Merge remote-tracking branch 'refs/remotes/origin/main' into dohooo/team-cloud-sandbox

Author: dohooo

CHANGELOG.md

@@ -1,5 +1,13 @@
 # Changelog
 
+## 0.39.1
+
+### Patch Changes
+
+- [#848](https://github.com/dohooo/helmor/pull/848) [`eac697c`](https://github.com/dohooo/helmor/commit/eac697c62d0bbf434433421db09d63ac11f25893) Thanks [@natllian](https://github.com/natllian)! - Improve Claude model handling:
+  - The default Claude model is pinned to Opus 4.8 (1M context) so it can't silently switch to a different model when the bundled Claude CLI updates; existing sessions and settings keep the same model.
+  - Terminal mode is now limited to official Claude models — custom (BYOK) Claude models run in GUI mode instead, since the terminal can't carry their custom provider settings.
+
 ## 0.39.0
 
 ### Minor Changes

package.json

@@ -2,7 +2,7 @@
 	"name": "helmor",
 	"description": "The local-first IDE for coding agent orchestration.",
 	"private": true,
-	"version": "0.39.0",
+	"version": "0.39.1",
 	"packageManager": "bun@1.3.2",
 	"type": "module",
 	"workspaces": [

sidecar/src/model-catalog.ts

@@ -11,24 +11,25 @@ const MODEL_CATALOG: Record<Provider, readonly ProviderModelInfo[]> = {
 	claude: [
 		// Fable 5 leads the list as the most capable pick, but it burns limits
 		// ~2x faster than Opus — `useEnsureDefaultModel` pins the app default
-		// to the `default` (Opus) entry below, NOT to the first entry. No fast
+		// to the Opus 4.8 entry below, NOT to the first entry. No fast
 		// mode (Opus 4.6+ only).
 		{
 			id: "claude-fable-5[1m]",
 			label: "Fable 5 1M",
 			cliModel: "claude-fable-5[1m]",
 			effortLevels: ["low", "medium", "high", "xhigh", "max"],
 		},
-		// `default` resolves to the newest Opus the bundled claude-code knows
-		// about — in 2.1.170 that is Opus 4.8 (1M context, adaptive thinking,
-		// default high effort, fast mode at 2x rate / 2.5x speed). Kept as
-		// `default` (rather than pinned `claude-opus-4-8`) so it stays the
-		// auto-latest pick AND remains the app default selection (see
-		// `useEnsureDefaultModel`, which prefers id == "default").
+		// App default selection (see `useEnsureDefaultModel`, which pins this
+		// id). Pinned to the explicit `claude-opus-4-8[1m]` wire id — the `[1m]`
+		// suffix selects the 1M-context variant, matching the label. We do NOT
+		// use the CLI's `default` sentinel: it resolves to whatever the bundled
+		// claude-code decides is "default" (non-deterministic across CLI bumps),
+		// whereas a pinned id is stable. Bump when a newer Opus ships. MUST stay
+		// in sync with the Rust catalog (`official_claude_section`).
 		{
-			id: "default",
+			id: "claude-opus-4-8[1m]",
 			label: "Opus 4.8 1M",
-			cliModel: "default",
+			cliModel: "claude-opus-4-8[1m]",
 			effortLevels: ["low", "medium", "high", "xhigh", "max"],
 			supportsFastMode: true,
 		},

sidecar/test/claude-session-manager.test.ts

@@ -493,8 +493,9 @@ describe("ClaudeSessionManager.sendMessage", () => {
 			models.map((m) => [m.id, m.supportsFastMode]),
 		);
 
-		// `default` now resolves to Opus 4.8, which supports fast mode.
-		expect(bySupports.default).toBe(true);
+		// Opus 4.8 1M supports fast mode.
+		expect(bySupports["claude-opus-4-8[1m]"]).toBe(true);
+		expect(bySupports.default).toBeUndefined();
 		expect(bySupports.sonnet).toBeUndefined();
 		expect(bySupports["claude-opus-4-7[1m]"]).toBeUndefined();
 		expect(bySupports["claude-opus-4-6[1m]"]).toBe(true);
@@ -539,7 +540,7 @@ describe("ClaudeSessionManager.sendMessage", () => {
 			{
 				sessionId: `helmor-sess-effort-${level}`,
 				prompt: "test",
-				model: "default",
+				model: "claude-opus-4-8[1m]",
 				cwd: undefined,
 				resume: undefined,
 				permissionMode: undefined,
@@ -562,7 +563,7 @@ describe("ClaudeSessionManager.sendMessage", () => {
 			{
 				sessionId: "helmor-sess-effort-bogus",
 				prompt: "test",
-				model: "default",
+				model: "claude-opus-4-8[1m]",
 				cwd: undefined,
 				resume: undefined,
 				permissionMode: undefined,
@@ -585,7 +586,7 @@ describe("ClaudeSessionManager.sendMessage", () => {
 			{
 				sessionId: "helmor-sess-mcp-blocking",
 				prompt: "test",
-				model: "default",
+				model: "claude-opus-4-8[1m]",
 				cwd: undefined,
 				resume: undefined,
 				permissionMode: undefined,
@@ -633,7 +634,7 @@ describe("ClaudeSessionManager.sendMessage", () => {
 			{
 				sessionId: "helmor-sess-fm",
 				prompt: "hi",
-				model: "default",
+				model: "claude-opus-4-8[1m]",
 				cwd: undefined,
 				resume: undefined,
 				permissionMode: undefined,
@@ -675,7 +676,7 @@ describe("ClaudeSessionManager.sendMessage", () => {
 			{
 				sessionId: "helmor-sess-fm-init",
 				prompt: "hi",
-				model: "default",
+				model: "claude-opus-4-8[1m]",
 				cwd: undefined,
 				resume: undefined,
 				permissionMode: undefined,
@@ -714,7 +715,7 @@ describe("ClaudeSessionManager.sendMessage", () => {
 			{
 				sessionId: "helmor-sess-fm-on",
 				prompt: "hi",
-				model: "default",
+				model: "claude-opus-4-8[1m]",
 				cwd: undefined,
 				resume: undefined,
 				permissionMode: undefined,
@@ -1771,9 +1772,9 @@ describe("ClaudeSessionManager.listModels", () => {
 				effortLevels: ["low", "medium", "high", "xhigh", "max"],
 			},
 			{
-				id: "default",
+				id: "claude-opus-4-8[1m]",
 				label: "Opus 4.8 1M",
-				cliModel: "default",
+				cliModel: "claude-opus-4-8[1m]",
 				effortLevels: ["low", "medium", "high", "xhigh", "max"],
 				supportsFastMode: true,
 			},

src-tauri/Cargo.lock

@@ -1,6 +1,6 @@
 [package]
 name = "helmor"
-version = "0.39.0"
+version = "0.39.1"
 description = "The local-first IDE for coding agent orchestration."
 authors = ["Caspian Zhao", "Nathan Lian"]
 edition = "2021"

src-tauri/Cargo.toml

@@ -821,9 +821,12 @@ mod tests {
     #[test]
     fn resolve_model_infers_provider() {
         let _env = crate::testkit::TestEnv::new("resolve-model-infers-provider");
-        let claude = resolve_model("default", None);
+        // Hint-less ids that match no other provider infer to claude and pass
+        // through verbatim (legacy "default" rows are normalized by the DB
+        // migration, so resolve_model needs no special case for it).
+        let claude = resolve_model("claude-opus-4-8[1m]", None);
         assert_eq!(claude.provider, "claude");
-        assert_eq!(claude.cli_model, "default");
+        assert_eq!(claude.cli_model, "claude-opus-4-8[1m]");
 
         let codex = resolve_model("gpt-5.4", None);
         assert_eq!(codex.provider, "codex");

src-tauri/src/agents.rs

@@ -277,30 +277,29 @@ fn official_claude_section() -> AgentModelSection {
         options: vec![
             // Fable 5 leads the list as the most capable pick, but it burns
             // limits ~2x faster than Opus — `useEnsureDefaultModel` therefore
-            // pins the app default to the `default` (Opus) entry below, NOT
+            // pins the app default to the Opus 4.8 entry below, NOT
             // to options[0]. No fast mode (Opus 4.6+ only).
             claude_model(
                 "claude-fable-5[1m]",
                 "Fable 5 1M",
                 &["low", "medium", "high", "xhigh", "max"],
                 false,
             ),
-            // `default` resolves to the newest Opus the bundled claude-code
-            // knows about — 2.1.170 maps it to Opus 4.8 (1M context, adaptive
-            // thinking, default high effort, fast mode at 2x rate / 2.5x
-            // speed). Kept as `default` so it stays the auto-latest pick and
-            // remains the app's default selection (see
-            // `useEnsureDefaultModel`, which prefers id == "default"). MUST
-            // stay in sync with `sidecar/src/model-catalog.ts`.
+            // App default selection (see `useEnsureDefaultModel`, which pins
+            // this id). Pinned to the explicit `claude-opus-4-8[1m]` wire id —
+            // the `[1m]` suffix selects the 1M-context variant, matching the
+            // label. We do NOT use the CLI's `default` sentinel: it resolves to
+            // whatever the bundled claude-code decides (non-deterministic
+            // across CLI bumps), whereas a pinned id is stable. Bump when a
+            // newer Opus ships. MUST stay in sync with
+            // `sidecar/src/model-catalog.ts`.
             claude_model(
-                "default",
+                "claude-opus-4-8[1m]",
                 "Opus 4.8 1M",
                 &["low", "medium", "high", "xhigh", "max"],
                 true,
             ),
-            // Explicit 4.7 pin — this slot used to BE `default`; now that
-            // `default` advanced to 4.8 we surface 4.7 as its own selectable
-            // entry, above 4.6.
+            // Explicit 4.7 pin, above 4.6.
             claude_model(
                 "claude-opus-4-7[1m]",
                 "Opus 4.7 1M",
@@ -900,7 +899,7 @@ mod tests {
                 .collect::<Vec<_>>(),
             vec![
                 "claude-fable-5[1m]",
-                "default",
+                "claude-opus-4-8[1m]",
                 "claude-opus-4-7[1m]",
                 "claude-opus-4-6[1m]",
                 "sonnet",
@@ -970,7 +969,7 @@ mod tests {
                 .collect::<Vec<_>>(),
             vec![
                 "claude-fable-5[1m]",
-                "default",
+                "claude-opus-4-8[1m]",
                 "claude-opus-4-7[1m]",
                 "claude-opus-4-6[1m]",
                 "sonnet",
@@ -993,10 +992,11 @@ mod tests {
     #[test]
     fn resolve_claude_model() {
         let _env = crate::testkit::TestEnv::new("resolve-claude-model");
-        let m = resolve_model("default", None);
+        // The pinned Opus 4.8 1M id resolves to itself.
+        let m = resolve_model("claude-opus-4-8[1m]", None);
         assert_eq!(m.provider, "claude");
-        assert_eq!(m.cli_model, "default");
-        assert_eq!(m.id, "default");
+        assert_eq!(m.cli_model, "claude-opus-4-8[1m]");
+        assert_eq!(m.id, "claude-opus-4-8[1m]");
         assert!(m.supports_effort);
     }
 
@@ -1441,15 +1441,15 @@ mod tests {
             &ids[..4],
             &[
                 "claude-fable-5[1m]",
-                "default",
+                "claude-opus-4-8[1m]",
                 "claude-opus-4-7[1m]",
                 "claude-opus-4-6[1m]"
             ],
             "Fable 5 must lead, with Opus 4.8 (default) / 4.7 / 4.6 beneath it"
         );
 
         // Fable 5: most capable, leads the list, but is NOT the app default
-        // (too expensive) — `useEnsureDefaultModel` pins to id == "default".
+        // (too expensive) — `useEnsureDefaultModel` pins to the Opus 4.8 id.
         // No fast mode (Opus 4.6+ only); full effort tiers incl. xhigh.
         let fable = &claude.options[0];
         assert_eq!(fable.label, "Fable 5 1M");
@@ -1460,11 +1460,11 @@ mod tests {
             vec!["low", "medium", "high", "xhigh", "max"]
         );
 
-        // `default` → Opus 4.8: stays the app default selection, supports
-        // fast mode, and keeps the xhigh effort tier.
+        // Opus 4.8: the app default selection, supports fast mode, and keeps
+        // the xhigh effort tier. Pinned to its explicit `[1m]` wire id.
         let default = &claude.options[1];
         assert_eq!(default.label, "Opus 4.8 1M");
-        assert_eq!(default.cli_model, "default");
+        assert_eq!(default.cli_model, "claude-opus-4-8[1m]");
         assert!(default.supports_fast_mode, "Opus 4.8 supports fast mode");
         assert_eq!(
             default.effort_levels,
@@ -1490,10 +1490,12 @@ mod tests {
     #[test]
     fn claude_default_no_longer_collides_with_cursor_auto() {
         let _env = crate::testkit::TestEnv::new("claude-default-no-longer-collides-with-c");
-        // `default` belongs to Claude (Opus 4.8 1M). Cursor's Auto is
-        // `cursor-default`. They MUST resolve to different providers
-        // even when the picker / persistence flow doesn't pass a hint —
-        // this is the regression the namespace prefix exists to prevent.
+        // A hint-less bare `default` still infers to claude; cursor's Auto is
+        // the namespaced `cursor-default`. They MUST resolve to different
+        // providers even without a hint — the regression the namespace prefix
+        // exists to prevent. (Claude no longer ships a `default` model id; any
+        // legacy occurrence is normalized by the DB migration before it gets
+        // here, so resolve_model just passes it through.)
         let claude = resolve_model("default", None);
         assert_eq!(claude.provider, "claude");
         assert_eq!(claude.cli_model, "default");

src-tauri/src/agents/catalog.rs

@@ -414,9 +414,11 @@ fn import_workspace_db_records(conn: &Connection, workspace_id: &str) -> Result<
     )
     .context("Failed to import sessions")?;
 
-    // 3b. Remap legacy "opus-1m" model ID (CLI no longer accepts it)
+    // 3b. Remap legacy "opus-1m" model ID to the pinned Opus 4.8 1M wire id
+    // (CLI no longer accepts "opus-1m"; claude no longer uses the "default"
+    // sentinel).
     conn.execute(
-        "UPDATE main.sessions SET model = 'default' WHERE model = 'opus-1m' AND workspace_id = ?1",
+        "UPDATE main.sessions SET model = 'claude-opus-4-8[1m]' WHERE model = 'opus-1m' AND workspace_id = ?1",
         [workspace_id],
     )
     .ok();