pennyrouter 0.2.18 → 0.2.19

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "pennyrouter",
-  "version": "0.2.18",
+  "version": "0.2.19",
   "description": "Install and manage PennyRouter local coding-agent integrations.",
   "homepage": "https://pennyrouter.com",
   "bugs": {

package/src/harnesses/claude-code.js

@@ -4,15 +4,19 @@ import { configureClaudeCodeStatusline, restoreClaudeCodeStatusline } from "../s
 const CONFIG_PATH = "~/.claude/settings.json";
 const LEGACY_MODEL_LABEL = "PennyRouter Bundle";
 const CONNECTOR_SETTING = "disableClaudeAiConnectors";
-// The slot values remain genuine Anthropic ids so Claude Code recognizes their actual context
-// windows (Haiku 200K; the 1M-capable models 1M) and sends the native id on the wire. Claude
-// Code separately supports *_MODEL_NAME and *_MODEL_DESCRIPTION for the picker, so Penny can
-// present a friendly alias without sacrificing model recognition.
+// The slot values remain genuine Anthropic ids so Claude Code recognizes them and sends the
+// native id on the wire. Claude Code separately supports *_MODEL_NAME and *_MODEL_DESCRIPTION
+// for the picker, so Penny can present a friendly alias without sacrificing model recognition.
+//
+// The 1M-context variant is a `[1m]` suffix on the id — Claude Code does NOT infer a 1M window
+// from a bare id, so a bare `claude-sonnet-5` gets 200K and auto-compacts around 150K. The
+// suffix rides the wire; the gateway strips it for the upstream model and turns it into the
+// `context-1m` beta header. Haiku has no 1M variant, so it stays bare.
 export const CLAUDE_CODE_MODELS = {
   haiku: "claude-haiku-4-5-20251001",
-  sonnet: "claude-sonnet-5",
-  opus: "claude-opus-5",
-  fable: "claude-fable-5",
+  sonnet: "claude-sonnet-5[1m]",
+  opus: "claude-opus-5[1m]",
+  fable: "claude-fable-5[1m]",
   custom: "Penny Custom",             // -> dynamic (PennyRouter cost-routing)
 };
 const CLAUDE_CODE_MODEL_PRESENTATION = {

package/src/launch.test.js

@@ -103,7 +103,10 @@ async function main() {
     assert.ok(!claude.args.includes("--plugin-dir"));
     assert.equal(claude.env.ANTHROPIC_AUTH_TOKEN, "pr-super-secret");
     assert.equal(claude.env.ANTHROPIC_BASE_URL, "https://gateway.example/");
-    assert.equal(claude.env.ANTHROPIC_DEFAULT_OPUS_MODEL, "claude-opus-5");
+    // 1M-capable slots carry the `[1m]` suffix; Haiku has no 1M variant and stays bare.
+    // Per-slot suffixes, not the global CLAUDE_CODE_MAX_CONTEXT_TOKENS override, which
+    // would wrongly hand Haiku a 1M window too.
+    assert.equal(claude.env.ANTHROPIC_DEFAULT_OPUS_MODEL, "claude-opus-5[1m]");
     assert.equal(claude.env.ANTHROPIC_DEFAULT_HAIKU_MODEL, "claude-haiku-4-5-20251001");
     assert.equal(claude.env.ANTHROPIC_DEFAULT_HAIKU_MODEL_NAME, "Haiku 4.5 (PennyRouter)");
     assert.equal(claude.env.ANTHROPIC_DEFAULT_HAIKU_MODEL_DESCRIPTION, "Fast, economical Claude model · 200K context");
@@ -128,7 +131,7 @@ async function main() {
     });
     const persistedClaude = JSON.parse(fs.readFileSync(path.join(home, ".claude", "settings.json"), "utf8"));
     assert.equal(persistedClaude.env.ANTHROPIC_DEFAULT_HAIKU_MODEL, "claude-haiku-4-5-20251001");
-    assert.equal(persistedClaude.env.ANTHROPIC_DEFAULT_SONNET_MODEL, "claude-sonnet-5");
+    assert.equal(persistedClaude.env.ANTHROPIC_DEFAULT_SONNET_MODEL, "claude-sonnet-5[1m]");
     assert.equal("CLAUDE_CODE_MAX_CONTEXT_TOKENS" in persistedClaude.env, false);
 
     await removePennyHarness("claude-code");