@hackerrank/astra-cli 0.1.17 → 0.1.19

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@hackerrank/astra-cli",
-  "version": "0.1.17",
+  "version": "0.1.19",
   "description": "Minimal zero-dependency AI coding agent for the HackerRank AI Gateway.",
   "type": "module",
   "bin": {

package/src/agent.js

@@ -105,7 +105,11 @@ export class Agent {
 
   apiMessages() {
     return this.messages
-      .filter((m) => m.role === "system" || m.role === "user" || m.role === "assistant")
+      .filter(
+        (m) =>
+          (m.role === "system" || m.role === "user") ||
+          (m.role === "assistant" && m.content !== "")
+      )
       .map((m) => ({ role: m.role, content: m.content }));
   }
 
@@ -131,7 +135,7 @@ export class Agent {
       throw err;
     }
     this.lastContextTokens = usage.prompt_tokens;
-    this.add("assistant", content, { usage });
+    if (content !== "") this.add("assistant", content, { usage });
 
     this.onStep({
       step: this.nSteps,

package/src/model.js

@@ -66,8 +66,9 @@ export class GatewayModel {
     this.reportedCostUsd = 0;
     this.estimatedCostUsd = 0;
     this.costSource = null; // "reported" | "estimated" | "mixed" | null
-    // Preferred token-cap parameter; swapped automatically on 400 if needed.
-    this._tokenParam = "max_tokens";
+    // GPT-family gateway routes use the newer completion-token parameter.
+    // Other OpenAI-compatible providers continue using the legacy parameter.
+    this._tokenParam = /^gpt(?:-|$)/i.test(model) ? "max_completion_tokens" : "max_tokens";
     // Optional AbortSignal; when set and aborted, in-flight requests are
     // cancelled (used to force-quit a long model call on a second Ctrl+C).
     this.signal = null;