codeshark-cli 0.1.5 → 0.1.6

package/dist/index.js

@@ -250,7 +250,10 @@ async function main() {
             if (status === "available") {
                 console.log(`    ${hex("#4ade80", "✓ Available")}  ${model.label}`);
             }
-            else {
+            else if (status === "busy") {
+                console.log(`    ${hex("#fbbf24", "~ Lane busy")}  ${model.label}${reason ? dim(` — ${reason}`) : ""}`);
+            }
+            else if (status === "unavailable") {
                 console.log(`    ${hex("#f87171", "✗ Unavailable")}  ${model.label}${reason ? dim(` — ${reason}`) : ""}`);
             }
         }

package/dist/modelAvailability.js

@@ -2,7 +2,9 @@ import { DEFAULT_GATEWAY_URL, envApiKey } from "./config.js";
 import { MODELS } from "./models.js";
 const statuses = new Map();
 const reasons = new Map();
-const TIMEOUT_MS = 8_000;
+const TIMEOUT_MS = 15_000;
+/** Match the gateway's per-IP inflight cap so probes never queue behind themselves. */
+const MAX_CONCURRENT_CHECKS = 2;
 export function modelAvailability(modelId) {
     return statuses.get(modelId) ?? "checking";
 }
@@ -16,6 +18,25 @@ export function modelAvailabilitySnapshot() {
         reason: modelAvailabilityReason(model.id),
     }));
 }
+/**
+ * Classify a failed probe. "busy" covers transient lane problems (rate
+ * limits, upstream hiccups, timeouts) where the model is likely fine but
+ * the shared free lane is momentarily overloaded — those must NOT read as
+ * "model removed". Only explicit client errors mean the model itself is
+ * not servable right now.
+ */
+export function classifyProbeFailure(statusOrError) {
+    if (statusOrError instanceof Error) {
+        if (statusOrError.name === "AbortError")
+            return { status: "busy", reason: "lane busy — timed out" };
+        return { status: "busy", reason: statusOrError.message || "network error" };
+    }
+    if (statusOrError === 429)
+        return { status: "busy", reason: "rate limited — try again shortly" };
+    if (statusOrError >= 500)
+        return { status: "busy", reason: `lane busy — HTTP ${statusOrError}` };
+    return { status: "unavailable", reason: `HTTP ${statusOrError}` };
+}
 function endpointFor(model, cfg) {
     if (model.provider === "gemini") {
         return {
@@ -53,8 +74,12 @@ async function checkOne(model, cfg) {
                 body: JSON.stringify({ contents: [{ role: "user", parts: [{ text: "Reply with OK." }] }] }),
                 signal: controller.signal,
             });
-            if (!response.ok)
-                throw new Error(`HTTP ${response.status}`);
+            if (!response.ok) {
+                const { status, reason } = classifyProbeFailure(response.status);
+                statuses.set(model.id, status);
+                reasons.set(model.id, reason);
+                return;
+            }
         }
         else {
             const response = await fetch(target.url, {
@@ -63,15 +88,20 @@ async function checkOne(model, cfg) {
                 body: JSON.stringify({ model: model.model, messages: [{ role: "user", content: "Reply with OK." }], stream: false, max_tokens: 4 }),
                 signal: controller.signal,
             });
-            if (!response.ok)
-                throw new Error(`HTTP ${response.status}`);
+            if (!response.ok) {
+                const { status, reason } = classifyProbeFailure(response.status);
+                statuses.set(model.id, status);
+                reasons.set(model.id, reason);
+                return;
+            }
         }
         statuses.set(model.id, "available");
         reasons.delete(model.id);
     }
     catch (error) {
-        statuses.set(model.id, "unavailable");
-        reasons.set(model.id, error instanceof Error && error.name === "AbortError" ? "check timed out" : error instanceof Error ? error.message : String(error));
+        const { status, reason } = classifyProbeFailure(error instanceof Error ? error : new Error(String(error)));
+        statuses.set(model.id, status);
+        reasons.set(model.id, reason);
     }
     finally {
         clearTimeout(timer);
@@ -82,5 +112,13 @@ export async function checkModelAvailability(cfg) {
         statuses.set(model.id, "checking");
         reasons.delete(model.id);
     }
-    await Promise.all(MODELS.map((model) => checkOne(model, cfg)));
+    // Throttle to the gateway's per-IP inflight limit: firing all probes at
+    // once makes them queue behind each other and time out spuriously.
+    const queue = [...MODELS];
+    const workers = Array.from({ length: Math.min(MAX_CONCURRENT_CHECKS, queue.length) }, async () => {
+        for (let model = queue.shift(); model; model = queue.shift()) {
+            await checkOne(model, cfg);
+        }
+    });
+    await Promise.all(workers);
 }

package/dist/repl.js

@@ -48,9 +48,14 @@ export function printModelInfo() {
     for (const m of MODELS) {
         const active = m.id === (findModel(cfg.model ?? "")?.id ?? DEFAULT_MODEL_ID);
         const status = modelAvailability(m.id);
-        const marker = status === "available" ? hex("#4ade80", "●") : status === "unavailable" ? hex("#f87171", "✗") : dim("○");
-        const note = status === "unavailable" ? `Unavailable${modelAvailabilityReason(m.id) ? ` — ${modelAvailabilityReason(m.id)}` : ""}` : m.notes;
-        console.log(`    ${marker} ${m.label.padEnd(28)} ${dim(m.context.padEnd(5))} ${status === "unavailable" ? hex("#f87171", note) : note}`);
+        const marker = status === "available" ? hex("#4ade80", "●") : status === "unavailable" ? hex("#f87171", "✗") : status === "busy" ? hex("#fbbf24", "◐") : dim("○");
+        const note = status === "unavailable"
+            ? `Unavailable${modelAvailabilityReason(m.id) ? ` — ${modelAvailabilityReason(m.id)}` : ""}`
+            : status === "busy"
+                ? `Lane busy${modelAvailabilityReason(m.id) ? ` — ${modelAvailabilityReason(m.id)}` : ""} — should still work`
+                : m.notes;
+        const styled = status === "unavailable" ? hex("#f87171", note) : status === "busy" ? hex("#fbbf24", note) : note;
+        console.log(`    ${marker} ${m.label.padEnd(28)} ${dim(m.context.padEnd(5))} ${styled}`);
     }
     console.log("");
     console.log(dim("  Switch with: /model <name>"));
@@ -81,6 +86,9 @@ export function switchModel(query) {
         console.log(dim(`  ${modelAvailabilityReason(entry.id) ?? "The boot health check failed."}`));
         return;
     }
+    if (modelAvailability(entry.id) === "busy") {
+        console.log(hex("#fbbf24", `  ⚠ ${entry.label}'s shared lane is busy right now — switching anyway. Requests may wait or retry.`));
+    }
     cfg.model = entry.id;
     saveConfig(cfg);
     console.log(`  ${hex("#4ade80", "✓")} Switched to ${bold(entry.label)} ${dim(`(${entry.context} context)`)}`);

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "codeshark-cli",
-  "version": "0.1.5",
+  "version": "0.1.6",
   "description": "An open-source terminal coding agent. Eight model choices, project tools, and approval before every tool action.",
   "type": "module",
   "bin": {