@theokit/sdk 4.17.1 → 4.18.1

package/dist/eval.cjs

@@ -6255,10 +6255,12 @@ __export(run_until_exports, {
 async function* runUntilImpl(agent, goal, options, deps) {
   const maxTurns = options?.maxTurns ?? 20;
   const maxFails = options?.maxConsecutiveJudgeFailures ?? 3;
+  const tokenBudget = options?.tokenBudget;
   const signal = options?.signal;
   const isAborted = () => signal !== void 0 && signal.aborted;
   let turn = 0;
   let consecutiveFailures = 0;
+  let tokensUsed = 0;
   let lastResponse = "";
   if (isAborted()) {
     yield {
@@ -6266,20 +6268,44 @@ async function* runUntilImpl(agent, goal, options, deps) {
       status: "paused",
       reason: "aborted via AbortSignal before first turn"
     };
-    return { status: "paused", turnsUsed: 0, finalResponse: void 0 };
+    return { status: "paused", turnsUsed: 0, tokensUsed, finalResponse: void 0 };
   }
   yield { type: "status_change", status: "active", reason: "Goal started" };
   while (turn < maxTurns) {
     if (isAborted()) {
       yield { type: "status_change", status: "paused", reason: "aborted via AbortSignal" };
-      return { status: "paused", turnsUsed: turn, finalResponse: lastResponse || void 0 };
+      return {
+        status: "paused",
+        turnsUsed: turn,
+        tokensUsed,
+        finalResponse: lastResponse || void 0
+      };
     }
     turn += 1;
     yield { type: "turn_start", turn, goal };
     const continuationPrompt = turn === 1 ? goal : composeContinuation(goal, lastResponse);
     const run = await agent.send(continuationPrompt);
+    const cancelOnAbort = () => {
+      void run.cancel?.();
+    };
+    if (signal !== void 0) {
+      if (signal.aborted) cancelOnAbort();
+      else signal.addEventListener("abort", cancelOnAbort, { once: true });
+    }
     const result = await run.wait();
+    if (signal !== void 0) signal.removeEventListener("abort", cancelOnAbort);
     lastResponse = result.result ?? "";
+    const turnTokens = result.usage?.totalTokens;
+    if (typeof turnTokens === "number") tokensUsed += turnTokens;
+    if (isAborted()) {
+      yield { type: "status_change", status: "paused", reason: "aborted via AbortSignal" };
+      return {
+        status: "paused",
+        turnsUsed: turn,
+        tokensUsed,
+        finalResponse: lastResponse || void 0
+      };
+    }
     yield { type: "agent_response", turn, content: lastResponse };
     const judgeOpts = {};
     if (options?.judgeModel !== void 0) judgeOpts.judgeModel = options.judgeModel;
@@ -6305,6 +6331,7 @@ async function* runUntilImpl(agent, goal, options, deps) {
         return {
           status: "failed",
           turnsUsed: turn,
+          tokensUsed,
           finalResponse: lastResponse || void 0
         };
       }
@@ -6313,7 +6340,12 @@ async function* runUntilImpl(agent, goal, options, deps) {
     }
     if (judgment.verdict === "done") {
       yield { type: "status_change", status: "completed", reason: judgment.reason };
-      return { status: "completed", turnsUsed: turn, finalResponse: lastResponse || void 0 };
+      return {
+        status: "completed",
+        turnsUsed: turn,
+        tokensUsed,
+        finalResponse: lastResponse || void 0
+      };
     }
     if (judgment.verdict === "skipped") {
       yield {
@@ -6321,7 +6353,25 @@ async function* runUntilImpl(agent, goal, options, deps) {
         status: "completed",
         reason: `skipped: ${judgment.reason}`
       };
-      return { status: "completed", turnsUsed: turn, finalResponse: lastResponse || void 0 };
+      return {
+        status: "completed",
+        turnsUsed: turn,
+        tokensUsed,
+        finalResponse: lastResponse || void 0
+      };
+    }
+    if (tokenBudget !== void 0 && tokensUsed >= tokenBudget) {
+      yield {
+        type: "status_change",
+        status: "budget_limited",
+        reason: `token budget (${tokenBudget}) reached: ${tokensUsed} used`
+      };
+      return {
+        status: "budget_limited",
+        turnsUsed: turn,
+        tokensUsed,
+        finalResponse: lastResponse || void 0
+      };
     }
     yield { type: "continuation", turn, prompt: continuationPrompt };
   }
@@ -6330,13 +6380,37 @@ async function* runUntilImpl(agent, goal, options, deps) {
     status: "failed",
     reason: `max turns (${maxTurns}) exhausted`
   };
-  return { status: "failed", turnsUsed: turn, finalResponse: lastResponse || void 0 };
+  return {
+    status: "failed",
+    turnsUsed: turn,
+    tokensUsed,
+    finalResponse: lastResponse || void 0
+  };
 }
 function composeContinuation(goal, lastResponse) {
-  return `Continue working toward the goal: ${goal}
-
-Your last response was:
-${lastResponse.slice(0, 1e3)}`;
+  return [
+    "Continue working toward the active goal.",
+    "",
+    `<objective>
+${goal}
+</objective>`,
+    "",
+    "Continuation behavior:",
+    "- This goal persists across turns. Keep the full objective intact; if it cannot be finished now,",
+    "  make concrete progress toward the real requested end state and do not redefine success around a",
+    "  smaller or easier task.",
+    "",
+    "Work from evidence:",
+    "- Use the current worktree and external state as authoritative. Inspect the current state before",
+    "  relying on it. Improve, replace, or remove existing work as needed to satisfy the actual objective.",
+    "",
+    "Completion audit:",
+    "- Before deciding the goal is achieved, treat completion as unproven and verify it against the actual",
+    "  current state, requirement by requirement. Treat uncertain or indirect evidence as not achieved.",
+    "",
+    `Your last response was:
+${lastResponse.slice(0, 1e3)}`
+  ].join("\n");
 }
 var init_run_until = __esm({
   "src/internal/runtime/lifecycle/run-until.ts"() {
@@ -18576,7 +18650,7 @@ function localAgentRunUntil(agent, goal, options) {
         status: "paused",
         reason: "runUntil() requires an explicit goal (durable objectives removed in v4.0)"
       };
-      return { status: "paused", turnsUsed: 0, finalResponse: void 0 };
+      return { status: "paused", turnsUsed: 0, tokensUsed: 0, finalResponse: void 0 };
     }
     return yield* runUntilImpl2(agent, goal, options, await buildDeps());
   }