feat(firebase-ai-logic-basics): extract text, multimodal, chat, and streaming capability guides

[AustinBenoit]

Summary (Stack 2/5)

Extracts the four core generation patterns out of the per-platform setup guides into unified cross-platform capability guides under references/sdk/capabilities/ covering Android (Kotlin), iOS (Swift), Flutter (Dart), and Web (JavaScript) side by side.

Changes

• Adds text_generation.md, multimodal_input.md, chat.md, and streaming.md.
• Removes temporary in-file usage sections from android_setup.md and flutter_setup.md (reaching their final setup-only state).
• Updates SKILL.md routing table.

[gemini-code-assist]

Code Review

This pull request reorganizes the Firebase AI Logic SDK documentation by extracting capability-specific examples (such as text generation, multimodal input, chat, and streaming) from platform setup guides into dedicated markdown files. The review feedback highlights several API and syntax errors in the newly created guides across multiple SDKs. Specifically, the Dart SafetySetting constructor is incorrectly documented as accepting three parameters instead of two, the Swift examples incorrectly reference non-existent InlineDataPart and FileDataPart types instead of ModelContent.Part enums, and the Web JavaScript snippets incorrectly include a type property inside Part objects.

[gemini-code-assist]

In the Dart firebase_ai (and underlying google_generative_ai) SDK, the SafetySetting constructor only accepts two positional parameters: category and threshold. There is no third method parameter or HarmBlockMethod type, so this description is incorrect and will mislead developers.

[gemini-code-assist]

Remove the third positional argument (null) from the SafetySetting constructor calls, as the Dart SDK only accepts two positional parameters (category and threshold). Passing a third argument will cause a compilation error.

Suggested change

	// Third positional argument is an optional HarmBlockMethod; pass null.
	final safetySettings = [
	SafetySetting(HarmCategory.harassment, HarmBlockThreshold.high, null),
	SafetySetting(HarmCategory.hateSpeech, HarmBlockThreshold.medium, null),
	];
	final safetySettings = [
	SafetySetting(HarmCategory.harassment, HarmBlockThreshold.high),
	SafetySetting(HarmCategory.hateSpeech, HarmBlockThreshold.medium),
	];

[gemini-code-assist]

In the Swift SDK (FirebaseAILogic), there are no InlineDataPart or FileDataPart types. Instead, you should use the ModelContent.Part.inlineData enum case to wrap raw data with a MIME type.

Suggested change

	let pdf = try InlineDataPart(data: Data(contentsOf: pdfURL), mimeType: "application/pdf")
	// e.g. audio instead: InlineDataPart(data: audioData, mimeType: "audio/mpeg")
	let pdf = ModelContent.Part.inlineData(data: try Data(contentsOf: pdfURL), mimeType: "application/pdf")
	// e.g. audio instead: ModelContent.Part.inlineData(data: audioData, mimeType: "audio/mpeg")

[gemini-code-assist]

In the Swift SDK, use ModelContent.Part.fileData instead of FileDataPart to reference files stored in Cloud Storage for Firebase.

[gemini-code-assist]

In the Firebase AI Web SDK, the Part object for inline data does not contain a type property. It should only contain the inlineData property. Adding type: "inlineData" is incorrect and may cause validation or runtime errors.

Suggested change

	return {
	type: "inlineData",
	inlineData: {
	data: await base64EncodedDataPromise,
	mimeType: file.type,
	},
	};
	return {
	inlineData: {
	data: await base64EncodedDataPromise,
	mimeType: file.type,
	},
	};

[gemini-code-assist]

Remove the type: "inlineData" property from the inline data part object, as it is not supported by the Web SDK's Part type.

Suggested change

	const pdfPart = { type: "inlineData", inlineData: { data: base64Pdf, mimeType: "application/pdf" } };
	const pdfPart = { inlineData: { data: base64Pdf, mimeType: "application/pdf" } };

[gemini-code-assist]

In the Firebase AI Web SDK, the Part object for text does not contain a type property. It should only contain the text property. Adding type: "text" is incorrect and may cause validation or runtime errors.

Suggested change

	{
	role: "user",
	parts: [{ type: "text", text: "Hello, I am a developer." }],
	},
	{
	role: "model",
	parts: [{ type: "text", text: "Great to meet you. How can I help with code?" }],
	},
	history: [
	{
	role: "user",
	parts: [{ text: "Hello, I am a developer." }],
	},
	{
	role: "model",
	parts: [{ text: "Great to meet you. How can I help with code?" }],
	},
	],