@hackerrank/astra-cli 0.1.4 → 0.1.5

package/README.md

@@ -228,7 +228,7 @@ credits — that's a quota issue, not a bug.
                         whatever runs already exist on disk, then exit
 -C, --cwd <path>        Working directory for commands (default: cwd)
 -o, --output <path>     Also write trajectory JSON here (autonomous mode)
--s, --steps <n>         Step limit (default: 40)
+-s, --steps <n>         Turn limit (unset = no limit)
 -w, --wall <seconds>    Wall-clock limit (default: 0 = none)
     --timeout <seconds> Per-command timeout (default: 60)
     --max-output <n>    Max chars of command output kept (default: 16000)
@@ -237,7 +237,7 @@ credits — that's a quota issue, not a bug.
     --resume <id>       Resume a saved session
     --sessions          List saved sessions and exit
 -y, --yolo              Auto-run commands (always on in autonomous mode)
--q, --quiet             Do not stream steps (autonomous mode)
+-q, --quiet             Do not stream turns (autonomous mode)
 -h, --help              Show help
 ```
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@hackerrank/astra-cli",
-  "version": "0.1.4",
+  "version": "0.1.5",
   "description": "Minimal zero-dependency AI coding agent for the HackerRank AI Gateway.",
   "type": "module",
   "bin": {

package/src/agent.js

@@ -38,7 +38,7 @@ export class Agent {
     this.model = model;
     this.env = env;
     this.mode = opts.mode === "interactive" ? "interactive" : "autonomous";
-    this.stepLimit = opts.stepLimit ?? 40;
+    this.stepLimit = opts.stepLimit ?? 0;
     this.wallTimeLimitSeconds = opts.wallTimeLimitSeconds ?? 0;
     this.maxConsecutiveFormatErrors = opts.maxConsecutiveFormatErrors ?? 3;
     this.outputPath = opts.outputPath ?? null;

package/src/bench.js

@@ -282,7 +282,7 @@ export async function runCell({
   taskFile,
   projectMetadata,
   root,
-  steps = 40,
+  steps = 0,
   wall = 0,
   timeout = 60,
   maxOutputChars = 16000,

package/src/cli.js

@@ -45,7 +45,7 @@
  *       --bench-root <dir>  Root folder for bench runs (default: ./bench)
  *   -C, --cwd <path>        Working directory for commands (default: cwd)
  *   -o, --output <path>     Also write trajectory JSON here (bench mode)
- *   -s, --steps <n>         Step limit (default: 40)
+ *   -s, --steps <n>         Turn limit (unset = no limit)
  *   -w, --wall <seconds>    Wall-clock limit (default: 0 = none)
  *       --timeout <seconds> Per-command timeout (default: 60)
  *       --max-output <n>    Max chars of command output kept (default: 16000)
@@ -55,7 +55,7 @@
  *       --sessions          List saved sessions and exit
  *   -y, --yolo              Auto-run commands without confirmation
  *                           (always on in bench mode)
- *   -q, --quiet             Do not stream steps (bench mode)
+ *   -q, --quiet             Do not stream turns (bench mode)
  *   -h, --help              Show this help
  *
  * API key resolution (first hit wins):
@@ -113,7 +113,7 @@ function reasoningKwargs(level) {
 }
 
 function parseArgs(argv) {
-  const args = { steps: 40, wall: 0, timeout: 60, quiet: false, "max-output": 16000, _provided: new Set() };
+  const args = { steps: 0, wall: 0, timeout: 60, quiet: false, "max-output": 16000, _provided: new Set() };
   const alias = {
     "-m": "model", "--model": "model",
     "-t": "task", "--task": "task",
@@ -419,11 +419,15 @@ async function main() {
     if (benchRun) {
       console.error(
         `\x1b[2m[astra] bench · ${benchRun.slug}/${benchRun.runId} ` +
-          `· model=${modelId}${reasoning ? ` reasoning=${reasoning}` : ""} steps<=${args.steps}\x1b[0m`
+          `· model=${modelId}${reasoning ? ` reasoning=${reasoning}` : ""}` +
+          `${Number(args.steps) > 0 ? ` turns<=${args.steps}` : ""}\x1b[0m`
       );
       console.error(`\x1b[2m[astra] workspace -> ${benchRun.workspace}\x1b[0m`);
     } else {
-      console.error(`\x1b[2m[astra] bench · model=${modelId} cwd=${cmdCwd} steps<=${args.steps}\x1b[0m`);
+      console.error(
+        `\x1b[2m[astra] bench · model=${modelId} cwd=${cmdCwd}` +
+          `${Number(args.steps) > 0 ? ` turns<=${args.steps}` : ""}\x1b[0m`
+      );
     }
   }
 
@@ -451,7 +455,7 @@ async function main() {
     const pt = model.totalPromptTokens;
     const ct = model.totalCompletionTokens;
     console.error(
-      `\n\x1b[1m[astra] exit=${result.exit_status} steps=${agent.nSteps} calls=${model.nCalls}\x1b[0m`
+      `\n\x1b[1m[astra] exit=${result.exit_status} turns=${agent.nSteps}\x1b[0m`
     );
     console.error(
       `\x1b[1m[astra] tokens: prompt=${fmt(pt)} completion=${fmt(ct)} total=${fmt(pt + ct)} ` +
@@ -576,7 +580,7 @@ async function runBenchMatrix({ args, apiKey, models, reasonings, repeat, task,
       const status = r.error ? `model_error: ${r.error.split("\n")[0]}` : r.status;
       console.error(
         `  ${mark} ${String(r.model).padEnd(20)} ${String(r.reasoning).padEnd(8)} ` +
-          `${String(status).padEnd(16)} ${String(r.steps).padStart(3)} steps · ` +
+          `${String(status).padEnd(16)} ${String(r.steps).padStart(3)} turns · ` +
           `${fmt(r.total_tokens)} tok · ${fmtUsd(r.cost_usd)}` +
           `${r.cost_source === "estimated" ? "~" : ""}`
       );
@@ -620,7 +624,7 @@ function printLeaderboard(board) {
   board.sort((a, b) => b.completion_rate - a.completion_rate || a.avg_steps - b.avg_steps);
   console.error(
     `\n\x1b[1m${"model".padEnd(20)} ${"reason".padEnd(8)} ${"complete".padEnd(8)} ` +
-      `${"steps".padStart(6)} ${"tokens".padStart(9)}  ${"cost".padStart(9)}  source\x1b[0m`
+      `${"turns".padStart(6)} ${"tokens".padStart(9)}  ${"cost".padStart(9)}  source\x1b[0m`
   );
   for (const g of board) {
     const completed = `${g.completed}/${g.runs}`;
@@ -665,23 +669,22 @@ function printSessions() {
 
 function printEvent(msg) {
   // Bench mode keeps output compact: skip the verbose system/user/assistant
-  // message bodies (per-step progress is shown by printStep). Only surface a
+  // message bodies (per-turn progress is shown by printStep). Only surface a
   // non-empty terminal exit event so the run's submission is still visible.
   if (msg.role !== "exit" || !msg.content?.trim()) return;
   console.error(`\x1b[35m--- SUBMISSION ---\x1b[0m\n${msg.content}\n`);
 }
 
-/** Compact per-step status line with exact token usage from the API. */
+/** Compact per-turn status line with exact token usage from the API. */
 function printStep(s) {
-  const limit = s.stepLimit > 0 ? `/${s.stepLimit}` : "";
   const cached = s.usage.cached_tokens ? ` (cached ${fmt(s.usage.cached_tokens)})` : "";
   const cost = s.costUsd != null
-    ? ` · ${fmtUsd(s.costUsd)}${s.costKind === "estimated" ? "~" : ""}`
+    ? ` │ ${fmtUsd(s.costUsd)}${s.costKind === "estimated" ? "~" : ""}`
     : "";
   console.error(
-    `\x1b[36m[astra] step ${s.step}${limit} · call ${s.nCalls} · ` +
-      `↑${fmt(s.usage.prompt_tokens)} ↓${fmt(s.usage.completion_tokens)} tok · ` +
-      `ctx ${fmt(s.contextTokens)}${cached}${cost} · ${s.elapsedSeconds.toFixed(1)}s\x1b[0m`
+    `\x1b[36m[astra] turn ${s.step} │ ` +
+      `↑${fmt(s.usage.prompt_tokens)} ↓${fmt(s.usage.completion_tokens)} tok │ ` +
+      `ctx ${fmt(s.contextTokens)}${cached}${cost} │ ${s.elapsedSeconds.toFixed(1)}s\x1b[0m`
   );
 }
 

package/src/project.js

@@ -14,7 +14,7 @@ const DEFAULT_BENCH = {
   models: [],
   reasoning: [""],
   repeat: 1,
-  steps: 40,
+  steps: 0,
   wall_seconds: 0,
   command_timeout_seconds: 60,
 };
@@ -169,7 +169,7 @@ export function loadProject(projectPath) {
       models: stringArray(bench.models, "bench.models", DEFAULT_BENCH.models),
       reasoning: stringArray(bench.reasoning, "bench.reasoning", DEFAULT_BENCH.reasoning),
       repeat: positiveInteger(bench.repeat, "bench.repeat", DEFAULT_BENCH.repeat),
-      steps: positiveInteger(bench.steps, "bench.steps", DEFAULT_BENCH.steps),
+      steps: positiveInteger(bench.steps, "bench.steps", DEFAULT_BENCH.steps, { allowZero: true }),
       wall: positiveInteger(bench.wall_seconds, "bench.wall_seconds", DEFAULT_BENCH.wall_seconds, { allowZero: true }),
       timeout: positiveInteger(bench.command_timeout_seconds, "bench.command_timeout_seconds", DEFAULT_BENCH.command_timeout_seconds),
     },