@sayknow-cli/ai 0.3.11 → 0.3.12

package/dist/types/model-retirements.d.ts

@@ -0,0 +1,6 @@
+export declare const RETIRED_MODEL_KEYS: readonly ["google-antigravity/gemini-3.1-pro-high"];
+export declare function isRetiredModelKey(provider: string, modelId: string): boolean;
+export declare function isRetiredModel(model: {
+    provider: string;
+    id: string;
+}): boolean;

package/dist/types/models.d.ts

@@ -1,6 +1,14 @@
-import MODELS from "./models.json";
 import type { Api, KnownProvider, Model, Usage } from "./types";
-export type GeneratedProvider = keyof typeof MODELS;
+/**
+ * Static bundled model registry loaded lazily from `models.json`.
+ *
+ * This module intentionally exposes compile-time defaults only.
+ * It does not include runtime discovery, models.dev overlays, or on-disk cache state.
+ *
+ * For runtime-aware resolution, use `createModelManager()` / `resolveProviderModels()`.
+ */
+type BundledCatalog = typeof import("./models.json");
+export type GeneratedProvider = keyof BundledCatalog;
 export declare function getBundledModel<TApi extends Api = Api>(provider: GeneratedProvider, modelId: string): Model<TApi>;
 export declare function getBundledProviders(): KnownProvider[];
 export declare function getBundledModels(provider: GeneratedProvider): Model<Api>[];
@@ -10,3 +18,4 @@ export declare function calculateCost<TApi extends Api>(model: Model<TApi>, usag
  * Returns false if either model is null or undefined.
  */
 export declare function modelsAreEqual<TApi extends Api>(a: Model<TApi> | null | undefined, b: Model<TApi> | null | undefined): boolean;
+export {};

package/dist/types/providers/google-gemini-headers.d.ts

@@ -3,6 +3,9 @@
  * Uses the same format as the official Gemini CLI (v0.35+):
  * GeminiCLI/VERSION/MODEL (PLATFORM; ARCH; SURFACE)
  */
+export declare const GEMINI_CLI_VERSION_ENV = "SKC_AI_GEMINI_CLI_VERSION";
+export declare const LEGACY_GEMINI_CLI_VERSION_ENV = "PI_AI_GEMINI_CLI_VERSION";
+export declare const DEFAULT_GEMINI_CLI_VERSION = "0.50.0";
 export declare function getGeminiCliUserAgent(modelId?: string): string;
 export declare const getGeminiCliHeaders: (modelId?: string) => {
     "User-Agent": string;

package/dist/types/utils/discovery/antigravity.d.ts

@@ -51,6 +51,12 @@ export interface FetchAntigravityDiscoveryModelsOptions {
     signal?: AbortSignal;
     /** Optional fetch implementation override for tests. */
     fetcher?: typeof fetch;
+    /**
+     * Provider id the caller assigns to returned models. Scopes retired-selector
+     * filtering (e.g. `google-gemini-cli` reuses this helper and remaps rows).
+     * Default: `google-antigravity`.
+     */
+    targetProvider?: "google-antigravity" | "google-gemini-cli";
 }
 /**
  * Fetches discoverable Antigravity models and normalizes them into canonical model entries.

package/package.json

@@ -1,7 +1,7 @@
 {
 	"type": "module",
 	"name": "@sayknow-cli/ai",
-	"version": "0.3.11",
+	"version": "0.3.12",
 	"description": "Unified LLM API with automatic model discovery and provider configuration",
 	"homepage": "https://sayknow-cli.com",
 	"author": "jaybeyond",
@@ -43,7 +43,7 @@
 	"dependencies": {
 		"@anthropic-ai/sdk": "^0.94.0",
 		"@bufbuild/protobuf": "^2.12.0",
-		"@sayknow-cli/utils": "0.3.11",
+		"@sayknow-cli/utils": "0.3.12",
 		"openai": "^6.36.0",
 		"partial-json": "^0.1.7",
 		"zod": "4.4.3"

package/src/model-cache.ts

@@ -3,7 +3,7 @@
  * Replaces per-provider JSON files with a single cache.db.
  */
 import { Database } from "bun:sqlite";
-import { getModelDbPath } from "@sayknow-cli/utils";
+import { getModelDbPath } from "@sayknow-cli/utils/dirs";
 import type { Api, Model } from "./types";
 
 const CACHE_SCHEMA_VERSION = 3;

package/src/model-manager.ts

@@ -1,8 +1,8 @@
 import { readModelCache, writeModelCache } from "./model-cache";
+import { isRetiredModel, isRetiredModelKey } from "./model-retirements";
 import { applyGeneratedModelPolicies, enrichModelThinking } from "./model-thinking";
 import { type GeneratedProvider, getBundledModels } from "./models";
 import type { Api, Model, Provider } from "./types";
-import { isRecord } from "./utils";
 
 const DEFAULT_CACHE_TTL_MS = 2 * 60 * 60 * 1000;
 const NON_AUTHORITATIVE_RETRY_MS = 5 * 60 * 1000;
@@ -85,7 +85,14 @@ function passModelList<TApi extends Api>(value: unknown): Model<TApi>[] {
 	}
 	const out: Model<TApi>[] = [];
 	for (const item of value) {
-		if (item === null || typeof item !== "object" || typeof (item as { id: unknown }).id !== "string") {
+		if (item === null || typeof item !== "object") {
+			continue;
+		}
+		const candidate = item as { id?: unknown; provider?: unknown };
+		if (typeof candidate.id !== "string") {
+			continue;
+		}
+		if (typeof candidate.provider === "string" && isRetiredModelKey(candidate.provider, candidate.id)) {
 			continue;
 		}
 		out.push(enrichModelThinking(item as Model<TApi>));
@@ -382,7 +389,7 @@ function normalizeModelList<TApi extends Api>(value: unknown): Model<TApi>[] {
 	}
 	const models: Model<TApi>[] = [];
 	for (const item of value) {
-		if (isModelLike(item)) {
+		if (isModelLike(item) && !isRetiredModel(item)) {
 			models.push(enrichModelThinking(item as Model<TApi>));
 		}
 	}
@@ -441,6 +448,10 @@ function isModelLike(value: unknown): value is Model<Api> {
 	return true;
 }
 
+function isRecord(value: unknown): value is Record<string, unknown> {
+	return typeof value === "object" && value !== null;
+}
+
 function isModelInputArray(value: unknown): value is ("text" | "image")[] {
 	if (!Array.isArray(value) || value.length === 0) {
 		return false;

package/src/model-retirements.ts

@@ -0,0 +1,13 @@
+// Retired from advertised catalogs because Cloud Code Assist rejects live calls
+// with HTTP 400. The callable high-thinking path is gemini-3.1-pro-low:high.
+export const RETIRED_MODEL_KEYS = ["google-antigravity/gemini-3.1-pro-high"] as const;
+
+const RETIRED_MODEL_KEY_SET = new Set<string>(RETIRED_MODEL_KEYS);
+
+export function isRetiredModelKey(provider: string, modelId: string): boolean {
+	return RETIRED_MODEL_KEY_SET.has(`${provider}/${modelId}`);
+}
+
+export function isRetiredModel(model: { provider: string; id: string }): boolean {
+	return isRetiredModelKey(model.provider, model.id);
+}

package/src/model-thinking.ts

@@ -47,6 +47,7 @@ const DEFAULT_REASONING_EFFORTS_WITH_XHIGH_AND_MAX: readonly Effort[] = [
 const GEMINI_3_PRO_EFFORTS: readonly Effort[] = [Effort.Low, Effort.High];
 const GEMINI_3_FLASH_EFFORTS: readonly Effort[] = [Effort.Minimal, Effort.Low, Effort.Medium, Effort.High];
 const GPT_5_2_PLUS_EFFORTS: readonly Effort[] = [Effort.Low, Effort.Medium, Effort.High, Effort.XHigh];
+const GPT_5_6_PLUS_EFFORTS: readonly Effort[] = [Effort.Low, Effort.Medium, Effort.High, Effort.XHigh, Effort.Max];
 const GPT_5_5_DEFAULT_EFFORT = Effort.XHigh;
 
 const GPT_5_1_CODEX_MINI_EFFORTS: readonly Effort[] = [Effort.Medium, Effort.High];
@@ -60,7 +61,18 @@ type SemVer = {
 
 type GeminiKind = "pro" | "flash";
 type AnthropicKind = "opus" | "sonnet";
-type OpenAIVariant = "base" | "codex" | "codex-max" | "codex-mini" | "codex-spark" | "mini" | "max" | "nano";
+type OpenAIVariant =
+	| "base"
+	| "codex"
+	| "codex-max"
+	| "codex-mini"
+	| "codex-spark"
+	| "luna"
+	| "mini"
+	| "max"
+	| "nano"
+	| "sol"
+	| "terra";
 
 const CODEX_GPT_5_4_PRIORITY_BY_VARIANT: Partial<Record<OpenAIVariant, number>> = {
 	base: 0,
@@ -465,11 +477,25 @@ function applyGpt55ContextWindow(model: ApiModel<Api>, parsedModel: OpenAIModel)
 	}
 	return false;
 }
+const GPT_5_6_TIER_VARIANTS: ReadonlySet<OpenAIVariant> = new Set(["base", "sol", "terra", "luna"]);
+
+function applyGpt56ContextWindow(model: ApiModel<Api>, parsedModel: OpenAIModel): boolean {
+	if (!semverGte(parsedModel.version, "5.6") || !GPT_5_6_TIER_VARIANTS.has(parsedModel.variant)) {
+		return false;
+	}
+	// GPT-5.6 tiers enforce a ~373K usable prompt budget on both transports
+	// (matches the openai-codex live catalog), despite the 1M+ marketing window.
+	model.contextWindow = 373_000;
+	return true;
+}
 
 function applyOpenAICatalogPolicy(model: ApiModel<Api>, parsedModel: OpenAIModel): void {
 	if (applyGpt55ContextWindow(model, parsedModel)) {
 		return;
 	}
+	if (applyGpt56ContextWindow(model, parsedModel)) {
+		return;
+	}
 	// OpenAI code backend models: 400K figure includes output budget; input window is 272K.
 	if (parsedModel.variant.startsWith("codex") && parsedModel.variant !== "codex-spark") {
 		model.contextWindow = 272000;
@@ -582,6 +608,9 @@ function inferOpenAISupportedEfforts(model: OpenAIModel): readonly Effort[] {
 	if (model.variant === "codex-mini" && semverEqual(model.version, "5.1")) {
 		return GPT_5_1_CODEX_MINI_EFFORTS;
 	}
+	if (semverGte(model.version, "5.6")) {
+		return GPT_5_6_PLUS_EFFORTS;
+	}
 	if (semverGte(model.version, "5.2")) {
 		return GPT_5_2_PLUS_EFFORTS;
 	}
@@ -714,7 +743,10 @@ function parseAnthropicModel(modelId: string): AnthropicModel | null {
 }
 
 function parseOpenAIModel(modelId: string): OpenAIModel | null {
-	const match = /gpt-(\d+(?:\.\d+){0,2})(?:-(codex-spark|codex-mini|codex-max|codex|mini|max|nano))?$/.exec(modelId);
+	const match =
+		/gpt-(\d+(?:\.\d+){0,2})(?:-(codex-spark|codex-mini|codex-max|codex|luna|mini|max|nano|sol|terra))?$/.exec(
+			modelId,
+		);
 	if (!match) {
 		return null;
 	}

package/src/models.ts

@@ -1,26 +1,46 @@
+import { readFileSync } from "node:fs";
+import { isRetiredModelKey } from "./model-retirements";
 import { applyGeneratedModelPolicies, enrichModelThinking } from "./model-thinking";
-import MODELS from "./models.json" with { type: "json" };
+// `with { type: "file" }` is embedded by `bun build --compile` and resolves to
+// the bunfs path inside standalone binaries (and to the on-disk path in dev).
+// A plain `createRequire` of a `.json` listed as an extra compile entrypoint is
+// NOT emitted into the bunfs, and its cwd-fallback masks the failure whenever
+// the process runs inside a repo checkout — see PR body for the minimal repro.
+import modelsJsonPath from "./models.json" with { type: "file" };
 import type { Api, KnownProvider, Model, Usage } from "./types";
 import { isClaudeForcedToolChoiceIncapableModelId } from "./utils/tool-choice-capability";
 
 /**
- * Static bundled model registry loaded from `models.json`.
+ * Static bundled model registry loaded lazily from `models.json`.
  *
  * This module intentionally exposes compile-time defaults only.
  * It does not include runtime discovery, models.dev overlays, or on-disk cache state.
  *
  * For runtime-aware resolution, use `createModelManager()` / `resolveProviderModels()`.
  */
-const providerNames = Object.keys(MODELS) as KnownProvider[];
+type BundledCatalog = typeof import("./models.json");
+
+let bundledCatalog: BundledCatalog | undefined;
+let providerNames: KnownProvider[] | undefined;
 const providerModelRegistry: Map<string, Map<string, Model<Api>>> = new Map();
 
+function getBundledCatalog(): BundledCatalog {
+	// TS types a .json import as its contents; at runtime `with { type: "file" }`
+	// yields the file path (bunfs path in compiled binaries, disk path in dev).
+	bundledCatalog ??= JSON.parse(readFileSync(modelsJsonPath as unknown as string, "utf8")) as BundledCatalog;
+	return bundledCatalog;
+}
+
 function getProviderModels(provider: GeneratedProvider): Map<string, Model<Api>> | undefined {
 	const cached = providerModelRegistry.get(provider);
 	if (cached) return cached;
-	const models = MODELS[provider];
+	const models = getBundledCatalog()[provider];
 	if (!models) return undefined;
 	const providerModels = new Map<string, Model<Api>>();
 	for (const [id, model] of Object.entries(models)) {
+		if (isRetiredModelKey(provider, id)) {
+			continue;
+		}
 		providerModels.set(id, applyBundledCompatDefaults(enrichModelThinking(model as Model<Api>)));
 	}
 	providerModelRegistry.set(provider, providerModels);
@@ -52,7 +72,7 @@ function applyBundledCompatDefaults(model: Model<Api>): Model<Api> {
 	return policyModels[0] ?? normalized;
 }
 
-export type GeneratedProvider = keyof typeof MODELS;
+export type GeneratedProvider = keyof BundledCatalog;
 
 export function getBundledModel<TApi extends Api = Api>(provider: GeneratedProvider, modelId: string): Model<TApi> {
 	const providerModels = getProviderModels(provider);
@@ -62,6 +82,7 @@ export function getBundledModel<TApi extends Api = Api>(provider: GeneratedProvi
 export function getBundledProviders(): KnownProvider[] {
 	// Defensive copy: the old eager path returned a fresh Array.from(...), so
 	// callers may freely mutate their result without corrupting enumeration.
+	providerNames ??= Object.keys(getBundledCatalog()) as KnownProvider[];
 	return providerNames.slice();
 }
 

package/src/provider-models/google.ts

@@ -75,6 +75,7 @@ export function googleGeminiCliModelManagerOptions(
 						const models = await fetchAntigravityDiscoveryModels({
 							token,
 							endpoint,
+							targetProvider: "google-gemini-cli",
 						});
 						if (models === null) {
 							return null;

package/src/providers/anthropic.ts

@@ -586,15 +586,12 @@ const ANTHROPIC_BUILTIN_TOOL_NAMES = new Set(["web_search", "code_execution", "t
 export const applyClaudeToolPrefix = (name: string, prefixOverride: string = claudeToolPrefix) => {
 	if (!prefixOverride) return name;
 	if (ANTHROPIC_BUILTIN_TOOL_NAMES.has(name.toLowerCase())) return name;
-	const prefix = prefixOverride.toLowerCase();
-	if (name.toLowerCase().startsWith(prefix)) return name;
 	return `${prefixOverride}${name}`;
 };
 
 export const stripClaudeToolPrefix = (name: string, prefixOverride: string = claudeToolPrefix) => {
 	if (!prefixOverride) return name;
-	const prefix = prefixOverride.toLowerCase();
-	if (!name.toLowerCase().startsWith(prefix)) return name;
+	if (!name.startsWith(prefixOverride)) return name;
 	return name.slice(prefixOverride.length);
 };
 
@@ -1325,6 +1322,25 @@ export const streamAnthropic: StreamFunction<"anthropic-messages"> = (
 				| (ToolCall & { partialJson: string })
 			) & { index: number };
 			const blocks = output.content as Block[];
+			const blocksByAnthropicIndex = new Map<number, Block>();
+			const getBlockByAnthropicIndex = (anthropicIndex: number) => {
+				const block = blocksByAnthropicIndex.get(anthropicIndex);
+				if (!block) return { block: undefined, contentIndex: -1 };
+				return { block, contentIndex: blocks.indexOf(block) };
+			};
+			const trackBlockByAnthropicIndex = (anthropicIndex: number, block: Block) => {
+				// A duplicate start for an active index is a provider-envelope violation;
+				// finalize the orphaned block so no internal stream fields leak into output.
+				const orphaned = blocksByAnthropicIndex.get(anthropicIndex);
+				if (orphaned) {
+					if (orphaned.type === "toolCall" && orphaned.partialJson.trim()) {
+						orphaned.arguments = parseStreamingJson(orphaned.partialJson);
+					}
+					delete (orphaned as { index?: number }).index;
+					delete (orphaned as { partialJson?: string }).partialJson;
+				}
+				blocksByAnthropicIndex.set(anthropicIndex, block);
+			};
 			const idleTimeoutMs = options?.streamIdleTimeoutMs ?? getStreamIdleTimeoutMs();
 			const firstEventTimeoutMs = options?.streamFirstEventTimeoutMs ?? getStreamFirstEventTimeoutMs(idleTimeoutMs);
 			stream.push({ type: "start", partial: output });
@@ -1334,6 +1350,8 @@ export const streamAnthropic: StreamFunction<"anthropic-messages"> = (
 			let providerRetryAttempt = 0;
 			let thinkingRepairAttempted = false;
 			while (true) {
+				// Retries reset output.content; drop stale block correlations from the aborted attempt.
+				blocksByAnthropicIndex.clear();
 				activeAbortTracker = createAbortSourceTracker(options?.signal);
 				const firstEventTimeoutAbortError = new Error(
 					"Anthropic stream timed out while waiting for the first event",
@@ -1403,6 +1421,7 @@ export const streamAnthropic: StreamFunction<"anthropic-messages"> = (
 									index: event.index,
 								};
 								output.content.push(block);
+								trackBlockByAnthropicIndex(event.index, block);
 								stream.push({
 									type: "text_start",
 									contentIndex: output.content.length - 1,
@@ -1416,6 +1435,7 @@ export const streamAnthropic: StreamFunction<"anthropic-messages"> = (
 									index: event.index,
 								};
 								output.content.push(block);
+								trackBlockByAnthropicIndex(event.index, block);
 								stream.push({
 									type: "thinking_start",
 									contentIndex: output.content.length - 1,
@@ -1428,6 +1448,7 @@ export const streamAnthropic: StreamFunction<"anthropic-messages"> = (
 									index: event.index,
 								};
 								output.content.push(block);
+								trackBlockByAnthropicIndex(event.index, block);
 							} else if (event.content_block.type === "tool_use") {
 								streamedReplayUnsafeContent = true;
 								const block: Block = {
@@ -1441,6 +1462,7 @@ export const streamAnthropic: StreamFunction<"anthropic-messages"> = (
 									index: event.index,
 								};
 								output.content.push(block);
+								trackBlockByAnthropicIndex(event.index, block);
 								stream.push({
 									type: "toolcall_start",
 									contentIndex: output.content.length - 1,
@@ -1449,8 +1471,7 @@ export const streamAnthropic: StreamFunction<"anthropic-messages"> = (
 							}
 						} else if (event.type === "content_block_delta") {
 							if (event.delta.type === "text_delta") {
-								const index = blocks.findIndex(b => b.index === event.index);
-								const block = blocks[index];
+								const { block, contentIndex: index } = getBlockByAnthropicIndex(event.index);
 								if (block && block.type === "text") {
 									block.text += event.delta.text;
 									stream.push({
@@ -1461,8 +1482,7 @@ export const streamAnthropic: StreamFunction<"anthropic-messages"> = (
 									});
 								}
 							} else if (event.delta.type === "thinking_delta") {
-								const index = blocks.findIndex(b => b.index === event.index);
-								const block = blocks[index];
+								const { block, contentIndex: index } = getBlockByAnthropicIndex(event.index);
 								if (block && block.type === "thinking") {
 									block.thinking += event.delta.thinking;
 									stream.push({
@@ -1473,8 +1493,7 @@ export const streamAnthropic: StreamFunction<"anthropic-messages"> = (
 									});
 								}
 							} else if (event.delta.type === "input_json_delta") {
-								const index = blocks.findIndex(b => b.index === event.index);
-								const block = blocks[index];
+								const { block, contentIndex: index } = getBlockByAnthropicIndex(event.index);
 								if (block && block.type === "toolCall") {
 									block.partialJson += event.delta.partial_json;
 									block.arguments = parseStreamingJson(block.partialJson);
@@ -1486,17 +1505,16 @@ export const streamAnthropic: StreamFunction<"anthropic-messages"> = (
 									});
 								}
 							} else if (event.delta.type === "signature_delta") {
-								const index = blocks.findIndex(b => b.index === event.index);
-								const block = blocks[index];
+								const { block } = getBlockByAnthropicIndex(event.index);
 								if (block && block.type === "thinking") {
 									block.thinkingSignature = block.thinkingSignature || "";
 									block.thinkingSignature += event.delta.signature;
 								}
 							}
 						} else if (event.type === "content_block_stop") {
-							const index = blocks.findIndex(b => b.index === event.index);
-							const block = blocks[index];
+							const { block, contentIndex: index } = getBlockByAnthropicIndex(event.index);
 							if (block) {
+								blocksByAnthropicIndex.delete(event.index);
 								delete (block as { index?: number }).index;
 								if (block.type === "text") {
 									stream.push({
@@ -1513,7 +1531,9 @@ export const streamAnthropic: StreamFunction<"anthropic-messages"> = (
 										partial: output,
 									});
 								} else if (block.type === "toolCall") {
-									block.arguments = parseStreamingJson(block.partialJson);
+									if (block.partialJson.trim()) {
+										block.arguments = parseStreamingJson(block.partialJson);
+									}
 									delete (block as { partialJson?: string }).partialJson;
 									stream.push({
 										type: "toolcall_end",
@@ -2243,7 +2263,11 @@ function buildParams(
 		params.tools = convertTools(
 			context.tools,
 			isOAuthToken,
-			disableStrictTools || model.provider === "github-copilot",
+			// The Claude Code OAuth surface mishandles `strict: true` tools:
+			// streamed tool_use blocks arrive with empty/undefined arguments and
+			// occasionally corrupted names (works with PI_NO_STRICT=1). Never
+			// request strict tool use on OAuth requests.
+			disableStrictTools || isOAuthToken || model.provider === "github-copilot",
 			getAnthropicCompat(model).supportsEagerToolInputStreaming,
 		);
 	}

package/src/providers/google-gemini-headers.ts

@@ -3,8 +3,13 @@
  * Uses the same format as the official Gemini CLI (v0.35+):
  * GeminiCLI/VERSION/MODEL (PLATFORM; ARCH; SURFACE)
  */
+export const GEMINI_CLI_VERSION_ENV = "SKC_AI_GEMINI_CLI_VERSION";
+export const LEGACY_GEMINI_CLI_VERSION_ENV = "PI_AI_GEMINI_CLI_VERSION";
+export const DEFAULT_GEMINI_CLI_VERSION = "0.50.0";
+
 export function getGeminiCliUserAgent(modelId = "gemini-3.1-pro-preview"): string {
-	const version = process.env.SKC_AI_GEMINI_CLI_VERSION || process.env.PI_AI_GEMINI_CLI_VERSION || "0.49.0";
+	const version =
+		process.env[GEMINI_CLI_VERSION_ENV] || process.env[LEGACY_GEMINI_CLI_VERSION_ENV] || DEFAULT_GEMINI_CLI_VERSION;
 	const platform = process.platform === "win32" ? "win32" : process.platform;
 	const arch = process.arch === "x64" ? "x64" : process.arch;
 	return `GeminiCLI/${version}/${modelId} (${platform}; ${arch}; terminal)`;

package/src/utils/discovery/antigravity.ts

@@ -1,7 +1,7 @@
 import * as z from "zod/v4";
+import { isRetiredModelKey } from "../../model-retirements";
 import { getAntigravityUserAgent } from "../../providers/google-gemini-headers";
 import type { Model } from "../../types";
-import { toPositiveNumber } from "../../utils";
 
 const DEFAULT_ANTIGRAVITY_DISCOVERY_ENDPOINTS = [
 	"https://daily-cloudcode-pa.googleapis.com",
@@ -162,6 +162,12 @@ export interface FetchAntigravityDiscoveryModelsOptions {
 	signal?: AbortSignal;
 	/** Optional fetch implementation override for tests. */
 	fetcher?: typeof fetch;
+	/**
+	 * Provider id the caller assigns to returned models. Scopes retired-selector
+	 * filtering (e.g. `google-gemini-cli` reuses this helper and remaps rows).
+	 * Default: `google-antigravity`.
+	 */
+	targetProvider?: "google-antigravity" | "google-gemini-cli";
 }
 
 /**
@@ -174,6 +180,7 @@ export async function fetchAntigravityDiscoveryModels(
 	options: FetchAntigravityDiscoveryModelsOptions,
 ): Promise<Model<"google-gemini-cli">[] | null> {
 	const fetcher = options.fetcher ?? fetch;
+	const targetProvider = options.targetProvider ?? "google-antigravity";
 	const endpoints = options.endpoint
 		? [trimTrailingSlashes(options.endpoint)]
 		: DEFAULT_ANTIGRAVITY_DISCOVERY_ENDPOINTS.map(trimTrailingSlashes);
@@ -214,7 +221,7 @@ export async function fetchAntigravityDiscoveryModels(
 		const models: Model<"google-gemini-cli">[] = [];
 
 		for (const [modelId, model] of Object.entries(parsed.models ?? {})) {
-			if (ANTIGRAVITY_DISCOVERY_DENYLIST.has(modelId)) {
+			if (ANTIGRAVITY_DISCOVERY_DENYLIST.has(modelId) || isRetiredModelKey(targetProvider, modelId)) {
 				continue;
 			}
 			if (model.isInternal === true) {
@@ -256,6 +263,13 @@ function parseAntigravityDiscoveryResponse(value: unknown): AntigravityDiscovery
 	return parsed.data;
 }
 
+function toPositiveNumber(value: unknown, fallback: number): number {
+	if (typeof value !== "number" || !Number.isFinite(value) || value <= 0) {
+		return fallback;
+	}
+	return value;
+}
+
 function trimTrailingSlashes(value: string): string {
 	return value.replace(/\/+$/, "");
 }