Google Developer Knowledge — Official Google Docs Skill

Use this skill when the task needs current official Google developer documentation for Google Cloud, Firebase, Android, Google AI, Gemini CLI, Flutter, Go, Maps, Web, TensorFlow, or other products indexed by the Developer Knowledge corpus.

Core Rule

Use mcporter capability discovery from references/mcporter-workflow.md. Select only a healthy server exposing Google Developer Knowledge search, full-document retrieval, and grounded-answer capabilities. Do not assume a server alias, URL, tool-name prefix, authentication arrangement, or native tool wrapper.

Tool Choice

1. Start with search_documents for most research. It returns official documentation chunks with parent, id, and content.
2. Use get_documents with returned parent document names when snippets are insufficient or citations need full-page context.
3. Use answer_query for a quick grounded synthesis only when its references are sufficient. If it returns 429, times out, or lacks adequate references, fall back to search_documents plus get_documents.

Decision Checkpoints

1. 🔴 CHECKPOINT — Capability discovery: Parse the selected route returned by discovery into gdk_server, gdk_search_tool, gdk_get_tool, and gdk_answer_tool. If discovery finds no compatible server, record the reason and use REST only with DEVELOPERKNOWLEDGE_API_KEY or OAuth credentials plus a quota project; otherwise use an official-document fallback.
2. 🔴 CHECKPOINT — Evidence sufficiency: For version-sensitive, pricing, quota, security, or migration claims, retrieve the relevant full document with get_documents. A search snippet or an answer_query response alone is not enough.
3. 🔴 CHECKPOINT — Retrieval budget: Retrieve no more than three parent documents by default. Ask before broad extraction, offline use, or a larger document set.
4. 🔴 CHECKPOINT — Output trace: State the selected server or fallback, the retrieval method, and source URLs or document URIs. If the needed detail is absent, mark it unverified rather than infer it.

Quick Commands

# Run capability discovery from references/mcporter-workflow.md first.
gdk_route='<selected-route-json>'
gdk_server="$(jq -r '.server' <<<"$gdk_route")"
gdk_search_tool="$(jq -r '.tools.search' <<<"$gdk_route")"
gdk_get_tool="$(jq -r '.tools.get' <<<"$gdk_route")"
gdk_answer_tool="$(jq -r '.tools.answer' <<<"$gdk_route")"

mcporter call "$gdk_server.$gdk_search_tool" \
  --args '{"query":"Cloud Run minimum instances concurrency pricing"}' \
  --output json

mcporter call "$gdk_server.$gdk_get_tool" \
  --args '{"names":["documents/docs.cloud.google.com/run/docs/about-concurrency"]}' \
  --output json

mcporter call "$gdk_server.$gdk_answer_tool" \
  --args '{"query":"Compare Cloud Run and Cloud Functions for a new HTTP microservice."}' \
  --output json

Workflow

1. Define one focused documentation question.
2. Pass Capability discovery and retain the returned route and tool names for the task.
3. Search with search_documents.
4. Pull the best parent documents with get_documents when evidence requires full context.
5. Pass Evidence sufficiency, then synthesize only from retrieved evidence.
6. Pass Retrieval budget and Output trace before responding.

Coverage and Limits

• Corpus covers public pages from Google developer domains such as cloud.google.com, docs.cloud.google.com, firebase.google.com, developer.android.com, ai.google.dev, developers.google.com, web.dev, go.dev, docs.flutter.dev, and www.tensorflow.org.
• Some generated API reference prefixes are excluded from the corpus, especially language-specific Google Cloud reference directories.
• Data freshness target is usually 24-48 hours after documentation publication.
• Full documents are Markdown and can be large; retrieve only the pages needed for the answer.

Detailed Reference

Use references/mcporter-workflow.md for command patterns, field shapes, troubleshooting, and source-backed answer patterns.

Use references/api-fallback.md only after capability discovery fails, when validating a client integration, or when the task specifically asks for direct Developer Knowledge REST API calls. Never attempt REST without an API key or OAuth credentials plus a quota project.

Retrieval Record

For answers that depend on current Google behavior, include a concise record:

• method: mcp, rest-api-key, rest-oauth, or official-document.
• server: selected MCP server, or none when fallback was used.
• fallback_reason: none, discovery failure, rate limit, authentication failure, or corpus gap.
• sources: official URLs or document URIs used.
• verification: full-document, search-snippet-only, or unverified.

Anti-Patterns

Official Documents

These are the official documents this skill was built from:

• https://developers.google.com/knowledge/api
• https://developers.google.com/knowledge/quickstart
• https://developers.google.com/knowledge/howto
• https://developers.google.com/knowledge/answer-query
• https://developers.google.com/knowledge/mcp
• https://developers.google.com/knowledge/reference/mcp
• https://developers.google.com/knowledge/reference/mcp/tools_list/search_documents
• https://developers.google.com/knowledge/reference/mcp/tools_list/answer_query
• https://developers.google.com/knowledge/reference/mcp/tools_list/get_documents
• https://developers.google.com/knowledge/reference/corpus-reference