@theokit/sdk 2.13.1 → 2.15.0

package/dist/define-tool.d.ts

@@ -1,4 +1,5 @@
 import type { z as ZodNamespace, ZodType } from "zod";
+import type { SanitizeOptions } from "./sanitize/types.js";
 import type { CustomTool } from "./types/agent.js";
 /**
  * Spec accepted by {@link defineTool}. `inputSchema` is a Zod schema; the
@@ -15,6 +16,13 @@ export interface DefineToolSpec<T extends ZodType> {
     inputSchema: T;
     /** Handler invoked with the parsed input. Type is inferred via `z.infer<T>`. */
     handler: (input: ZodNamespace.infer<T>) => string | Promise<string>;
+    /**
+     * Sanitize the raw model-emitted args BEFORE schema validation (`@theokit/sdk/sanitize`).
+     * `true` trims whitespace; an object opts into coercion / JSON-repair. Coercion is schema-aware
+     * against this tool's `inputSchema`. Absent ⇒ args reach validation untouched. Sanitize is
+     * hygiene, not a validity bypass — a genuinely invalid arg still raises `ZodError`.
+     */
+    sanitize?: boolean | SanitizeOptions;
 }
 /**
  * Type-safe builder for {@link CustomTool}. Converts a Zod schema to JSON

package/dist/{errors-9yw4UQwX.d.cts → errors-Bart0ptP.d.cts}

@@ -1,4 +1,4 @@
-import { p as RunOperation } from './run-TMdc7gmo.cjs';
+import { p as RunOperation } from './run-DXy_MVwz.cjs';
 
 /**
  * Public type contract for the Budget enforcement primitive

package/dist/{errors-DFiY-NHK.d.ts → errors-DJuuubJK.d.ts}

@@ -1,4 +1,4 @@
-import { p as RunOperation } from './run-TMdc7gmo.js';
+import { p as RunOperation } from './run-DXy_MVwz.js';
 
 /**
  * Public type contract for the Budget enforcement primitive

package/dist/errors.d.cts

@@ -1,2 +1,2 @@
-export { A as AgentDisposedError, c as AgentRunError, d as AgentRunErrorCode, e as AuthenticationError, g as BudgetExceededError, C as ConfigurationError, u as CredentialPoolExhaustedError, E as ErrorCode, m as ErrorMetadata, I as IntegrationNotConnectedError, n as InvalidTaskIdError, K as KnownAgentRunErrorCode, M as MemoryAdapterError, o as MemoryAdapterErrorCode, N as NetworkError, R as RateLimitError, p as TaskNotFoundError, T as TheokitAgentError, U as UnknownAgentError, q as UnsupportedBudgetOperationError, r as UnsupportedRunOperationError, s as UnsupportedTaskOperationError, t as isTransientError } from './errors-9yw4UQwX.cjs';
-import './run-TMdc7gmo.cjs';
+export { A as AgentDisposedError, c as AgentRunError, d as AgentRunErrorCode, e as AuthenticationError, g as BudgetExceededError, C as ConfigurationError, u as CredentialPoolExhaustedError, E as ErrorCode, m as ErrorMetadata, I as IntegrationNotConnectedError, n as InvalidTaskIdError, K as KnownAgentRunErrorCode, M as MemoryAdapterError, o as MemoryAdapterErrorCode, N as NetworkError, R as RateLimitError, p as TaskNotFoundError, T as TheokitAgentError, U as UnknownAgentError, q as UnsupportedBudgetOperationError, r as UnsupportedRunOperationError, s as UnsupportedTaskOperationError, t as isTransientError } from './errors-Bart0ptP.cjs';
+import './run-DXy_MVwz.cjs';

package/dist/eval.cjs

@@ -1457,6 +1457,7 @@ function isEmptyRound(result) {
   return (result.result ?? "").trim() === "";
 }
 function classifyRound(result, round, maxRounds, emptyStreak) {
+  if (result.stoppedByDoomLoop === true) return "no_progress";
   if (result.stoppedAtIterationLimit !== true) return "done";
   if (isEmptyRound(result) && emptyStreak >= 1) return "no_progress";
   if (round >= maxRounds) return "step_limit";
@@ -4490,6 +4491,7 @@ function applyScriptMetrics(base, script) {
   if (script.usage !== void 0) base.usage = script.usage;
   if (script.cost !== void 0) base.cost = script.cost;
   if (script.stoppedAtIterationLimit === true) base.stoppedAtIterationLimit = true;
+  if (script.stoppedByDoomLoop === true) base.stoppedByDoomLoop = true;
 }
 
 // src/internal/runtime/cloud/cloud-run.ts
@@ -8271,6 +8273,93 @@ function evaluateBudgetGate(tracker) {
   }
 }
 
+// src/internal/agent-loop/doom-loop-tracker.ts
+init_errors();
+function createDoomLoopTracker(option) {
+  if (option === false) return void 0;
+  return new DoomLoopTracker(option);
+}
+var DEFAULT_CONFIG = { softThreshold: 3, hardThreshold: 5 };
+function assertValidThresholds(soft, hard) {
+  for (const [label, value] of [
+    ["softThreshold", soft],
+    ["hardThreshold", hard]
+  ]) {
+    if (!Number.isInteger(value) || value < 1) {
+      throw new ConfigurationError(
+        `doomLoop.${label} must be a positive integer (received ${value}).`,
+        { code: "invalid_doom_loop_threshold" }
+      );
+    }
+  }
+}
+function sortKeys(value) {
+  if (value === null || typeof value !== "object") return value;
+  if (Array.isArray(value)) return value.map(sortKeys);
+  const out = {};
+  for (const key of Object.keys(value).sort()) {
+    out[key] = sortKeys(value[key]);
+  }
+  return out;
+}
+function signatureOf(call) {
+  const { input } = call;
+  let inputSig;
+  if (input === null || input === void 0) inputSig = "null";
+  else if (typeof input !== "object") inputSig = String(input);
+  else {
+    try {
+      inputSig = JSON.stringify(sortKeys(input)) ?? "null";
+    } catch {
+      inputSig = String(input);
+    }
+  }
+  return `${call.name}\0${inputSig}`;
+}
+var DoomLoopTracker = class {
+  #config;
+  #lastSignature = "";
+  #count = 0;
+  constructor(config) {
+    const softThreshold = config?.softThreshold ?? DEFAULT_CONFIG.softThreshold;
+    const hardThreshold = config?.hardThreshold ?? DEFAULT_CONFIG.hardThreshold;
+    assertValidThresholds(softThreshold, hardThreshold);
+    this.#config = { softThreshold, hardThreshold };
+  }
+  inspect(call) {
+    const signature = signatureOf(call);
+    this.#count = signature === this.#lastSignature ? this.#count + 1 : 1;
+    this.#lastSignature = signature;
+    const count = this.#count;
+    if (count >= this.#config.hardThreshold) {
+      return {
+        kind: "hard",
+        message: `Detected ${count} consecutive identical calls to \`${call.name}\`; stopping to avoid a loop.`
+      };
+    }
+    if (count === this.#config.softThreshold) {
+      return {
+        kind: "soft",
+        message: `Detected ${count} consecutive identical calls to \`${call.name}\`; try a different approach.`
+      };
+    }
+    return { kind: "ok" };
+  }
+  reset() {
+    this.#lastSignature = "";
+    this.#count = 0;
+  }
+};
+function firstDoomLoopVerdict(tracker, calls) {
+  let escalation = { kind: "ok" };
+  for (const call of calls) {
+    const v = tracker.inspect(call);
+    if (v.kind === "hard") return v;
+    if (v.kind === "soft" && escalation.kind === "ok") escalation = v;
+  }
+  return escalation;
+}
+
 // src/internal/budget/usage-accumulator.ts
 var UsageAccumulator = class {
   input = 0;
@@ -8451,6 +8540,7 @@ async function initLoopContext(inputs) {
     tools,
     finalText: "",
     finalStatus: "finished",
+    doomLoop: createDoomLoopTracker(inputs.doomLoop),
     usage: new UsageAccumulator(),
     nudgeAttempts: 0,
     stopFeedbackAttempts: 0,
@@ -9475,6 +9565,7 @@ async function runAgentLoop(inputs) {
       ctx.finalStatus = "error";
     }
     sendSpan?.setAttribute("status", ctx.finalStatus);
+    if (ctx.stoppedByDoomLoop === true) sendSpan?.setAttribute("stoppedByDoomLoop", true);
     if (inputs.telemetry?.includeContent === true && ctx.finalText.length > 0) {
       sendSpan?.addEvent("response", { content: ctx.finalText });
     }
@@ -9501,7 +9592,8 @@ async function runAgentLoop(inputs) {
       ...usage !== void 0 ? { usage } : {},
       ...cost !== void 0 ? { cost } : {},
       ...ctx.error !== void 0 ? { error: ctx.error } : {},
-      ...ctx.stoppedAtIterationLimit === true ? { stoppedAtIterationLimit: true } : {}
+      ...ctx.stoppedAtIterationLimit === true ? { stoppedAtIterationLimit: true } : {},
+      ...ctx.stoppedByDoomLoop === true ? { stoppedByDoomLoop: true } : {}
     };
   } finally {
     if (ctxRef !== void 0 && ctxRef.memoryProviderHandle !== void 0 && inputs.memoryProvider !== void 0) {
@@ -9660,8 +9752,26 @@ async function continueOrTerminate(inputs, ctx, llmOutput) {
     }
   }
   pushToolConversationSteps(ctx, llmOutput.toolCalls, toolResults);
+  if (await inspectDoomLoop(inputs, ctx, llmOutput.toolCalls) === "stop") return "done";
   return handleToolErrorContinuation(inputs, ctx, toolResults);
 }
+async function inspectDoomLoop(inputs, ctx, toolCalls) {
+  if (ctx.doomLoop === void 0) return "continue";
+  const verdict = firstDoomLoopVerdict(ctx.doomLoop, toolCalls);
+  if (verdict.kind === "hard") {
+    ctx.stoppedByDoomLoop = true;
+    await emitAssistantTextStep(
+      inputs,
+      ctx,
+      verdict.message ?? "Stopped: repeated identical tool calls made no progress."
+    );
+    return "stop";
+  }
+  if (verdict.kind === "soft") {
+    ctx.messages.push({ role: "user", content: [{ type: "text", text: verdict.message ?? "" }] });
+  }
+  return "continue";
+}
 
 // src/internal/llm/fallback-client.ts
 init_errors();
@@ -10950,6 +11060,115 @@ function toOllamaTools(tools) {
     }
   }));
 }
+var cachedJsonrepair;
+function loadJsonrepair() {
+  if (cachedJsonrepair === void 0) {
+    const req = module$1.createRequire((typeof document === 'undefined' ? require('u' + 'rl').pathToFileURL(__filename).href : (_documentCurrentScript && _documentCurrentScript.tagName.toUpperCase() === 'SCRIPT' && _documentCurrentScript.src || new URL('eval.cjs', document.baseURI).href)));
+    cachedJsonrepair = req("jsonrepair").jsonrepair;
+  }
+  return cachedJsonrepair;
+}
+function isPlainObject(v) {
+  return v !== null && typeof v === "object" && !Array.isArray(v);
+}
+function toFiniteNumber(raw) {
+  if (raw === "") return void 0;
+  const n = Number(raw);
+  return Number.isFinite(n) && String(n) === raw ? n : void 0;
+}
+function tryJson(raw, repair) {
+  const t = raw.trimStart();
+  if (!(t.startsWith("{") || t.startsWith("["))) return void 0;
+  try {
+    return JSON.parse(repair ? loadJsonrepair()(t) : t);
+  } catch {
+    return void 0;
+  }
+}
+function heuristicCoerce(raw, repairJson) {
+  if (raw === "true") return true;
+  if (raw === "false") return false;
+  if (raw === "null") return null;
+  const n = toFiniteNumber(raw);
+  if (n !== void 0) return n;
+  const json = tryJson(raw, false) ?? (repairJson ? tryJson(raw, true) : void 0);
+  return json === void 0 ? raw : json;
+}
+function coerceCandidates(raw, repairJson) {
+  const out = [];
+  if (raw === "true") out.push(true);
+  else if (raw === "false") out.push(false);
+  else if (raw === "null") out.push(null);
+  const n = toFiniteNumber(raw);
+  if (n !== void 0) out.push(n);
+  const json = tryJson(raw, false) ?? (repairJson ? tryJson(raw, true) : void 0);
+  if (json !== void 0) out.push(json);
+  out.push(raw);
+  return out;
+}
+function objectShape(schema) {
+  const shape = schema?.shape;
+  return shape !== null && typeof shape === "object" ? shape : void 0;
+}
+
+// src/sanitize/sanitize-tool-input.ts
+function applyTrim(key, value, ctx) {
+  const trimmed = value.trim();
+  if (trimmed !== value) ctx.notes.push(`trimmed "${key}"`);
+  return trimmed;
+}
+function applyCoerce(key, raw, ctx) {
+  const field = ctx.shape?.[key];
+  let coerced = raw;
+  if (field) {
+    for (const candidate of coerceCandidates(raw, ctx.repairJson)) {
+      if (field.safeParse(candidate).success) {
+        coerced = candidate;
+        break;
+      }
+    }
+  } else {
+    coerced = heuristicCoerce(raw, ctx.repairJson);
+  }
+  if (coerced !== raw) ctx.notes.push(`coerced "${key}"`);
+  return coerced;
+}
+function applyRepair(key, value, ctx) {
+  const repaired = tryJson(value, true);
+  if (repaired === void 0) return value;
+  ctx.notes.push(`repaired json "${key}"`);
+  return repaired;
+}
+function sanitizeString(key, value, ctx) {
+  let out = ctx.trim ? applyTrim(key, value, ctx) : value;
+  if (ctx.coerce && typeof out === "string") out = applyCoerce(key, out, ctx);
+  if (ctx.repairJson && !ctx.coerce && typeof out === "string") out = applyRepair(key, out, ctx);
+  return out;
+}
+function walk(input, ctx, depth) {
+  const out = {};
+  for (const [key, value] of Object.entries(input)) {
+    if (typeof value === "string") out[key] = sanitizeString(key, value, ctx);
+    else if (ctx.deep && depth < ctx.maxDepth && isPlainObject(value))
+      out[key] = walk(value, ctx, depth + 1);
+    else out[key] = value;
+  }
+  return out;
+}
+function sanitizeToolInput(input, options) {
+  if (!isPlainObject(input)) return { value: input, changed: false, notes: [] };
+  const ctx = {
+    trim: options?.trim,
+    coerce: options?.coerce ?? false,
+    repairJson: options?.repairJson ?? false,
+    deep: options?.deep ?? false,
+    maxDepth: options?.maxDepth ?? 8,
+    shape: objectShape(options?.schema),
+    notes: []
+  };
+  const value = walk(input, ctx, 0);
+  return { value, changed: ctx.notes.length > 0, notes: ctx.notes };
+}
 
 // src/internal/llm/hermes-tool-extract.ts
 var HERMES_BLOCK = /<function=\s*([^>\s]+)\s*>([\s\S]*?)<\/tool_call>/g;
@@ -10975,9 +11194,9 @@ function parseHermesParams(inner) {
     const key = param[1];
     const value = param[2];
     if (key === void 0 || value === void 0) continue;
-    input[key.trim()] = value.trim();
+    input[key.trim()] = value;
   }
-  return input;
+  return sanitizeToolInput(input, { trim: true }).value;
 }
 
 // src/internal/llm/openai.ts
@@ -12097,6 +12316,8 @@ function buildLoopInputs(options, runId, userText) {
     // M1-2: per-send iteration ceiling (validated above). The loop reads
     // inputs.maxIterations (default 8 when unset).
     ...maxIterations !== void 0 ? { maxIterations } : {},
+    // Doom-loop guard config (default on; `false` disables, object tunes thresholds).
+    ...options.sendOptions.doomLoop !== void 0 ? { doomLoop: options.sendOptions.doomLoop } : {},
     // D315-D317 — tool lifecycle hooks (cost tracking + audit + retry/alert)
     ...options.agentOptions.onToolStart !== void 0 ? { onToolStart: options.agentOptions.onToolStart } : {},
     ...options.agentOptions.onToolEnd !== void 0 ? { onToolEnd: options.agentOptions.onToolEnd } : {},
@@ -12229,6 +12450,7 @@ var RealLocalRun = class extends FixtureRunBase {
     if (output.usage !== void 0) this.script.usage = output.usage;
     if (output.cost !== void 0) this.script.cost = output.cost;
     if (output.stoppedAtIterationLimit === true) this.script.stoppedAtIterationLimit = true;
+    if (output.stoppedByDoomLoop === true) this.script.stoppedByDoomLoop = true;
     if (output.error !== void 0 && this.script.errorDetail === void 0) {
       this.script.errorDetail = {
         message: output.error.message,
@@ -15787,7 +16009,7 @@ var JsonlParseError = class extends Error {
   }
   line;
 };
-function isPlainObject(value) {
+function isPlainObject2(value) {
   return typeof value === "object" && value !== null && !Array.isArray(value);
 }
 function tryParseObjectLine(line) {
@@ -15798,7 +16020,7 @@ function tryParseObjectLine(line) {
   } catch {
     return void 0;
   }
-  return isPlainObject(parsed) ? parsed : void 0;
+  return isPlainObject2(parsed) ? parsed : void 0;
 }
 function loadJsonl(path, opts = {}) {
   const text = fs.readFileSync(path, "utf8");
@@ -15814,7 +16036,7 @@ function loadJsonl(path, opts = {}) {
     } catch {
       throw new JsonlParseError(`line ${lineNumber}: invalid JSON`, lineNumber);
     }
-    if (!isPlainObject(parsed)) {
+    if (!isPlainObject2(parsed)) {
       throw new JsonlParseError(`line ${lineNumber}: not a JSON object`, lineNumber);
     }
     out.push(opts.map ? opts.map(parsed, lineNumber) : parsed);