jeo-code 0.12.2 → 0.12.4

package/CHANGELOG.md

@@ -9,6 +9,32 @@ The README mirrors the latest 5 entries — regenerate with `bun run changelog:s
 
 ## [Unreleased]
 
+## [0.12.4] - 2026-10-17
+_Refresh Antigravity provider (`antigravity/*`) account catalogs with `antigravity/hub/2.23.1` so entitled Gemini 4 Argon and Gemini 3.8 Flash tiers appear directly in `/model provider antigravity`._
+
+### Fixed
+- **Antigravity provider client identity & Argon catalog visibility** (`src/ai/providers/antigravity.ts`, `src/ai/model-discovery.ts`) — updated `getAntigravityUserAgent()` from `antigravity/1.104.0` to `antigravity/hub/2.23.1` (`jeopi` v16.6.3 parity) so `daily-cloudcode-pa.googleapis.com/v1internal:fetchAvailableModels` returns entitled `gemini-4-argon-{low,medium,high}` and `gemini-3.8-flash-*` tiers. Preserved named non-recommended chat models (`displayName`) alongside `agentModelSorts`/`tieredModelIds`/`defaultAgentModelId` while excluding explicit non-chat (`tabModelIds`, `imageGenerationModelIds`, `audioTranscriptionModelIds`, `deprecatedModelIds`), internal, and unnamed rows.
+- **Client-UA-scoped Antigravity model cache invalidation** (`src/ai/model-cache.ts`) — included `getAntigravityUserAgent()` in `antigravityAccountScope` so client User-Agent updates immediately invalidate stale cached model lists without waiting for the 6-hour TTL.
+- **In-name `-medium` thinking depth recognition** (`src/ai/providers/gemini.ts`, `src/ai/model-catalog.ts`) — `geminiNamedEffort` and `inferCatalogMetadata` now recognize `-medium` model suffixes (`gemini-4-argon-medium`, `gemini-3.8-flash-medium`) so unset `reasoningEffort` defaults to `medium` thinking depth.
+
+### Verified
+- Focused Antigravity/Gemini/discovery/catalog regression suites (`175 pass, 0 fail`), `bun run typecheck`, `bun run build`, `bun run pack:check`, and live `antigravity/gemini-4-argon-{low,medium,high}` discovery + `antigravity/gemini-4-argon-medium` inference (`HTTP 200`).
+
+## [0.12.3] - 2026-10-17
+_Anthropic static model catalog trimmed to claude-5.x with live API discovery, and `/route off` strictly fixes the active model with no failover._
+
+### Changed
+- **Anthropic claude-4.x entries removed from the static catalog** (`src/ai/model-catalog.ts`) — `claude-opus-4-8`, `claude-opus-4-7`, `claude-sonnet-4-6`, `claude-opus-4-6` dropped. Static catalog is now a fallback-only claude-5.x set (`sonnet-5-5`, `opus-5-5`, `sonnet-5`, `fable-5`, `mythos-5`).
+- **`sonnet` / `opus` built-in aliases updated to claude-5.5** (`src/ai/model-registry.ts`) — `sonnet` → `claude-sonnet-5-5`, `opus` → `claude-opus-5-5`.
+- **Live `https://api.anthropic.com/v1/models` is the primary model source** — when logged in, the API returns the account's actual available models; the static catalog serves as fallback only.
+- **`/route off` fixes the model with no failover** (`src/commands/launch.ts`, `src/commands/launch/route-slash.ts`, `src/agent/task-tool.ts`) — when routing is off (via config or `/route off`), neither main turns nor delegated subagents bail early or fail over to equivalent models on 429/refusal, and `/route` status reports `routing: off (this session) — model fixed, no failover`.
+
+### Fixed
+- **Provider-qualified catalog entries no longer hijack bare model IDs** (`src/ai/model-catalog.ts`, `src/ai/model-catalog-compat.ts`) — `findCatalogModel`, `findCatalogEntry`, and `catalogMetadata` only match a bare `providerModel` when the catalog row itself is not provider-qualified (`!canonical.startsWith(`${provider}/`)`), preventing `antigravity/claude-sonnet-4-6` from misrouting bare `claude-sonnet-4-6` to `antigravity`.
+
+### Verified
+- `bun run typecheck`, `bun run pack:check`, and full regression suites green.
+
 ## [0.12.2] - 2026-10-16
 _Select Argon through the official agy full agent, with its own live catalog and session, separately from jeo's Cloud Code Assist backend._
 

package/README.ja.md

@@ -416,10 +416,10 @@ CI は `.github/workflows/npm-publish.yml` で公開します — GitHub リリ
 
 <!-- CHANGELOG:START (auto-generated from CHANGELOG.md — run `bun run changelog:sync`) -->
 - **[Unreleased]**
+- **[0.12.4]** (2026-10-17) — Refresh Antigravity provider (`antigravity/*`) account catalogs with `antigravity/hub/2.23.1` so entitled Gemini 4 Argon and Gemini 3.8 Flash tiers appear directly in `/model provider antigravity`.
+- **[0.12.3]** (2026-10-17) — Anthropic static model catalog trimmed to claude-5.x with live API discovery, and `/route off` strictly fixes the active model with no failover.
 - **[0.12.2]** (2026-10-16) — Select Argon through the official agy full agent, with its own live catalog and session, separately from jeo's Cloud Code Assist backend.
 - **[0.12.1]** (2026-10-16) — Logged-in OpenAI subscription models were invisible on proxy-configured machines.
-- **[0.12.0]** (2026-10-16) — Karpathy behavioral guardrails become always-on discipline, and skill routing gains an ECC single-harness inventory plus a Jev-style advisory router.
-- **[0.11.6]** (2026-10-16) — Live Antigravity account model discovery now serves the actual account's available models instead of assuming static availability.
 
 See [CHANGELOG.md](CHANGELOG.md) for the full history.
 <!-- CHANGELOG:END -->

package/README.ko.md

@@ -417,10 +417,10 @@ CI는 `.github/workflows/npm-publish.yml`로 배포합니다 — GitHub 릴리
 
 <!-- CHANGELOG:START (auto-generated from CHANGELOG.md — run `bun run changelog:sync`) -->
 - **[Unreleased]**
+- **[0.12.4]** (2026-10-17) — Refresh Antigravity provider (`antigravity/*`) account catalogs with `antigravity/hub/2.23.1` so entitled Gemini 4 Argon and Gemini 3.8 Flash tiers appear directly in `/model provider antigravity`.
+- **[0.12.3]** (2026-10-17) — Anthropic static model catalog trimmed to claude-5.x with live API discovery, and `/route off` strictly fixes the active model with no failover.
 - **[0.12.2]** (2026-10-16) — Select Argon through the official agy full agent, with its own live catalog and session, separately from jeo's Cloud Code Assist backend.
 - **[0.12.1]** (2026-10-16) — Logged-in OpenAI subscription models were invisible on proxy-configured machines.
-- **[0.12.0]** (2026-10-16) — Karpathy behavioral guardrails become always-on discipline, and skill routing gains an ECC single-harness inventory plus a Jev-style advisory router.
-- **[0.11.6]** (2026-10-16) — Live Antigravity account model discovery now serves the actual account's available models instead of assuming static availability.
 
 See [CHANGELOG.md](CHANGELOG.md) for the full history.
 <!-- CHANGELOG:END -->

package/README.md

@@ -409,10 +409,10 @@ Huge thanks to [gajae-code](https://github.com/Yeachan-Heo/gajae-code) for the i
 
 <!-- CHANGELOG:START (auto-generated from CHANGELOG.md — run `bun run changelog:sync`) -->
 - **[Unreleased]**
+- **[0.12.4]** (2026-10-17) — Refresh Antigravity provider (`antigravity/*`) account catalogs with `antigravity/hub/2.23.1` so entitled Gemini 4 Argon and Gemini 3.8 Flash tiers appear directly in `/model provider antigravity`.
+- **[0.12.3]** (2026-10-17) — Anthropic static model catalog trimmed to claude-5.x with live API discovery, and `/route off` strictly fixes the active model with no failover.
 - **[0.12.2]** (2026-10-16) — Select Argon through the official agy full agent, with its own live catalog and session, separately from jeo's Cloud Code Assist backend.
 - **[0.12.1]** (2026-10-16) — Logged-in OpenAI subscription models were invisible on proxy-configured machines.
-- **[0.12.0]** (2026-10-16) — Karpathy behavioral guardrails become always-on discipline, and skill routing gains an ECC single-harness inventory plus a Jev-style advisory router.
-- **[0.11.6]** (2026-10-16) — Live Antigravity account model discovery now serves the actual account's available models instead of assuming static availability.
 
 See [CHANGELOG.md](CHANGELOG.md) for the full history.
 <!-- CHANGELOG:END -->

package/README.zh.md

@@ -414,10 +414,10 @@ CI 通过 `.github/workflows/npm-publish.yml` 发布 — GitHub 发布 release
 
 <!-- CHANGELOG:START (auto-generated from CHANGELOG.md — run `bun run changelog:sync`) -->
 - **[Unreleased]**
+- **[0.12.4]** (2026-10-17) — Refresh Antigravity provider (`antigravity/*`) account catalogs with `antigravity/hub/2.23.1` so entitled Gemini 4 Argon and Gemini 3.8 Flash tiers appear directly in `/model provider antigravity`.
+- **[0.12.3]** (2026-10-17) — Anthropic static model catalog trimmed to claude-5.x with live API discovery, and `/route off` strictly fixes the active model with no failover.
 - **[0.12.2]** (2026-10-16) — Select Argon through the official agy full agent, with its own live catalog and session, separately from jeo's Cloud Code Assist backend.
 - **[0.12.1]** (2026-10-16) — Logged-in OpenAI subscription models were invisible on proxy-configured machines.
-- **[0.12.0]** (2026-10-16) — Karpathy behavioral guardrails become always-on discipline, and skill routing gains an ECC single-harness inventory plus a Jev-style advisory router.
-- **[0.11.6]** (2026-10-16) — Live Antigravity account model discovery now serves the actual account's available models instead of assuming static availability.
 
 See [CHANGELOG.md](CHANGELOG.md) for the full history.
 <!-- CHANGELOG:END -->

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "jeo-code",
-  "version": "0.12.2",
+  "version": "0.12.4",
   "description": "Clean, highly optimized AI coding agent using spec-first loop",
   "type": "module",
   "main": "src/cli.ts",

package/src/agent/task-tool.ts

@@ -433,6 +433,10 @@ export async function runSubagentOnce(
   // "false positive -> try a different model" design this implements.
   const safetyFallbackAvailable = (): boolean => safetyFallbackCandidates().length > 0;
 
+  // Failover (reroute to an equivalent model) is part of routing: it only runs when
+  // `routing.enabled` is on. `launch.ts` passes the parent session's EFFECTIVE state
+  // (`/route on|off` override ?? config) in `config.routing.enabled`.
+  const failoverEnabled = !!opts.config.routing?.enabled;
   let activeModel = initialModel;
   emit({ role: role.id, kind: "start", detail: taskText, maxSteps, model: activeModel });
   const runOnce = () => {
@@ -448,8 +452,7 @@ export async function runSubagentOnce(
       signal,
       steer,
       tools: subagentToolset(role),
-      rateLimitFallbackAvailable,
-      safetyFallbackAvailable,
+      ...(failoverEnabled ? { rateLimitFallbackAvailable, safetyFallbackAvailable } : {}),
       events: {
         onStep: n => { currentStep = n; },
         // Live reasoning preview (native extended-thinking models only — the JSON-protocol
@@ -494,7 +497,10 @@ export async function runSubagentOnce(
   };
 
   let result = await runOnce();
-  for (let attempt = 0; attempt < MAX_SUBAGENT_REROUTES; attempt++) {
+  // `routing.enabled` off = the model is FIXED: no failover reroute at all (the
+  // engine bail predicates above are withheld too, so a 429 rides its normal
+  // backoff ladder instead of bailing early with nowhere to go).
+  for (let attempt = 0; failoverEnabled && attempt < MAX_SUBAGENT_REROUTES; attempt++) {
     const rateLimited = !result.done && isRateLimitError(new Error(result.doneReason ?? ""));
     const safetyBail = !result.done && !rateLimited && isSafetyFallbackBail(result.doneReason);
     if (!rateLimited && !safetyBail) break;

package/src/ai/model-cache.ts

@@ -22,6 +22,7 @@ import { jeoEnv } from "../util/env";
 import { recordLiveCodexModels, recordLiveProviderModels, setAntigravityAccountScope, setOpenAIOauthAccountScope } from "./model-catalog";
 import type { Credential } from "../auth";
 import { getStoredOAuth, snapshotProvider } from "../auth/storage";
+import { getAntigravityUserAgent } from "./providers/antigravity";
 import { extractChatgptAccountId } from "./providers/openai-responses";
 import type { ProviderName } from "./types";
 
@@ -87,7 +88,7 @@ export async function antigravityAccountScope(credential?: Credential): Promise<
       || process.env.GOOGLE_CLOUD_PROJECT?.trim()
       || process.env.GOOGLE_CLOUD_PROJECT_ID?.trim()
       || "";
-    return createHash("sha256").update(`${providerSlot}\u0000${identity}\u0000${projectId}`, "utf8").digest("hex");
+    return createHash("sha256").update(`${providerSlot}\u0000${identity}\u0000${projectId}\u0000${getAntigravityUserAgent()}`, "utf8").digest("hex");
   } catch {
     return undefined;
   }

package/src/ai/model-catalog-compat.ts

@@ -55,7 +55,11 @@ export function findCatalogEntry(id: string): ModelCatalogEntry | undefined {
   const direct = findCatalogModel(id);
   if (direct) return adapt(direct);
   const n = normalizeModelId(id);
-  const hit = MODEL_CATALOG.find(m => normalizeModelId(m.canonical) === n || normalizeModelId(m.providerModel) === n);
+  const hit = MODEL_CATALOG.find(m =>
+    normalizeModelId(m.canonical) === n
+    || normalizeModelId(`${m.provider}/${m.providerModel}`) === n
+    || (!m.canonical.startsWith(`${m.provider}/`) && normalizeModelId(m.providerModel) === n),
+  );
   return hit ? adapt(hit) : undefined;
 }
 

package/src/ai/model-catalog.ts

@@ -80,7 +80,9 @@ const STD: ThinkLevel[] = ["low", "medium", "high"];
  *  (`cheapestCredentialed`/`strongestCredentialed`/`compareStrengthAscending`)
  *  now tiebreak on `releaseDate` directly rather than relying on position. */
 export const MODEL_CATALOG: readonly CatalogModel[] = [
-  // Anthropic (newest first: sonnet-5-5 > opus-5-5 > sonnet-5 > fable-5 > opus-4-8 > opus-4-7 > sonnet-4-6 > opus-4-6). 4.5-and-below dropped from jeo's exposed model list per user direction.
+  // Anthropic (newest first: sonnet-5-5 > opus-5-5 > sonnet-5 > fable-5 > mythos-5).
+  // claude-4.x and below removed: live API discovery (https://api.anthropic.com/v1/models)
+  // returns the account's actual available models; the static catalog is fallback-only.
   // 5.5 generation (platform.claude.com/docs/en/models/{opus-5-5,sonnet-5-5}/overview): dateless pinned wire ids
   // `claude-opus-5-5` (released 2026-09-22) / `claude-sonnet-5-5` (2026-09-28), 1M ctx / 128k sync max output, adaptive
   // thinking (Opus 5.5 always on; Sonnet 5.5 on by default), text+image input. Canonical == wire id; anthropic.ts already
@@ -92,17 +94,7 @@ export const MODEL_CATALOG: readonly CatalogModel[] = [
   // Mythos 5 is limited-availability (Project Glasswing) but callable by id for approved accounts.
   { canonical: "claude-fable-5", provider: "anthropic", providerModel: "claude-fable-5", contextTokens: 1_000_000, maxOutputTokens: 128_000, thinking: FULL, images: true, releaseDate: "2026-06" },
   { canonical: "claude-mythos-5", provider: "anthropic", providerModel: "claude-mythos-5", contextTokens: 1_000_000, maxOutputTokens: 128_000, thinking: FULL, images: true, limitedAvailability: true, releaseDate: "2026-06" },
-  // NOTE: opus 4.6+ use Anthropic ADAPTIVE thinking (type:"adaptive" + output_config.effort).
-  // opus 4.7/4.8 OMIT visible thought unless the request opts into `display: "summarized"` —
-  // anthropic.ts sets that on the adaptive transport so reasoning streams again (gjc parity).
-  // The nativizable path still replays signature-only thinking blocks for cross-turn continuity.
-  // The 4.6 generation onward is 1M context / 128k sync max output (Anthropic docs:
-  // Opus 4.8 & Sonnet 5 comparison table; Opus 4.6/4.7, Sonnet 4.6 share the gen's
-  // dateless-snapshot 1M/128k + 300k batch-output beta).
-  { canonical: "claude-opus-4-8", provider: "anthropic", providerModel: "claude-opus-4-8", contextTokens: 1_000_000, maxOutputTokens: 128_000, thinking: FULL, images: true, releaseDate: "2026-05" },
-  { canonical: "claude-opus-4-7", provider: "anthropic", providerModel: "claude-opus-4-7", contextTokens: 1_000_000, maxOutputTokens: 128_000, thinking: FULL, images: true, releaseDate: "2026-04" },
-  { canonical: "claude-sonnet-4-6", provider: "anthropic", providerModel: "claude-sonnet-4-6", contextTokens: 1_000_000, maxOutputTokens: 128_000, thinking: FULL, images: true, releaseDate: "2026-02" },
-  { canonical: "claude-opus-4-6", provider: "anthropic", providerModel: "claude-opus-4-6", contextTokens: 1_000_000, maxOutputTokens: 128_000, thinking: FULL, images: true, releaseDate: "2026-02" },
+
   // OpenAI (newest first: gpt-5.6 > gpt-5.5 > gpt-5.4 > o4-mini/o3/gpt-4.1 > o3-mini > gpt-4o-mini > gpt-4o)
   { canonical: "gpt-5.6", provider: "openai", providerModel: "gpt-5.6", contextTokens: 400_000, maxOutputTokens: 128_000, thinking: FULL, images: true, releaseDate: "2026-06" },
   { canonical: "gpt-5.5", provider: "openai", providerModel: "gpt-5.5", contextTokens: 400_000, maxOutputTokens: 128_000, thinking: FULL, images: true, releaseDate: "2026-04" },
@@ -526,7 +518,11 @@ export function formatTokens(n: number): string {
 /** Exact lookup by canonical id or provider model id. */
 export function findCatalogModel(idOrModel: string): CatalogModel | undefined {
   const q = idOrModel.trim();
-  return MODEL_CATALOG.find(m => m.canonical === q || m.providerModel === q || `${m.provider}/${m.providerModel}` === q);
+  return MODEL_CATALOG.find(m =>
+    m.canonical === q
+    || `${m.provider}/${m.providerModel}` === q
+    || (!m.canonical.startsWith(`${m.provider}/`) && m.providerModel === q),
+  );
 }
 
 /**
@@ -622,7 +618,7 @@ export function inferCatalogMetadata(modelId: string): CatalogModel | undefined
       providerModel: id,
       contextTokens: 1_000_000,
       maxOutputTokens: 65_536,
-      thinking: !reasons ? [] : big3 || id.includes("thinking") || id.includes("-high") || id.includes("-low") ? FULL : STD,
+      thinking: !reasons ? [] : big3 || id.includes("thinking") || id.includes("-high") || id.includes("-medium") || id.includes("-low") ? FULL : STD,
       images: true,
       company: antigravity ? "Google Antigravity" : "Google",
     };
@@ -653,7 +649,10 @@ export function catalogMetadata(modelId: string): CatalogModel | undefined {
   if (direct) return direct;
   // Tolerate provider-prefixed or bare provider model ids.
   const bare = modelId.replace(/^[a-z-]+\//, "");
-  const hit = MODEL_CATALOG.find(m => m.providerModel === bare || m.providerModel.endsWith(`/${bare}`) || m.canonical === bare);
+  const hit = MODEL_CATALOG.find(m =>
+    m.canonical === bare
+    || (!m.canonical.startsWith(`${m.provider}/`) && (m.providerModel === bare || m.providerModel.endsWith(`/${bare}`))),
+  );
   if (hit) return hit;
   // Last resort: infer capabilities from the model family so a brand-new
   // revision still surfaces reasoning/thinking like its catalogued siblings.

package/src/ai/model-discovery.ts

@@ -212,7 +212,7 @@ function isGeminiChatModel(id: string): boolean {
 }
 
 type CodexModelRow = { slug?: string; id?: string; visibility?: string; supported_in_api?: boolean; priority?: number };
-type AntigravityModelRow = { slug?: string; id?: string; name?: string; isInternal?: boolean; model?: string };
+type AntigravityModelRow = { slug?: string; id?: string; name?: string; displayName?: string; isInternal?: boolean; model?: string };
 
 function normalizeAntigravityModelId(id: string): string {
   return id.replace(/^models\//, "").replace(/^antigravity\//, "");
@@ -250,8 +250,9 @@ export function parseModelsBody(provider: ProviderName, body: unknown, opts: { o
     // (MODEL_PLACEHOLDER_*) and must never be surfaced. The response's OWN
     // metadata decides what to show — no hard-coded model lists:
     // Agent sorts and tiered model ids are complementary chat offerings, not
-    // interchangeable allowlists. Include both plus the declared default;
-    // only older responses without these fields need non-chat role filtering.
+    // interchangeable allowlists. Include both plus the declared default and
+    // named non-recommended chat models; exclude explicit non-chat roles and
+    // unnamed unreferenced rows.
     const payload = body as {
       models?: Record<string, AntigravityModelRow> | AntigravityModelRow[];
       agentModelSorts?: { groups?: { modelIds?: string[] }[] }[];
@@ -280,13 +281,16 @@ export function parseModelsBody(provider: ProviderName, body: unknown, opts: { o
     if (typeof payload.defaultAgentModelId === "string") agentIds.add(normalizeAntigravityModelId(payload.defaultAgentModelId));
     const hasChatModelMetadata = Array.isArray(payload.agentModelSorts)
       || payload.tieredModelIds !== undefined || payload.defaultAgentModelId !== undefined;
-    const nonChat = new Set([
+    const explicitNonChat = new Set([
       ...roleIds(payload.tabModelIds),
       ...roleIds(payload.imageGenerationModelIds),
       ...roleIds(payload.audioTranscriptionModelIds),
+      ...deprecated,
+    ]);
+    const nonChat = new Set([
+      ...explicitNonChat,
       ...roleIds(payload.commitMessageModelIds),
       ...roleIds(payload.mqueryModelIds),
-      ...deprecated,
     ]);
     const rawModels = payload.models;
     const entries = Array.isArray(rawModels)
@@ -297,7 +301,10 @@ export function parseModelsBody(provider: ProviderName, body: unknown, opts: { o
       .filter(([id, model]) =>
         id.length > 0 &&
         model?.isInternal !== true &&
-        (hasChatModelMetadata ? agentIds.has(id) : !nonChat.has(id)))
+        !explicitNonChat.has(id) &&
+        (hasChatModelMetadata
+          ? agentIds.size > 0 && (agentIds.has(id) || (!nonChat.has(id) && typeof model?.displayName === "string" && model.displayName.trim().length > 0))
+          : !nonChat.has(id)))
       .map(([id]) => `antigravity/${id}`);
   }
   if (provider === "gemini") {

package/src/ai/model-registry.ts

@@ -9,18 +9,18 @@ export interface ModelAliases {
 export const BUILTIN_ALIASES: ModelAliases = {
   fast: "ollama/qwen2.5:0.5b",
   local: "ollama/qwen2.5:0.5b",
-  sonnet: "claude-sonnet-4-6",
-  opus: "claude-opus-4-6",
+  sonnet: "claude-sonnet-5-5",
+  opus: "claude-opus-5-5",
   gpt: "gpt-5.5",
   flash: "gemini-2.5-flash",
   grok: "grok-4.3",
   kimi: "kimi-k2-0711-preview",
-  // Anthropic 5.5 generation: spellings that name the version. Bare `opus`/`sonnet` above stay on the
-  // 4.6 ids (pinned backward-compat); users opt into 5.5 by naming the version.
+  // Anthropic 5.5 generation: versioned spellings as convenience aliases.
   "opus-5.5": "claude-opus-5-5",
   "sonnet-5.5": "claude-sonnet-5-5",
   "opus5.5": "claude-opus-5-5",
   "sonnet5.5": "claude-sonnet-5-5",
+
   // Antigravity's backend DEPRECATED this wire id (deprecatedModelIds:
   // { "gemini-3.1-pro-high": { newModelId: "gemini-pro-agent" } }) — keep
   // existing configs/roles pinned to the old id working on the successor.

package/src/ai/providers/antigravity.ts

@@ -28,10 +28,10 @@ function antigravityClaudeThinkingBudget(effort: CallOptions["reasoningEffort"])
 const ENDPOINTS = [ANTIGRAVITY_DAILY_ENDPOINT, ANTIGRAVITY_SANDBOX_ENDPOINT] as const;
 
 export function getAntigravityUserAgent(): string {
-  const version = process.env.PI_AI_ANTIGRAVITY_VERSION || "1.104.0";
+  const version = process.env.PI_AI_ANTIGRAVITY_VERSION || "2.23.1";
   const os = process.platform === "win32" ? "windows" : process.platform;
   const arch = process.arch === "x64" ? "amd64" : process.arch === "ia32" ? "386" : process.arch;
-  return `antigravity/${version} ${os}/${arch}`;
+  return `antigravity/hub/${version} ${os}/${arch}`;
 }
 
 function antigravityModelId(model: string): string {

package/src/ai/providers/gemini.ts

@@ -48,6 +48,7 @@ export function geminiThinkingBudget(model: string, effort?: CallOptions["reason
  *  the documented thought-token burn protection). Expects a lowercased id. */
 function geminiNamedEffort(m: string): CallOptions["reasoningEffort"] | undefined {
   return m.includes("-high") ? "high"
+    : m.includes("-medium") ? "medium"
     : m.includes("-low") ? "low"
     : m.includes("thinking") ? "medium"
     : undefined;

package/src/commands/launch/route-slash.ts

@@ -44,6 +44,11 @@ export interface RouteSlashResult {
   lines: string[];
 }
 
+/** Off means a FIXED model: no prompt routing AND no equivalent-model failover. */
+function stateLabel(enabled: boolean): string {
+  return enabled ? "routing: on (this session)" : "routing: off (this session) — model fixed, no failover";
+}
+
 const USAGE = "Usage: /route [status|on|off|why|history [n]|save|on save|off save]";
 
 function isRealDecision(decision: RouteDecision | { note: string } | null): decision is RouteDecision {
@@ -92,7 +97,7 @@ export async function runRouteSlash(input: string, ctx: RouteSlashCtx): Promise<
   const effective = ctx.sessionRouteOverride ?? ctx.routingConfigEnabled;
 
   if (!sub || sub === "status") {
-    const lines = [`routing: ${effective ? "on" : "off"} (this session)`];
+    const lines = [stateLabel(effective)];
     if (isRealDecision(ctx.lastRouteDecision)) {
       lines.push(`last decision: ${ctx.lastRouteDecision.tier} → ${ctx.lastRouteDecision.model} (${ctx.lastRouteDecision.source}: ${ctx.lastRouteDecision.signals.join(", ") || "none"}, confidence ${ctx.lastRouteDecision.confidence.toFixed(2)})`);
     }
@@ -113,9 +118,9 @@ export async function runRouteSlash(input: string, ctx: RouteSlashCtx): Promise<
     const enabled = sub === "on";
     if (arg === "save") {
       await saveConfigPatch(raw => ({ routing: { ...raw.routing, enabled } }));
-      return { sessionRouteOverride: enabled, lines: [`routing: ${enabled ? "on" : "off"} (this session) — saved to ~/.jeo/config.json`] };
+      return { sessionRouteOverride: enabled, lines: [`${stateLabel(enabled)} — saved to ~/.jeo/config.json`] };
     }
-    return { sessionRouteOverride: enabled, lines: [`routing: ${enabled ? "on" : "off"} (this session)`] };
+    return { sessionRouteOverride: enabled, lines: [stateLabel(enabled)] };
   }
 
   // Bare `/route save` persists whatever is CURRENTLY effective (a prior

package/src/commands/launch.ts

@@ -1659,7 +1659,10 @@ export async function runLaunchCommand(args: string[]): Promise<void> {
         const fullTools = {
           ...DEFAULT_TOOLS,
           task: createTaskTool({
-            config: { ...turnConfig, defaultModel: activeModel },
+            // Subagents inherit the parent's EFFECTIVE routing state (`/route on|off`
+            // override ?? config) — task-tool.ts gates its failover reroute on it.
+            config: { ...turnConfig, defaultModel: activeModel, routing: { ...turnConfig.routing, enabled: routingEnabled } },
+
             signal: ac.signal,
             steer: drainSteer,
             registry: subagentRegistry,
@@ -1710,12 +1713,13 @@ export async function runLaunchCommand(args: string[]): Promise<void> {
         // mutated in place (`.add`) as the fallback loop below tries more models — this
         // closure always reads its live contents, never a stale snapshot.
         const attemptedRouteModels = new Set<string>([activeModel]);
-        // A NON-routed turn (routing disabled, or the user pinned a model via /model)
-        // still needs the equivalent-pool fallback on a usage-limit/rate-limit/credential
-        // failure — the exact "5-hour usage window exhausted, model just stops working"
-        // case this loop exists to fix. Synthesize a same-size-class RouteDecision
-        // (inferTierForModel) so `equivalentRouteFallback` has a tier/pool to search even
-        // with routing off; a real `routed` decision (routing enabled) is used unchanged.
+        // The synthesized decision below is only the BASE for `equivalentRouteFallback`'s
+        // tier/pool search when the pre-call router produced no decision (routing on but
+        // fail-open, or a /model pin with `/route on`). It is inert when routing is off:
+        // the reroute loop and the engine's bail predicates are both gated on
+        // `routingEnabled`, so `/route off` means a FIXED model with no failover.
+        // (`inferTierForModel` gives the same-size-class tier/pool to search.)
+
         let fallbackBaseDecision: RouteDecision = routed ?? {
           model: activeModel,
           thinking: activeThinking,
@@ -1785,8 +1789,10 @@ export async function runLaunchCommand(args: string[]): Promise<void> {
           signal: ac.signal,
           sessionKey: cacheKey,
           steer: drainSteer,
-          rateLimitFallbackAvailable,
-          safetyFallbackAvailable,
+          // Withheld when routing is off: the bail predicates only make sense paired
+          // with the reroute loop below, which `routingEnabled` also gates.
+          ...(routingEnabled ? { rateLimitFallbackAvailable, safetyFallbackAvailable } : {}),
+
           events: wrapEvents(withStepPersistence({ ...withToolDetailCapture(tui ? tui.events() : streamEvents), onBeforeDone }, persistTurnTail), opik),
         });
         const routeFailureReason = (doneReason: unknown): string | null => {
@@ -1866,7 +1872,8 @@ export async function runLaunchCommand(args: string[]): Promise<void> {
         const routeFallbackMax = Number.isFinite(routeFallbackMaxRaw) && routeFallbackMaxRaw >= 0
           ? Math.floor(routeFallbackMaxRaw)
           : 3;
-        for (let routeFallbackAttempt = 0; routeFallbackAttempt < routeFallbackMax; routeFallbackAttempt++) {
+        for (let routeFallbackAttempt = 0; routingEnabled && routeFallbackAttempt < routeFallbackMax; routeFallbackAttempt++) {
+
           // A loop guard is a local safety boundary, not a provider failure. A
           // fresh loop on an equivalent model loses the guard state and could
           // re-execute a previously skipped mutation.
@@ -1962,10 +1969,20 @@ export async function runLaunchCommand(args: string[]): Promise<void> {
     }
     // A completed turn with an empty done-reason must NOT masquerade as a step-limit
     // failure ("reached the 3-step limit" after the model called done at step 3).
-    const reply = stripSafetyFallbackTag(result.doneReason ?? "")
+    const baseReply = stripSafetyFallbackTag(result.doneReason ?? "")
       || (result.done
         ? `(done in ${result.steps} step${result.steps === 1 ? "" : "s"} — the model returned no summary)`
         : `(reached the ${result.steps}-step limit without signaling done)`);
+    // Routing off = the model is fixed and nothing fails over. When the turn died on
+    // an account-level limit, say so and name the one switch that re-enables failover.
+    const failoverHint = !routingEnabled && !result.done && (() => {
+      const failure = new Error(result.doneReason ?? "");
+      return isRateLimitError(failure) || isUsageLimitError(failure);
+    })()
+      ? "\n(routing is off — the model is fixed and no equivalent-model failover was attempted. Run '/route on' to allow failover.)"
+      : "";
+    const reply = `${baseReply}${failoverHint}`;
+
     if (sessionNotifyEndpoint) {
       const usageStr = result.usage ? formatUsage(result.usage) : undefined;
       if (notifyRedact) {