent-agnostic and the fix is just a few lines of instruction text in the skill markdown files.

Additional context

• agents-cli version: 0.5.1
• Coding agents tested: Claude Code, Antigravity
• OS: Windows (ARM64)

The fix is lightweight — it's a few lines of instruction text in the existing skill markdown files, not a code change. Happy to provide specific examples of APIs/model names that caused failures if that would be helpful.