@tangle-network/agent-app 0.43.66 → 0.43.68

package/dist/chunk-UP33Z633.js.map

@@ -0,0 +1 @@
+{"version":3,"sources":["../src/work-product/queue.ts"],"sourcesContent":["/**\n * The review queue is a PROJECTION, not a store — a client-safe pure fold of\n * existing sources into queue items (the `/missions` events.ts pattern: pure\n * data, re-validation at JSON boundaries). The only genuinely-new durable\n * state behind it is the {@link WorkProductRecord} row and its status\n * machine; everything else reads what already exists:\n *\n *  - intake: a chat thread for the engagement scope with NO record yet\n *  - missing_info: the open record's thread has a PENDING `/interactions` ask\n *  - working: record status `draft` (the live token tail stays on the chat\n *    surface's existing running-turns endpoint — the projection tracks no\n *    live runs, per the reuse-the-primitive invariant)\n *  - ready_for_review / changes_requested / approved / blocked: read directly\n *    off `WorkProductRecord.status` (blocked surfaces its unresolved count)\n */\n\nimport {\n  isWorkProductStatus,\n  unresolvedBlockingExceptions,\n  type WorkProductProvenance,\n  type WorkProductRecord,\n  type WorkProductRef,\n} from './types'\n\nexport type ReviewQueueState =\n  | 'intake'\n  | 'missing_info'\n  | 'working'\n  | 'ready_for_review'\n  | 'changes_requested'\n  | 'approved'\n  | 'blocked'\n\n/** One row of the review queue projection for an engagement scope */\nexport interface ReviewQueueItem {\n  scopeKey: string\n  state: ReviewQueueState\n  threadId: string | null\n  workProduct?: WorkProductRef & { title: string; kind: string }\n  /** The pending `/interactions` ask parking this scope, when any. */\n  pendingAsk?: { interactionId: string; title: string }\n  blockingExceptions: number\n  failedChecks: number\n  provenance?: Pick<WorkProductProvenance, 'profileHash' | 'servingModels'>\n  updatedAt: number\n}\n\n/** An engagement-scoped chat thread — the intake candidate source. Products\n *  that scope threads already carry a scopeKey-style column. */\nexport interface ReviewQueueThread {\n  scopeKey: string\n  threadId: string\n  updatedAt: number\n}\n\n/** A pending `/interactions` ask on a thread (from the existing list\n *  endpoint) — the missing_info source. */\nexport interface ReviewQueuePendingAsk {\n  threadId: string\n  interactionId: string\n  title: string\n}\n\n/** Existing-source inputs the projection folds — no new stores */\nexport interface ReviewQueueInputs {\n  workProducts: readonly WorkProductRecord[]\n  /** Engagement threads with no work product yet → intake items. */\n  threads?: readonly ReviewQueueThread[]\n  /** Pending asks by thread → missing_info override on open records. */\n  pendingAsks?: readonly ReviewQueuePendingAsk[]\n}\n\nconst OPEN_STATUSES = new Set(['draft', 'blocked', 'ready', 'changes_requested'])\n\n/** The record that represents a scope in the queue: its single open row when\n *  one exists, else its latest approved version. Superseded rows are history\n *  and never surface. */\nfunction currentRecordPerScope(records: readonly WorkProductRecord[]): Map<string, WorkProductRecord> {\n  const byScope = new Map<string, WorkProductRecord>()\n  for (const record of records) {\n    if (record.status === 'superseded') continue\n    const held = byScope.get(record.scopeKey)\n    if (!held) {\n      byScope.set(record.scopeKey, record)\n      continue\n    }\n    const heldOpen = OPEN_STATUSES.has(held.status)\n    const recordOpen = OPEN_STATUSES.has(record.status)\n    if (recordOpen !== heldOpen) {\n      if (recordOpen) byScope.set(record.scopeKey, record)\n      continue\n    }\n    if (record.version > held.version || (record.version === held.version && record.updatedAt > held.updatedAt)) {\n      byScope.set(record.scopeKey, record)\n    }\n  }\n  return byScope\n}\n\nfunction stateOf(record: WorkProductRecord, pendingAsk: ReviewQueuePendingAsk | undefined): ReviewQueueState {\n  switch (record.status) {\n    case 'ready':\n      return 'ready_for_review'\n    case 'changes_requested':\n      return 'changes_requested'\n    case 'approved':\n      return 'approved'\n    case 'blocked':\n      return pendingAsk ? 'missing_info' : 'blocked'\n    case 'draft':\n      return pendingAsk ? 'missing_info' : 'working'\n    // 'superseded' is filtered before this switch.\n    default:\n      return 'working'\n  }\n}\n\n/** Fold the existing sources into queue items, newest first. */\nexport function projectReviewQueue(inputs: ReviewQueueInputs): ReviewQueueItem[] {\n  const asksByThread = new Map<string, ReviewQueuePendingAsk>()\n  for (const ask of inputs.pendingAsks ?? []) {\n    if (!asksByThread.has(ask.threadId)) asksByThread.set(ask.threadId, ask)\n  }\n\n  const items: ReviewQueueItem[] = []\n  const byScope = currentRecordPerScope(inputs.workProducts)\n  for (const [scopeKey, record] of byScope) {\n    const pendingAsk = record.threadId ? asksByThread.get(record.threadId) : undefined\n    const item: ReviewQueueItem = {\n      scopeKey,\n      state: stateOf(record, pendingAsk),\n      threadId: record.threadId,\n      workProduct: {\n        id: record.id,\n        version: record.version,\n        title: record.artifact?.title ?? scopeKey,\n        kind: record.artifact?.kind ?? '',\n      },\n      blockingExceptions: unresolvedBlockingExceptions(record.exceptions).length,\n      failedChecks: record.checks.filter((check) => !check.passed).length,\n      provenance: { profileHash: record.provenance.profileHash, servingModels: record.provenance.servingModels },\n      updatedAt: record.updatedAt,\n    }\n    if (pendingAsk) item.pendingAsk = { interactionId: pendingAsk.interactionId, title: pendingAsk.title }\n    items.push(item)\n  }\n\n  // Intake: an engagement thread with no record for its scope.\n  for (const thread of inputs.threads ?? []) {\n    if (byScope.has(thread.scopeKey)) continue\n    if (items.some((item) => item.scopeKey === thread.scopeKey)) continue\n    items.push({\n      scopeKey: thread.scopeKey,\n      state: 'intake',\n      threadId: thread.threadId,\n      blockingExceptions: 0,\n      failedChecks: 0,\n      updatedAt: thread.updatedAt,\n    })\n  }\n\n  return items.sort((a, b) => b.updatedAt - a.updatedAt)\n}\n\n/** Re-validate one JSON-boundary row into a queue item; null for junk. The\n *  client-side twin of the server projection, for payloads that cross a\n *  fetch boundary. */\nexport function parseReviewQueueItem(raw: unknown): ReviewQueueItem | null {\n  if (!raw || typeof raw !== 'object' || Array.isArray(raw)) return null\n  const record = raw as Record<string, unknown>\n  const states: readonly ReviewQueueState[] = [\n    'intake',\n    'missing_info',\n    'working',\n    'ready_for_review',\n    'changes_requested',\n    'approved',\n    'blocked',\n  ]\n  if (typeof record.scopeKey !== 'string' || record.scopeKey.length === 0) return null\n  if (!states.includes(record.state as ReviewQueueState)) return null\n  if (record.threadId !== null && typeof record.threadId !== 'string') return null\n  if (typeof record.blockingExceptions !== 'number' || typeof record.failedChecks !== 'number') return null\n  if (typeof record.updatedAt !== 'number') return null\n  const item: ReviewQueueItem = {\n    scopeKey: record.scopeKey,\n    state: record.state as ReviewQueueState,\n    threadId: record.threadId,\n    blockingExceptions: record.blockingExceptions,\n    failedChecks: record.failedChecks,\n    updatedAt: record.updatedAt,\n  }\n  const workProduct = record.workProduct as Record<string, unknown> | undefined\n  if (workProduct && typeof workProduct === 'object') {\n    if (\n      typeof workProduct.id === 'string' &&\n      typeof workProduct.version === 'number' &&\n      typeof workProduct.title === 'string' &&\n      typeof workProduct.kind === 'string'\n    ) {\n      item.workProduct = {\n        id: workProduct.id,\n        version: workProduct.version,\n        title: workProduct.title,\n        kind: workProduct.kind,\n      }\n    }\n  }\n  const pendingAsk = record.pendingAsk as Record<string, unknown> | undefined\n  if (pendingAsk && typeof pendingAsk === 'object') {\n    if (typeof pendingAsk.interactionId === 'string' && typeof pendingAsk.title === 'string') {\n      item.pendingAsk = { interactionId: pendingAsk.interactionId, title: pendingAsk.title }\n    }\n  }\n  const provenance = record.provenance as Record<string, unknown> | undefined\n  if (provenance && typeof provenance === 'object') {\n    if (\n      typeof provenance.profileHash === 'string' &&\n      Array.isArray(provenance.servingModels) &&\n      provenance.servingModels.every((model) => typeof model === 'string')\n    ) {\n      item.provenance = { profileHash: provenance.profileHash, servingModels: provenance.servingModels as string[] }\n    }\n  }\n  return item\n}\n\n/** Convenience guard used when a status string crosses a JSON boundary. */\nexport { isWorkProductStatus }\n"],"mappings":";;;;;AAwEA,IAAM,gBAAgB,oBAAI,IAAI,CAAC,SAAS,WAAW,SAAS,mBAAmB,CAAC;AAKhF,SAAS,sBAAsB,SAAuE;AACpG,QAAM,UAAU,oBAAI,IAA+B;AACnD,aAAW,UAAU,SAAS;AAC5B,QAAI,OAAO,WAAW,aAAc;AACpC,UAAM,OAAO,QAAQ,IAAI,OAAO,QAAQ;AACxC,QAAI,CAAC,MAAM;AACT,cAAQ,IAAI,OAAO,UAAU,MAAM;AACnC;AAAA,IACF;AACA,UAAM,WAAW,cAAc,IAAI,KAAK,MAAM;AAC9C,UAAM,aAAa,cAAc,IAAI,OAAO,MAAM;AAClD,QAAI,eAAe,UAAU;AAC3B,UAAI,WAAY,SAAQ,IAAI,OAAO,UAAU,MAAM;AACnD;AAAA,IACF;AACA,QAAI,OAAO,UAAU,KAAK,WAAY,OAAO,YAAY,KAAK,WAAW,OAAO,YAAY,KAAK,WAAY;AAC3G,cAAQ,IAAI,OAAO,UAAU,MAAM;AAAA,IACrC;AAAA,EACF;AACA,SAAO;AACT;AAEA,SAAS,QAAQ,QAA2B,YAAiE;AAC3G,UAAQ,OAAO,QAAQ;AAAA,IACrB,KAAK;AACH,aAAO;AAAA,IACT,KAAK;AACH,aAAO;AAAA,IACT,KAAK;AACH,aAAO;AAAA,IACT,KAAK;AACH,aAAO,aAAa,iBAAiB;AAAA,IACvC,KAAK;AACH,aAAO,aAAa,iBAAiB;AAAA;AAAA,IAEvC;AACE,aAAO;AAAA,EACX;AACF;AAGO,SAAS,mBAAmB,QAA8C;AAC/E,QAAM,eAAe,oBAAI,IAAmC;AAC5D,aAAW,OAAO,OAAO,eAAe,CAAC,GAAG;AAC1C,QAAI,CAAC,aAAa,IAAI,IAAI,QAAQ,EAAG,cAAa,IAAI,IAAI,UAAU,GAAG;AAAA,EACzE;AAEA,QAAM,QAA2B,CAAC;AAClC,QAAM,UAAU,sBAAsB,OAAO,YAAY;AACzD,aAAW,CAAC,UAAU,MAAM,KAAK,SAAS;AACxC,UAAM,aAAa,OAAO,WAAW,aAAa,IAAI,OAAO,QAAQ,IAAI;AACzE,UAAM,OAAwB;AAAA,MAC5B;AAAA,MACA,OAAO,QAAQ,QAAQ,UAAU;AAAA,MACjC,UAAU,OAAO;AAAA,MACjB,aAAa;AAAA,QACX,IAAI,OAAO;AAAA,QACX,SAAS,OAAO;AAAA,QAChB,OAAO,OAAO,UAAU,SAAS;AAAA,QACjC,MAAM,OAAO,UAAU,QAAQ;AAAA,MACjC;AAAA,MACA,oBAAoB,6BAA6B,OAAO,UAAU,EAAE;AAAA,MACpE,cAAc,OAAO,OAAO,OAAO,CAAC,UAAU,CAAC,MAAM,MAAM,EAAE;AAAA,MAC7D,YAAY,EAAE,aAAa,OAAO,WAAW,aAAa,eAAe,OAAO,WAAW,cAAc;AAAA,MACzG,WAAW,OAAO;AAAA,IACpB;AACA,QAAI,WAAY,MAAK,aAAa,EAAE,eAAe,WAAW,eAAe,OAAO,WAAW,MAAM;AACrG,UAAM,KAAK,IAAI;AAAA,EACjB;AAGA,aAAW,UAAU,OAAO,WAAW,CAAC,GAAG;AACzC,QAAI,QAAQ,IAAI,OAAO,QAAQ,EAAG;AAClC,QAAI,MAAM,KAAK,CAAC,SAAS,KAAK,aAAa,OAAO,QAAQ,EAAG;AAC7D,UAAM,KAAK;AAAA,MACT,UAAU,OAAO;AAAA,MACjB,OAAO;AAAA,MACP,UAAU,OAAO;AAAA,MACjB,oBAAoB;AAAA,MACpB,cAAc;AAAA,MACd,WAAW,OAAO;AAAA,IACpB,CAAC;AAAA,EACH;AAEA,SAAO,MAAM,KAAK,CAAC,GAAG,MAAM,EAAE,YAAY,EAAE,SAAS;AACvD;AAKO,SAAS,qBAAqB,KAAsC;AACzE,MAAI,CAAC,OAAO,OAAO,QAAQ,YAAY,MAAM,QAAQ,GAAG,EAAG,QAAO;AAClE,QAAM,SAAS;AACf,QAAM,SAAsC;AAAA,IAC1C;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,EACF;AACA,MAAI,OAAO,OAAO,aAAa,YAAY,OAAO,SAAS,WAAW,EAAG,QAAO;AAChF,MAAI,CAAC,OAAO,SAAS,OAAO,KAAyB,EAAG,QAAO;AAC/D,MAAI,OAAO,aAAa,QAAQ,OAAO,OAAO,aAAa,SAAU,QAAO;AAC5E,MAAI,OAAO,OAAO,uBAAuB,YAAY,OAAO,OAAO,iBAAiB,SAAU,QAAO;AACrG,MAAI,OAAO,OAAO,cAAc,SAAU,QAAO;AACjD,QAAM,OAAwB;AAAA,IAC5B,UAAU,OAAO;AAAA,IACjB,OAAO,OAAO;AAAA,IACd,UAAU,OAAO;AAAA,IACjB,oBAAoB,OAAO;AAAA,IAC3B,cAAc,OAAO;AAAA,IACrB,WAAW,OAAO;AAAA,EACpB;AACA,QAAM,cAAc,OAAO;AAC3B,MAAI,eAAe,OAAO,gBAAgB,UAAU;AAClD,QACE,OAAO,YAAY,OAAO,YAC1B,OAAO,YAAY,YAAY,YAC/B,OAAO,YAAY,UAAU,YAC7B,OAAO,YAAY,SAAS,UAC5B;AACA,WAAK,cAAc;AAAA,QACjB,IAAI,YAAY;AAAA,QAChB,SAAS,YAAY;AAAA,QACrB,OAAO,YAAY;AAAA,QACnB,MAAM,YAAY;AAAA,MACpB;AAAA,IACF;AAAA,EACF;AACA,QAAM,aAAa,OAAO;AAC1B,MAAI,cAAc,OAAO,eAAe,UAAU;AAChD,QAAI,OAAO,WAAW,kBAAkB,YAAY,OAAO,WAAW,UAAU,UAAU;AACxF,WAAK,aAAa,EAAE,eAAe,WAAW,eAAe,OAAO,WAAW,MAAM;AAAA,IACvF;AAAA,EACF;AACA,QAAM,aAAa,OAAO;AAC1B,MAAI,cAAc,OAAO,eAAe,UAAU;AAChD,QACE,OAAO,WAAW,gBAAgB,YAClC,MAAM,QAAQ,WAAW,aAAa,KACtC,WAAW,cAAc,MAAM,CAAC,UAAU,OAAO,UAAU,QAAQ,GACnE;AACA,WAAK,aAAa,EAAE,aAAa,WAAW,aAAa,eAAe,WAAW,cAA0B;AAAA,IAC/G;AAAA,EACF;AACA,SAAO;AACT;","names":[]}

package/dist/{chunk-FMDMI25K.js → chunk-V55WJSR4.js}

@@ -1,6 +1,6 @@
 import {
   formatBytes
-} from "./chunk-X3N2H6JE.js";
+} from "./chunk-AFNTRJQ7.js";
 
 // src/chat-routes/binary-sniff.ts
 function bytesStartWith(bytes, offset, signature) {
@@ -236,4 +236,4 @@ export {
   attachmentTotalSizeErrorMessage,
   createSandboxFileIndexRoute
 };
-//# sourceMappingURL=chunk-FMDMI25K.js.map
+//# sourceMappingURL=chunk-V55WJSR4.js.map

package/dist/{chunk-YKBDH2UY.js → chunk-WL7XHLDK.js}

@@ -10,7 +10,7 @@ import {
 import {
   dispatchAppTool,
   outcomeStatus
-} from "./chunk-7CTIUCQ4.js";
+} from "./chunk-UOAYS72M.js";
 
 // src/tools/capability.ts
 async function createCapabilityToken(userId, opts) {
@@ -126,4 +126,4 @@ export {
   restrictTaxonomy,
   handleAppToolRequest
 };
-//# sourceMappingURL=chunk-YKBDH2UY.js.map
+//# sourceMappingURL=chunk-WL7XHLDK.js.map

package/dist/{chunk-7CTIUCQ4.js → chunk-YEFFHORB.js}

@@ -105,74 +105,6 @@ function findCustomTool(name, tools) {
   return tools?.find((t) => t.name === name);
 }
 
-// src/tools/dispatch.ts
-async function dispatchAppTool(toolName, rawArgs, ctx, opts) {
-  try {
-    if (!isAppToolName(toolName)) {
-      const custom = findCustomTool(toolName, opts.customTools);
-      if (!custom) return { ok: false, code: "unknown_tool", message: `${toolName} is not an app tool.` };
-      const result = await custom.execute(rawArgs, ctx);
-      return { ok: true, result };
-    }
-    if (toolName === "submit_proposal") {
-      const type = String(rawArgs.type ?? "").trim();
-      const title = String(rawArgs.title ?? "").trim();
-      if (!type || !opts.taxonomy.proposalTypes.includes(type)) {
-        return { ok: false, code: "invalid_type", message: `type must be one of: ${opts.taxonomy.proposalTypes.join(", ")}.` };
-      }
-      if (!title) return { ok: false, code: "missing_title", message: "title is required." };
-      const description = rawArgs.description == null ? null : String(rawArgs.description);
-      let regulated = opts.taxonomy.regulatedTypes.includes(type);
-      if (opts.needsApproval) {
-        try {
-          regulated = await opts.needsApproval(type, { title, description }, ctx);
-        } catch {
-          regulated = true;
-        }
-      }
-      const r2 = await opts.handlers.submitProposal({ type, title, description, regulated }, ctx);
-      const { proposalId, deduped, status, ...extra } = r2;
-      const effectiveStatus = status ?? "queued_for_approval";
-      opts.onProduced?.({
-        type: "proposal_created",
-        proposalId,
-        title,
-        status: effectiveStatus === "executed" ? "executed" : "pending",
-        content: description ?? void 0
-      });
-      return { ok: true, result: { ...extra, status: effectiveStatus, proposalId, deduped, regulated } };
-    }
-    if (toolName === "schedule_followup") {
-      const r2 = await opts.handlers.scheduleFollowup(
-        { title: String(rawArgs.title ?? ""), dueDate: String(rawArgs.dueDate ?? ""), priority: rawArgs.priority },
-        ctx
-      );
-      return { ok: true, result: { followupId: r2.id, dueDate: r2.dueDate, deduped: r2.deduped } };
-    }
-    if (toolName === "render_ui") {
-      const r2 = await opts.handlers.renderUi({ title: String(rawArgs.title ?? ""), schema: rawArgs.schema }, ctx);
-      opts.onProduced?.({ type: "artifact", path: r2.path, content: r2.content });
-      return { ok: true, result: { path: r2.path } };
-    }
-    const r = await opts.handlers.addCitation(
-      { path: String(rawArgs.path ?? ""), quote: String(rawArgs.quote ?? ""), label: rawArgs.label },
-      ctx
-    );
-    return { ok: true, result: { citationId: r.citationId, path: r.path } };
-  } catch (err) {
-    if (err instanceof ToolInputError) return { ok: false, code: err.code, message: err.message, status: err.status };
-    return { ok: false, code: "app_tool_error", message: err instanceof Error ? err.message : String(err), status: 500 };
-  }
-}
-function outcomeStatus(outcome) {
-  return outcome.status ?? 400;
-}
-
-// src/tools/runtime.ts
-function createAppToolRuntimeExecutor(opts) {
-  return ({ toolName, args }) => dispatchAppTool(toolName, args, opts.ctx, opts);
-}
-
 export {
   ToolInputError,
   APP_TOOL_NAMES,
@@ -180,9 +112,6 @@ export {
   buildAppToolOpenAITools,
   defineAppTool,
   customToolToOpenAI,
-  findCustomTool,
-  dispatchAppTool,
-  outcomeStatus,
-  createAppToolRuntimeExecutor
+  findCustomTool
 };
-//# sourceMappingURL=chunk-7CTIUCQ4.js.map
+//# sourceMappingURL=chunk-YEFFHORB.js.map

package/dist/chunk-YEFFHORB.js.map

@@ -0,0 +1 @@
+{"version":3,"sources":["../src/tools/errors.ts","../src/tools/openai.ts","../src/tools/registry.ts"],"sourcesContent":["/** A correctable bad-input error a tool handler throws; the HTTP layer maps it\n *  to a 4xx with the code, the runtime layer to a failed tool_result. So the\n *  agent learns the call failed and can correct, instead of a silent success. */\nexport class ToolInputError extends Error {\n  constructor(\n    public code: string,\n    message: string,\n    public status = 400,\n  ) {\n    super(message)\n    this.name = 'ToolInputError'\n  }\n}\n","import type { AppToolTaxonomy, BuildAppToolsOptions } from './types'\n\n/** The four canonical app-tool names. Stable identifiers the model calls in\n *  both the sandbox (MCP server name) and runtime (function-tool name) paths. */\nexport const APP_TOOL_NAMES = ['submit_proposal', 'schedule_followup', 'render_ui', 'add_citation'] as const\n/** Resolve a valid application tool name from the predefined list of tool names */\nexport type AppToolName = (typeof APP_TOOL_NAMES)[number]\n\nconst NAME_SET = new Set<string>(APP_TOOL_NAMES)\n/** Determine if a string matches a valid application tool name */\nexport function isAppToolName(name: string): name is AppToolName {\n  return NAME_SET.has(name)\n}\n\n/** A minimal OpenAI Chat Completions function-tool shape — structurally\n *  compatible with `@tangle-network/agent-runtime`'s `OpenAIChatTool` without\n *  importing it (keeps this package runtime-free). */\nexport interface OpenAIFunctionTool {\n  type: 'function'\n  function: {\n    name: string\n    description: string\n    parameters: Record<string, unknown>\n  }\n}\n\n/**\n * Build the four app tools in OpenAI function-tool shape. `submit_proposal`'s\n * `type` enum is the product's {@link AppToolTaxonomy.proposalTypes}; the\n * model-facing descriptions and the follow-up priority enum default to the\n * Tangle reference vocabulary and can be retuned via {@link BuildAppToolsOptions}\n * (the tool names + JSON-Schema shapes stay fixed — they are mechanism). Pass\n * the result to the agent-runtime backend's `tools`.\n */\nexport function buildAppToolOpenAITools(\n  taxonomy: AppToolTaxonomy,\n  opts?: BuildAppToolsOptions,\n): OpenAIFunctionTool[] {\n  const d = opts?.descriptions\n  const priorityValues = opts?.priorityValues ?? ['low', 'medium', 'high']\n  const custom: OpenAIFunctionTool[] = (opts?.customTools ?? []).map((t) => ({\n    type: 'function',\n    function: { name: t.name, description: t.description, parameters: t.parameters },\n  }))\n  return [\n    {\n      type: 'function',\n      function: {\n        name: 'submit_proposal',\n        description:\n          d?.submit_proposal ??\n          'Route a regulated or state-changing action to a human for approval (a recommendation, contacting/soliciting a contact, outreach, a record/account change, scheduling). Queues it for a named certified human to approve before it executes.',\n        parameters: {\n          type: 'object',\n          properties: {\n            type: { type: 'string', enum: [...taxonomy.proposalTypes] },\n            title: { type: 'string', description: 'Short label for the approval queue.' },\n            description: { type: 'string', description: 'The full drafted message/recommendation, with sources.' },\n          },\n          required: ['type', 'title'],\n        },\n      },\n    },\n    {\n      type: 'function',\n      function: {\n        name: 'schedule_followup',\n        description:\n          d?.schedule_followup ??\n          'Register a dated cadence step (a reminder, chase, or check-in) on the follow-up calendar. Executes immediately.',\n        parameters: {\n          type: 'object',\n          properties: {\n            title: { type: 'string' },\n            dueDate: { type: 'string', description: 'ISO date YYYY-MM-DD.' },\n            priority: { type: 'string', enum: [...priorityValues] },\n          },\n          required: ['title', 'dueDate'],\n        },\n      },\n    },\n    {\n      type: 'function',\n      function: {\n        name: 'render_ui',\n        description:\n          d?.render_ui ??\n          'Show a generated view live in the workspace. Validates the OpenUI JSON and persists the artifact. Executes immediately.',\n        parameters: {\n          type: 'object',\n          properties: {\n            title: { type: 'string' },\n            schema: { type: 'object', description: 'The OpenUI JSON object.' },\n          },\n          required: ['title', 'schema'],\n        },\n      },\n    },\n    {\n      type: 'function',\n      function: {\n        name: 'add_citation',\n        description:\n          d?.add_citation ??\n          'Anchor a grounding reference: the exact quote from a file backing a figure or claim. Verifies the quote appears in the file. Executes immediately.',\n        parameters: {\n          type: 'object',\n          properties: {\n            path: { type: 'string', description: 'The vault file path.' },\n            quote: { type: 'string', description: 'The exact text from it.' },\n          },\n          required: ['path', 'quote'],\n        },\n      },\n    },\n    ...custom,\n  ]\n}\n","import { isAppToolName } from './openai'\nimport type { AppToolContext } from './types'\nimport type { OpenAIFunctionTool } from './openai'\n\n/**\n * A product-defined app tool — the open registration seam.\n *\n * The four built-ins (`submit_proposal`/`schedule_followup`/`render_ui`/\n * `add_citation`) are mechanism and stay hard-typed. This is how a product adds\n * a fifth+ tool (e.g. gtm-agent's `set_config`) WITHOUT forking the shell: the\n * `name` + JSON-Schema `parameters` are what the model sees, and `execute` is\n * dispatched through the SAME validation/outcome path as the built-ins — a\n * thrown {@link ToolInputError} becomes a correctable 4xx, any other throw an\n * internal error, and a tool call never silently \"succeeds\" without its effect.\n */\nexport interface AppToolDefinition<Args = Record<string, unknown>> {\n  /** Stable identifier the model calls (and the MCP server name). Must not\n   *  collide with a built-in app tool. */\n  name: string\n  /** Model-facing description. */\n  description: string\n  /** JSON-Schema for the parameters (the OpenAI `function.parameters` object). */\n  parameters: Record<string, unknown>\n  /** Default route path the per-turn MCP server / HTTP handler is mounted at.\n   *  Overridable per call via `paths`/`buildHttpMcpServer`. */\n  path?: string\n  /** Fulfil the call; the return value is the tool result the model sees.\n   *  `ctx` is the trusted per-turn identity (never from tool args). */\n  execute: (args: Args, ctx: AppToolContext) => Promise<unknown> | unknown\n}\n\n/**\n * Validate + brand a product tool definition. Throws when the name is empty or\n * collides with a built-in (those are reserved mechanism). Identity otherwise —\n * call it at module scope so a bad definition fails at boot, not first use.\n */\nexport function defineAppTool<Args = Record<string, unknown>>(def: AppToolDefinition<Args>): AppToolDefinition<Args> {\n  const name = def.name?.trim()\n  if (!name) throw new Error('defineAppTool: name is required')\n  if (isAppToolName(name)) throw new Error(`defineAppTool: \"${name}\" is a built-in app tool — choose a different name`)\n  if (typeof def.execute !== 'function') throw new Error(`defineAppTool: \"${name}\" needs an execute() handler`)\n  return def\n}\n\n/** The OpenAI function-tool def for a custom tool — appended to the built-ins by\n *  `buildAppToolOpenAITools`. */\nexport function customToolToOpenAI(def: AppToolDefinition): OpenAIFunctionTool {\n  return { type: 'function', function: { name: def.name, description: def.description, parameters: def.parameters } }\n}\n\n/** Find a registered custom tool by the name the model called. */\nexport function findCustomTool(\n  name: string,\n  tools: readonly AppToolDefinition[] | undefined,\n): AppToolDefinition | undefined {\n  return tools?.find((t) => t.name === name)\n}\n"],"mappings":";AAGO,IAAM,iBAAN,cAA6B,MAAM;AAAA,EACxC,YACS,MACP,SACO,SAAS,KAChB;AACA,UAAM,OAAO;AAJN;AAEA;AAGP,SAAK,OAAO;AAAA,EACd;AAAA,EANS;AAAA,EAEA;AAKX;;;ACRO,IAAM,iBAAiB,CAAC,mBAAmB,qBAAqB,aAAa,cAAc;AAIlG,IAAM,WAAW,IAAI,IAAY,cAAc;AAExC,SAAS,cAAc,MAAmC;AAC/D,SAAO,SAAS,IAAI,IAAI;AAC1B;AAsBO,SAAS,wBACd,UACA,MACsB;AACtB,QAAM,IAAI,MAAM;AAChB,QAAM,iBAAiB,MAAM,kBAAkB,CAAC,OAAO,UAAU,MAAM;AACvE,QAAM,UAAgC,MAAM,eAAe,CAAC,GAAG,IAAI,CAAC,OAAO;AAAA,IACzE,MAAM;AAAA,IACN,UAAU,EAAE,MAAM,EAAE,MAAM,aAAa,EAAE,aAAa,YAAY,EAAE,WAAW;AAAA,EACjF,EAAE;AACF,SAAO;AAAA,IACL;AAAA,MACE,MAAM;AAAA,MACN,UAAU;AAAA,QACR,MAAM;AAAA,QACN,aACE,GAAG,mBACH;AAAA,QACF,YAAY;AAAA,UACV,MAAM;AAAA,UACN,YAAY;AAAA,YACV,MAAM,EAAE,MAAM,UAAU,MAAM,CAAC,GAAG,SAAS,aAAa,EAAE;AAAA,YAC1D,OAAO,EAAE,MAAM,UAAU,aAAa,sCAAsC;AAAA,YAC5E,aAAa,EAAE,MAAM,UAAU,aAAa,yDAAyD;AAAA,UACvG;AAAA,UACA,UAAU,CAAC,QAAQ,OAAO;AAAA,QAC5B;AAAA,MACF;AAAA,IACF;AAAA,IACA;AAAA,MACE,MAAM;AAAA,MACN,UAAU;AAAA,QACR,MAAM;AAAA,QACN,aACE,GAAG,qBACH;AAAA,QACF,YAAY;AAAA,UACV,MAAM;AAAA,UACN,YAAY;AAAA,YACV,OAAO,EAAE,MAAM,SAAS;AAAA,YACxB,SAAS,EAAE,MAAM,UAAU,aAAa,uBAAuB;AAAA,YAC/D,UAAU,EAAE,MAAM,UAAU,MAAM,CAAC,GAAG,cAAc,EAAE;AAAA,UACxD;AAAA,UACA,UAAU,CAAC,SAAS,SAAS;AAAA,QAC/B;AAAA,MACF;AAAA,IACF;AAAA,IACA;AAAA,MACE,MAAM;AAAA,MACN,UAAU;AAAA,QACR,MAAM;AAAA,QACN,aACE,GAAG,aACH;AAAA,QACF,YAAY;AAAA,UACV,MAAM;AAAA,UACN,YAAY;AAAA,YACV,OAAO,EAAE,MAAM,SAAS;AAAA,YACxB,QAAQ,EAAE,MAAM,UAAU,aAAa,0BAA0B;AAAA,UACnE;AAAA,UACA,UAAU,CAAC,SAAS,QAAQ;AAAA,QAC9B;AAAA,MACF;AAAA,IACF;AAAA,IACA;AAAA,MACE,MAAM;AAAA,MACN,UAAU;AAAA,QACR,MAAM;AAAA,QACN,aACE,GAAG,gBACH;AAAA,QACF,YAAY;AAAA,UACV,MAAM;AAAA,UACN,YAAY;AAAA,YACV,MAAM,EAAE,MAAM,UAAU,aAAa,uBAAuB;AAAA,YAC5D,OAAO,EAAE,MAAM,UAAU,aAAa,0BAA0B;AAAA,UAClE;AAAA,UACA,UAAU,CAAC,QAAQ,OAAO;AAAA,QAC5B;AAAA,MACF;AAAA,IACF;AAAA,IACA,GAAG;AAAA,EACL;AACF;;;ACjFO,SAAS,cAA8C,KAAuD;AACnH,QAAM,OAAO,IAAI,MAAM,KAAK;AAC5B,MAAI,CAAC,KAAM,OAAM,IAAI,MAAM,iCAAiC;AAC5D,MAAI,cAAc,IAAI,EAAG,OAAM,IAAI,MAAM,mBAAmB,IAAI,yDAAoD;AACpH,MAAI,OAAO,IAAI,YAAY,WAAY,OAAM,IAAI,MAAM,mBAAmB,IAAI,8BAA8B;AAC5G,SAAO;AACT;AAIO,SAAS,mBAAmB,KAA4C;AAC7E,SAAO,EAAE,MAAM,YAAY,UAAU,EAAE,MAAM,IAAI,MAAM,aAAa,IAAI,aAAa,YAAY,IAAI,WAAW,EAAE;AACpH;AAGO,SAAS,eACd,MACA,OAC+B;AAC/B,SAAO,OAAO,KAAK,CAAC,MAAM,EAAE,SAAS,IAAI;AAC3C;","names":[]}

package/dist/eval-campaign/index.d.ts

@@ -2,88 +2,9 @@ import { JudgeVerdict } from '@tangle-network/agent-eval';
 export { EnsembleAggregate, JudgeVerdict, RunRecord, aggregateJudgeVerdicts } from '@tangle-network/agent-eval';
 import { Scenario, JudgeConfig } from '@tangle-network/agent-eval/campaign';
 export { CampaignResult, DispatchContext, Gate, JudgeConfig, JudgeDimension, JudgeScore, LabeledScenarioStore, MutableSurface, Mutator, Scenario, SurfaceProposer, defaultProductionGate, evolutionaryProposer, gepaProposer, paretoSignificanceGate, runCampaign } from '@tangle-network/agent-eval/campaign';
+export { T as TrustItem, a as TrustThresholds, b as TrustVerdict, t as trustVerdicts } from '../trust-gate-Dcm5xSva.js';
 export { SelfImproveBudget, SelfImproveOptions, SelfImproveResult, selfImprove } from '@tangle-network/agent-eval/contract';
 
-/**
- * Trust gate — decides whether an ensemble's scores are allowed to be BELIEVED,
- * one level up from {@link aggregateJudgeVerdicts} (which only reduces ONE
- * artifact's raters to a composite). A composite is a number; this is the check
- * that the number means anything. It is the code "Enforced by" for the
- * measurement-validation skill's after-gate ("is this result allowed to be
- * believed").
- *
- * Three checks, each fail-loud and named in `trustReasons`:
- *   (1) inter-rater reliability over the corpus ≥ `irrFloor` — raters that
- *       disagree no better than chance carry no signal to optimize against.
- *   (2) per-item rater spread ≤ `spreadCeiling` — for EACH item, raters must
- *       converge on THAT item.
- *   (3) surviving raters per item ≥ `minSurvivors` — a mean over one or two
- *       raters is an anecdote, not an ensemble.
- *
- * CRITICAL metric semantics — per-item spread is rater disagreement about the
- * SAME item: `max(score) − min(score)` across the raters that scored THAT item
- * (max over its dimensions), never pooled across different items or across the
- * baseline/candidate sides. Pooling reads a genuine quality gap BETWEEN items as
- * "the raters split" and so trips the gate exactly when the finding is largest —
- * the failure mode the after-gate exists to prevent. The corpus IRR (check 1)
- * leans on the substrate's `interRaterReliability`, whose expected-disagreement
- * denominator already pools across items, so genuine item-to-item variation
- * RAISES reliability rather than lowering it.
- */
-
-/** One item's raters: the per-judge verdicts {@link aggregateJudgeVerdicts}
- *  reduces, tagged with the item they scored so spread stays within-item. */
-interface TrustItem<D extends string = string> {
-    /** Stable item identifier — surfaces in `perItemSpread` and `trustReasons`. */
-    itemId: string;
-    /** The raters' verdicts for THIS item (one per judge call). A failed judge
-     *  (`perDimension: null`) is dropped before spread/IRR, never folded as 0. */
-    verdicts: readonly JudgeVerdict<D>[];
-}
-/** Thresholds for {@link trustVerdicts}. All overridable; defaults are the
- *  conservative after-gate bar. */
-interface TrustThresholds {
-    /** Minimum corpus inter-rater reliability (Krippendorff-style α). Below this
-     *  the raters agree no better than chance. Default 0.2. */
-    irrFloor?: number;
-    /** Maximum per-item rater spread (`max − min` over a single item's surviving
-     *  raters, across its dimensions). Above this the raters split ON THAT ITEM.
-     *  Default 0.5. */
-    spreadCeiling?: number;
-    /** Minimum surviving (non-failed) raters required per item. Default 3. */
-    minSurvivors?: number;
-}
-/** Result of the trust gate. `trustworthy` iff every check passed; `trustReasons`
- *  is empty iff `trustworthy`. */
-interface TrustVerdict {
-    /** True iff IRR ≥ floor AND every item's spread ≤ ceiling AND every item has
-     *  ≥ `minSurvivors` surviving raters. */
-    trustworthy: boolean;
-    /** One entry per FAILED check, each naming its number + the offending value.
-     *  Empty iff `trustworthy`. */
-    trustReasons: string[];
-    /** Corpus inter-rater reliability actually measured (the check-1 value). */
-    interRaterReliability: number;
-    /** Per-item spread (`max − min` over surviving raters, max over dimensions),
-     *  keyed by `itemId`. The check-2 input, surfaced for drill-down. */
-    perItemSpread: Record<string, number>;
-}
-/**
- * Decide whether an ensemble's per-item verdicts are trustworthy enough to
- * believe a lift computed from them. Pure: no LLM, no I/O, no clock, no random —
- * the same `items` + `thresholds` always yield the same verdict.
- *
- * Sibling to {@link aggregateJudgeVerdicts}: that reduces ONE item's raters to a
- * composite; this audits the raters ACROSS items and reports whether the
- * composites are believable. Run it on the corpus of held-out items before
- * reporting any lift over their scores.
- *
- * @throws if `items` is empty — an empty corpus has no measurable trust, and a
- *   silent `trustworthy: true` over zero evidence is the exact lie the gate
- *   exists to refuse.
- */
-declare function trustVerdicts<D extends string>(items: readonly TrustItem<D>[], thresholds?: TrustThresholds): TrustVerdict;
-
 /**
  * Eval-campaign — the app-shell's curated surface for a product's
  * self-improvement loop, NOT a reimplementation.
@@ -152,4 +73,4 @@ interface EnsembleJudgeConfig<TArtifact, TScenario extends Scenario, D extends s
  */
 declare function buildEnsembleJudge<TArtifact, TScenario extends Scenario, D extends string>(cfg: EnsembleJudgeConfig<TArtifact, TScenario, D>): JudgeConfig<TArtifact, TScenario>;
 
-export { type EnsembleJudgeConfig, type TrustItem, type TrustThresholds, type TrustVerdict, buildEnsembleJudge, trustVerdicts };
+export { type EnsembleJudgeConfig, buildEnsembleJudge };

package/dist/index.d.ts

@@ -17,7 +17,7 @@ export { KeyCrypto, KeyProvisioner, PlanLimit, PlatformBalanceInfo, PlatformBala
 export { HttpHeadProbeConfig, PreflightProbe, PreflightProbeResult, PreflightProbeVerdict, PreflightReport, RouterChatProbeConfig, SandboxAuthProbeConfig, formatPreflightReport, httpHeadProbe, routerChatProbe, runPreflight, sandboxAuthProbe } from './preflight/index.js';
 export { ObjectBody, ObjectKeyParts, ObjectStore, PutObjectOptions, R2LikeBucket, R2LikeObjectBody, R2LikeObjectHead, SignObjectUrlArgs, VerifyObjectUrlResult, assertSafeKeySegment, createProxiedArtifactRoute, createR2ObjectStore, objectKey, signObjectUrl, verifyObjectUrl } from './object-store/index.js';
 export { B as BULK_DELETE_MAX_THREADS, C as ChatStoreInputError, t as threadTitleFromMessage } from './core-7qIM7svy.js';
-export { C as ChatAttachmentKind, a as ChatAttachmentPart, b as ChatFilePart, c as ChatImagePart, d as ChatInteractionPart, e as ChatMentionKind, f as ChatMentionPart, g as ChatMessagePart, h as ChatNoticePart, i as ChatPartTime, j as ChatPlanPart, k as ChatReasoningPart, l as ChatStepFinishPart, m as ChatStepStartPart, n as ChatSubtaskPart, o as ChatTextPart, p as ChatToolPart, q as ChatToolState, r as ChatToolStatus, s as ChatUsageTokens, D as DEFAULT_ATTACHMENT_PROMPT_HEADER, S as StorableHarnessPartKind, t as attachmentInputToPart, u as attachmentKindForMime, v as attachmentPartsFromMessageParts, w as buildAttachmentPromptBlock, x as historyContentWithAttachments, y as isChatAttachmentPart, z as isChatInteractionPart, A as isChatMentionPart, B as isChatPlanPart, E as isChatStepFinishPart, F as isChatTextPart, G as isChatToolPart, H as mentionInputToPart, I as mentionPartsFromMessageParts, J as toChatMessageParts } from './parts-Bg8qcDvB.js';
+export { C as ChatAttachmentKind, a as ChatAttachmentPart, b as ChatFilePart, c as ChatImagePart, d as ChatInteractionPart, e as ChatMentionKind, f as ChatMentionPart, g as ChatMessagePart, h as ChatNoticePart, i as ChatPartTime, j as ChatPlanPart, k as ChatReasoningPart, l as ChatStepFinishPart, m as ChatStepStartPart, n as ChatSubtaskPart, o as ChatTextPart, p as ChatToolPart, q as ChatToolState, r as ChatToolStatus, s as ChatUsageTokens, t as ChatWorkProductPart, D as DEFAULT_ATTACHMENT_PROMPT_HEADER, S as StorableHarnessPartKind, u as attachmentInputToPart, v as attachmentKindForMime, w as attachmentPartsFromMessageParts, x as buildAttachmentPromptBlock, y as historyContentWithAttachments, z as isChatAttachmentPart, A as isChatInteractionPart, B as isChatMentionPart, E as isChatPlanPart, F as isChatStepFinishPart, G as isChatTextPart, H as isChatToolPart, I as isChatWorkProductPart, J as mentionInputToPart, K as mentionPartsFromMessageParts, L as toChatMessageParts } from './parts-2ymE5cs-.js';
 export { DeriveKeyOptions, createFieldCrypto, decodeHexKey, decryptAesGcm, decryptBytes, decryptWithKey, deriveKey, encryptAesGcm, encryptBytes, encryptWithKey } from './crypto/index.js';
 export { J as JsonRecord, M as MISSING_TOOL_TERMINAL_ERROR, a as MISSING_TOOL_TERMINAL_REASON, S as StreamEvent, b as asRecord, c as asString, d as attachmentPartKey, e as collapseRedundantTextParts, f as draftAssistantParts, g as encodeEvent, h as finalizeAssistantParts, i as finalizePendingInteractionParts, j as getPartKey, m as mergePersistedPart, n as normalizePersistedPart, k as normalizeTime, l as normalizeToolEvent, r as resolveToolId, o as resolveToolName, t as terminalizeDanglingAssistantToolUpdates, p as terminalizeDanglingToolPart, q as terminalizeDanglingToolParts } from './stream-normalizer-QYUl4vnl.js';
 export { PersistedChatMessageForTurn, ResolvedChatTurn, buildUserTextParts, messageHasTurnId, normalizeClientTurnId, resolveChatTurn } from './stream/index.js';
@@ -29,6 +29,9 @@ export { ChatPlan, ChatPlanPersistedPart, ChatPlanStatus, PLAN_SUBMITTED_EVENT,
 export { CreateDurableInteractionRoutePersistenceOptions, DurableAnswerIntentJournal, DurableAnswerIntentRecord, DurableAnswerIntentState, DurableChatConflictError, DurableChatError, DurableChatErrorCode, DurableChatEventProjection, DurableChatGoneError, DurableChatScope, DurableChatStateStore, DurableChatUnavailableError, DurableFollowUpReceipt, DurableInteractionAcknowledgement, DurableInteractionGuarantee, DurableInteractionProjection, DurableInteractionProjectionAdapter, DurableInteractionSettlement, DurableInteractionSettlementFactoryOptions, DurableInteractionSettlementOptions, DurablePlanAuthority, DurablePlanAuthorityCurrentResult, DurablePlanAuthorityDecision, DurablePlanAuthorityResult, DurablePlanAuthorization, DurablePlanCommandJournal, DurablePlanCommandKey, DurablePlanCommandRecord, DurablePlanCommandState, DurablePlanDecision, DurablePlanEffectRecord, DurablePlanProjection, DurablePlanRouteAuthorizeArgs, DurablePlanRouteOptions, DurablePlanRoutes, DurablePlanStateStore, DurablePlanStore, InMemoryDurableChatStateStore, InMemoryDurableChatStore, PreparedDurableInteractionAnswer, applyDurableInteractionAnswer, applyDurableInteractionAsk, applyDurableInteractionCancel, createDurableChatEventProjection, createDurableChatScope, createDurableInteractionProjectionAdapter, createDurableInteractionRoutePersistence, createDurableInteractionSettlement, createDurablePlanRoutes, createInMemoryDurableChatStateStore, durableChatScopeKey, durableInteractionIntentKey, normalizePlanDecision, planAuthorityIdempotencyKey, planCommandKey, planEffectKey, recordDurableInteractionAnswer, recordDurableInteractionCancel, stablePlanReceipt, upsertDurableInteractionAsk } from './durable-chat/index.js';
 export { CompleteMissionInput, CreateMissionInput, DEFAULT_MISSION_STEP_KINDS, InMemoryMissionStore, MISSION_CONTROL_CHANNEL_ID, MissionApprovalsPort, MissionAuditEvent, MissionConcurrencyError, MissionCostLedger, MissionEngine, MissionEngineOptions, MissionEventSink, MissionGateKind, MissionGateOptions, MissionGateProposal, MissionOutcome, MissionPlanRunOptions, MissionProposalResolution, MissionRecord, MissionService, MissionServiceOptions, MissionState, MissionStatus, MissionStep, MissionStepState, MissionStepStatus, MissionStorePort, MissionStreamEvent, MissionStreamStatus, MissionStreamStep, MissionStreamStepStatus, MissionUpdateGuard, MissionUpdatePatch, ParseMissionBlocksOptions, ParsedMission, ParsedMissionStep, PlanOutcome, RetryableStepError, SandboxDispatch, SandboxDispatchDoneResult, SandboxDispatchInProgressResult, SandboxDispatchInput, SandboxDispatchResult, SetStepStatusPatch, StepGateClassification, StepOutcome, applyMissionEvent, asMissionStreamEvent, budgetGateProposalId, buildAgentMissionPlan, createInMemoryMissionStore, createMissionEngine, createMissionService, isMissionStopRequested, isMissionTerminal, mergeMissionState, noopEventSink, parseMissionBlocks, parseSessionStreamEnvelope, reduceMissionEvents, stepGateProposalId, volumeGateProposalId } from './missions/index.js';
 export { S as StepAgentActivity, W as WithAgentActivity, s as stepAgentActivity } from './agent-activity-C8ZG0F0M.js';
+export { A as AgentCheckInput, E as EvidenceEntry, a as EvidenceLocator, b as ExceptionEntry, c as ExceptionSeverity, P as ProfileBacktestSummary, Q as QualityCheck, W as WorkProductArtifact, d as WorkProductAuditEvent, e as WorkProductParseResult, f as WorkProductPatch, g as WorkProductPersistedPart, h as WorkProductProvenance, i as WorkProductRecord, j as WorkProductRef, k as WorkProductStatus, l as WorkProductStorePort, m as WorkProductUpdateGuard, n as WorkProductVersionEntry, o as isWorkProductStatus, p as parseAgentCheckInput, q as parseArtifactInput, r as parseEvidenceInput, s as parseExceptionInput, t as persistedPartToWorkProduct, u as unresolvedBlockingExceptions, w as workProductToPersistedPart } from './types-CEchbvgz.js';
+export { CreateWorkProductInput, EVIDENCE_COVERAGE_CHECK, FinalizeWorkProductProvenanceInput, InMemoryWorkProductStore, MAX_WORK_PRODUCT_BATCH, SubmitWorkProductInput, WorkProductAuthorizeArgs, WorkProductOutcome, WorkProductProvenanceBase, WorkProductRouteAuthorization, WorkProductRoutes, WorkProductRoutesOptions, WorkProductService, WorkProductServiceOptions, WorkProductToolConfig, WorkProductVerdictBody, WorkProductVerdictInput, buildWorkProductTools, canTransitionWorkProduct, createInMemoryWorkProductStore, createWorkProductRoutes, createWorkProductService, finalizeWorkProductProvenance, isWorkProductTerminal, stampProvenance, validateWorkProductVerdictBody, workProductTrustInputs } from './work-product/index.js';
+export { R as ReviewQueueInputs, a as ReviewQueueItem, b as ReviewQueuePendingAsk, c as ReviewQueueState, d as ReviewQueueThread, p as parseReviewQueueItem, e as projectReviewQueue } from './queue-C24V13h9.js';
 export { AppToolDescriptor, AuthenticatedSandboxUser, BuildAppToolMcpServersOptions, BuildSandboxToolFileMountsOptions, DEFAULT_SANDBOX_RESOURCES, DriveSandboxTurnOptions, ENV_TOTAL_MAX_BYTES, ENV_VALUE_MAX_BYTES, EnsureWorkspaceSandboxOptions, LivenessProbeConfig, MemberSyncSeam, Outcome, PROVISION_PAYLOAD_MAX_BYTES, PeekWorkspaceSandboxOutcome, ProfileComposeOptions, PromptInputPart, ProviderResolutionConfig, ProvisionPayloadSections, ProvisionProfileSection, ResolveSandboxClientCredentialsOptions, ResolvedModel, SandboxApiCredentials, SandboxBuildContext, SandboxClientCredentials, SandboxCredentialEnvironment, SandboxExecChannel, SandboxExecOptions, SandboxFileBytesOutcome, SandboxFileSizeOutcome, SandboxPermissionLevel, SandboxResourceConfig, SandboxRestoreSpec, SandboxRuntimeAuthRefreshError, SandboxRuntimeConfig, SandboxRuntimeConnection, SandboxScope, SandboxStepTransition, SandboxTerminalTokenOptions, SandboxTerminalTokenResult, SandboxTerminalTokenSubject, SandboxTerminalWsMatch, SandboxToolPathOptions, SandboxToolSpec, ScopedTokenResult, SecretStore, StoppedSandboxResumeFailure, StoppedSandboxResumeRecovery, StreamSandboxPromptOptions, TerminalProxyIdentity, WorkspaceSandboxConnectionArgs, WorkspaceSandboxConnectionHandlerOptions, WorkspaceSandboxEnsureContext, WorkspaceSandboxInstanceLike, WorkspaceSandboxManager, WorkspaceSandboxManagerOptions, WorkspaceSandboxRuntimeProxyArgs, WorkspaceSandboxRuntimeProxyHandlerOptions, WorkspaceSandboxTerminalUpgradeHandlerOptions, WriteProfileFilesOptions, assertEnvWithinLimits, assertProvisionPayloadWithinCap, attachReasoningEffort, bearerSubprotocolToken, bearerToken, buildAppToolMcpServers, buildSandboxRuntimeProxyHeaders, buildSandboxToolFileMounts, buildSandboxToolPathSetupScript, classifySeveredStream, createSandboxTerminalToken, createWorkspaceSandboxConnectionHandler, createWorkspaceSandboxManager, createWorkspaceSandboxRuntimeProxyHandler, createWorkspaceSandboxTerminalUpgradeHandler, deferredCorpusHash, deleteSecret, detectInteractiveQuestion, driveSandboxTurn, encodeSandboxRuntimePath, ensureWorkspaceSandbox, flattenHistory, getClient, isSandboxTerminalWsUpgrade, isTerminalPromptEvent, matchSandboxTerminalWsPath, mergeExtraMcp, mergeHistoryIntoParts, mintSandboxScopedToken, mintTerminalProxyToken, peekWorkspaceSandbox, readSandboxBinaryBytes, readSecret, resetClientCache, resolveModel, resolveSandboxClientCredentials, runSandboxPrompt, runSandboxToolPathSetup, sandboxToolBinDir, sandboxToolPath, sandboxToolRootDir, secretStoreFromClient, shellQuote, splitDeferredProfileFiles, statSandboxFileSize, storeSecret, streamSandboxPrompt, syncSandboxMemberAdd, syncSandboxMemberRemove, syncSandboxMemberRole, terminalTokenFromRequest, verifySandboxTerminalToken, verifyTerminalProxyToken, writeProfileFilesToBox } from './sandbox/index.js';
 export { CookieOptions, JsonObject, KvLike, RateLimitResult, RequestContext, SecurityHeaderOptions, addSecurityHeaders, assertMediaUrl, checkRateLimit, clearCookieHeader, extractRequestContext, parseJsonObjectBody, readCookieValue, requireString, serializeCookie } from './web/index.js';
 export { BuildRedactedDocumentOptions, DEFAULT_REDACTION_PATTERNS, RedactForIngestionOptions, RedactedDocSegment, RedactedDocument, RedactionPattern, RedactionSpan, RevealResult, RevealSpanOptions, buildRedactedDocument, detectSpans, maskSpans, redactForIngestion, revealSpan } from './redact/index.js';
@@ -41,6 +44,7 @@ export { CompletionRequirement, CompletionVerdict, CorrectnessChecker, ProducedS
 export { F as FlowSpan, a as FlowTrace } from './flow-types-CqomVAUN.js';
 export { InteractionData, InteractionOutcome, InteractionRequest, modelProvider } from '@tangle-network/agent-interface';
 export { StorageConfig } from '@tangle-network/sandbox';
+export { T as TrustItem } from './trust-gate-Dcm5xSva.js';
 import '@tangle-network/agent-runtime/intelligence';
 import '@tangle-network/agent-knowledge';
 import 'zod';

package/dist/index.js

@@ -43,6 +43,24 @@ import {
   loopTraceEventsToFlowSpans,
   stepActivityFlowTrace
 } from "./chunk-FBVLEGEG.js";
+import {
+  EVIDENCE_COVERAGE_CHECK,
+  MAX_WORK_PRODUCT_BATCH,
+  buildWorkProductTools,
+  canTransitionWorkProduct,
+  createInMemoryWorkProductStore,
+  createWorkProductRoutes,
+  createWorkProductService,
+  finalizeWorkProductProvenance,
+  isWorkProductTerminal,
+  stampProvenance,
+  validateWorkProductVerdictBody,
+  workProductTrustInputs
+} from "./chunk-HRH7ASAG.js";
+import {
+  parseReviewQueueItem,
+  projectReviewQueue
+} from "./chunk-UP33Z633.js";
 import {
   HubExecClient,
   invokeIntegrationHub,
@@ -197,10 +215,21 @@ import {
   isChatStepFinishPart,
   isChatTextPart,
   isChatToolPart,
+  isChatWorkProductPart,
   mentionInputToPart,
   mentionPartsFromMessageParts,
   toChatMessageParts
-} from "./chunk-X3N2H6JE.js";
+} from "./chunk-AFNTRJQ7.js";
+import {
+  isWorkProductStatus,
+  parseAgentCheckInput,
+  parseArtifactInput,
+  parseEvidenceInput,
+  parseExceptionInput,
+  persistedPartToWorkProduct,
+  unresolvedBlockingExceptions,
+  workProductToPersistedPart
+} from "./chunk-F2CBC4DY.js";
 import {
   MISSING_TOOL_TERMINAL_ERROR,
   MISSING_TOOL_TERMINAL_REASON,
@@ -344,7 +373,7 @@ import {
   restrictTaxonomy,
   verifyCapabilityToken,
   verifyExpiringCapabilityToken
-} from "./chunk-YKBDH2UY.js";
+} from "./chunk-WL7XHLDK.js";
 import {
   DEFAULT_APP_TOOL_PATHS,
   DEFAULT_HEADER_NAMES,
@@ -371,7 +400,7 @@ import {
   runToolLoop,
   streamToolLoop,
   toLoopEvents
-} from "./chunk-JWBZ74TW.js";
+} from "./chunk-7ESQUSAC.js";
 import {
   DEFAULT_TANGLE_BILLING_ENFORCEMENT_ENV_VAR,
   DEFAULT_TANGLE_ROUTER_BASE_URL,
@@ -387,18 +416,20 @@ import {
   tangleExecutionKeyHttpError,
   trimOrNull
 } from "./chunk-JML7WKWU.js";
+import {
+  createAppToolRuntimeExecutor,
+  dispatchAppTool,
+  outcomeStatus
+} from "./chunk-UOAYS72M.js";
 import {
   APP_TOOL_NAMES,
   ToolInputError,
   buildAppToolOpenAITools,
-  createAppToolRuntimeExecutor,
   customToolToOpenAI,
   defineAppTool,
-  dispatchAppTool,
   findCustomTool,
-  isAppToolName,
-  outcomeStatus
-} from "./chunk-7CTIUCQ4.js";
+  isAppToolName
+} from "./chunk-YEFFHORB.js";
 import {
   __resetCatalogCache,
   buildCatalog,
@@ -440,6 +471,7 @@ export {
   DurableChatUnavailableError,
   ENV_TOTAL_MAX_BYTES,
   ENV_VALUE_MAX_BYTES,
+  EVIDENCE_COVERAGE_CHECK,
   EmailContentSchema,
   HubExecClient,
   INTERACTION_CANCEL_EVENT,
@@ -449,6 +481,7 @@ export {
   InMemoryDurableChatStateStore,
   InMemoryDurableChatStore,
   KNOWN_HARNESSES,
+  MAX_WORK_PRODUCT_BATCH,
   MCP_PROTOCOL_VERSIONS,
   MISSING_TOOL_TERMINAL_ERROR,
   MISSING_TOOL_TERMINAL_REASON,
@@ -505,8 +538,10 @@ export {
   buildSandboxToolPathSetupScript,
   buildScopedMcpServerEntry,
   buildUserTextParts,
+  buildWorkProductTools,
   canTransitionInteractionStatus,
   canTransitionPlanStatus,
+  canTransitionWorkProduct,
   cancelStatusFor,
   checkRateLimit,
   checkThemeContract,
@@ -538,6 +573,7 @@ export {
   createFieldCrypto,
   createInMemoryDurableChatStateStore,
   createInMemoryMissionStore,
+  createInMemoryWorkProductStore,
   createInteractionAnswerRoute,
   createKnowledgeLoop,
   createLlmCorrectnessChecker,
@@ -561,6 +597,8 @@ export {
   createTangleRouterModelConfig,
   createTcloudKeyProvisioner,
   createTokenRecallChecker,
+  createWorkProductRoutes,
+  createWorkProductService,
   createWorkspaceKeyManager,
   createWorkspaceSandboxConnectionHandler,
   createWorkspaceSandboxManager,
@@ -600,6 +638,7 @@ export {
   fieldAcceptsFreeText,
   finalizeAssistantParts,
   finalizePendingInteractionParts,
+  finalizeWorkProductProvenance,
   findCustomTool,
   flattenHistory,
   formatPreflightReport,
@@ -620,6 +659,7 @@ export {
   isChatStepFinishPart,
   isChatTextPart,
   isChatToolPart,
+  isChatWorkProductPart,
   isHarness,
   isMissionStopRequested,
   isMissionTerminal,
@@ -631,6 +671,8 @@ export {
   isTangleExecutionKeyError,
   isTerminalInteractionStatus,
   isTerminalPromptEvent,
+  isWorkProductStatus,
+  isWorkProductTerminal,
   lightTheme,
   listSessionInteractions,
   loopTraceEventsToFlowSpans,
@@ -659,17 +701,23 @@ export {
   noticePartKey,
   objectKey,
   outcomeStatus,
+  parseAgentCheckInput,
+  parseArtifactInput,
   parseAssetSpec,
+  parseEvidenceInput,
+  parseExceptionInput,
   parseInteractionAnswers,
   parseInteractionCancel,
   parseInteractionRequest,
   parseJsonObjectBody,
   parseMissionBlocks,
   parsePlanSubmittedEvent,
+  parseReviewQueueItem,
   parseSessionStreamEnvelope,
   peekWorkspaceSandbox,
   persistedPartToInteraction,
   persistedPartToPlan,
+  persistedPartToWorkProduct,
   planAuthorityIdempotencyKey,
   planCommandKey,
   planEffectKey,
@@ -678,6 +726,7 @@ export {
   planRevisionKey,
   planToPersistedPart,
   producedFromToolEvents,
+  projectReviewQueue,
   pumpBufferedTurn,
   questionInteractionContentSignature,
   readCookieValue,
@@ -729,6 +778,7 @@ export {
   splitDeferredProfileFiles,
   stablePlanReceipt,
   stampInteractionAnswers,
+  stampProvenance,
   statSandboxFileSize,
   stepActivityFlowTrace,
   stepAgentActivity,
@@ -753,8 +803,10 @@ export {
   toLoopEvents,
   traceEnv,
   trimOrNull,
+  unresolvedBlockingExceptions,
   upsertDurableInteractionAsk,
   validateInteractionAnswerBody,
+  validateWorkProductVerdictBody,
   verifyCapabilityToken,
   verifyCompletion,
   verifyExpiringCapabilityToken,
@@ -763,6 +815,8 @@ export {
   verifyTerminalProxyToken,
   volumeGateProposalId,
   weightedComposite,
+  workProductToPersistedPart,
+  workProductTrustInputs,
   writeProfileFilesToBox
 };
 //# sourceMappingURL=index.js.map

package/dist/model-resolution/index.d.ts

@@ -1,3 +1,102 @@
+/**
+ * Model failover — the answer to "one upstream hit a quota wall and the customer
+ * got a Bad Gateway even though the router had abundant healthy capacity".
+ *
+ * This is deliberately NOT a health probe. Probing before every turn buys a
+ * round-trip of latency on the happy path and still races the outage (a model
+ * healthy at probe time can 502 a second later). Failover is REACTIVE: run the
+ * preferred model, and on an upstream-outage signal move to the next model in
+ * the chain. Zero added latency when the preferred model works.
+ *
+ * Two facts drive the design, both measured against a live box during the
+ * 2026-07-25 Anthropic/DeepSeek outage:
+ *
+ * 1. An outage is NOT always a thrown error. The sandbox resolves with a
+ *    payload — `{ success: false, errorCode: 'provider_inference_unavailable' }`
+ *    — so a classifier that only inspects `catch` misses the customer-visible
+ *    case entirely. `isUpstreamUnavailable` inspects resolved values too.
+ * 2. Catalog membership is NOT liveness. The router's `/v1/models` still listed
+ *    every dead model during the outage, so `validateChatModelId` admitted
+ *    `claude-sonnet-4-6` while every call to it returned 502. Nothing upstream
+ *    of the actual call can be trusted to tell you a model is serving.
+ *
+ * Substrate-free: the caller injects `run`, so this composes with a sandbox
+ * turn, a router chat completion, or a test fake without importing any of them.
+ */
+/**
+ * Error codes that mean "this model's upstream is unavailable — a different
+ * model may still work". Deliberately excludes codes that would fail identically
+ * on every model (bad request, auth, content filter): retrying those down a
+ * chain burns latency and money to reach the same failure.
+ */
+declare const UPSTREAM_UNAVAILABLE_CODES: readonly string[];
+/** HTTP statuses that indicate an upstream capacity/availability problem. */
+declare const UPSTREAM_UNAVAILABLE_STATUSES: readonly number[];
+/**
+ * True when `signal` — a thrown error OR a resolved result payload — indicates
+ * the model's upstream is unavailable and another model is worth trying.
+ *
+ * Checked in order of decreasing confidence: explicit code, HTTP status, then
+ * message text. A resolved payload only counts as a failure when it carries an
+ * explicit failure marker (`success: false`, or an `error`/`errorCode` field) —
+ * a successful result is never misread as an outage.
+ */
+declare function isUpstreamUnavailable(signal: unknown): boolean;
+/** One model tried, and how it went. */
+interface ModelFailoverAttempt {
+    model: string;
+    ok: boolean;
+    /** Why this model was abandoned. Absent when `ok`. */
+    reason?: string;
+}
+/** The outcome of a failover run: the value plus the full attempt trail. */
+interface ModelFailoverResult<T> {
+    value: T;
+    /** The model that actually produced `value`. */
+    model: string;
+    attempts: ModelFailoverAttempt[];
+    /** True when the preferred (first) model did not serve the request. */
+    usedFallback: boolean;
+}
+/** Every model in the chain failed; carries the trail for logging. */
+declare class ModelFailoverExhaustedError extends Error {
+    readonly attempts: ModelFailoverAttempt[];
+    constructor(attempts: ModelFailoverAttempt[]);
+}
+/** Inputs to {@link runWithModelFailover}. */
+interface RunWithModelFailoverInput<T> {
+    /** Preferred model first, then fallbacks in descending preference. */
+    models: readonly string[];
+    /** Executes one turn with the given model. */
+    run: (model: string) => Promise<T>;
+    /**
+     * Classifies a resolved result as an upstream outage. Defaults to
+     * {@link isUpstreamUnavailable}, which understands the sandbox's
+     * `{ success: false, errorCode }` payload.
+     */
+    isUnavailableResult?: (result: T) => boolean;
+    /** Classifies a thrown error. Defaults to {@link isUpstreamUnavailable}. */
+    isUnavailableError?: (error: unknown) => boolean;
+    /** Observability hook fired each time a model is abandoned. */
+    onFallback?: (attempt: ModelFailoverAttempt, nextModel: string) => void;
+}
+/**
+ * Run `run` against the first model in `models` that does not report an upstream
+ * outage, falling through the chain in order.
+ *
+ * A non-outage failure (bad request, auth, content filter) is re-thrown
+ * immediately rather than retried down the chain — those fail identically on
+ * every model, so walking the chain would only multiply latency and spend.
+ *
+ * @throws ModelFailoverExhaustedError when every model reports an outage.
+ */
+declare function runWithModelFailover<T>(input: RunWithModelFailoverInput<T>): Promise<ModelFailoverResult<T>>;
+/**
+ * Build a failover chain: the preferred model first, then `fallbacks`, with
+ * duplicates removed so a model is never retried twice in one turn.
+ */
+declare function buildModelChain(preferred: string, fallbacks: readonly string[]): string[];
+
 /**
  * Canonical chat-model resolution — identical across every agent app.
  *
@@ -85,4 +184,4 @@ declare function isWellFormedModelId(modelId: string): boolean;
 /** Resolve unique catalog IDs associated with a given model including its canonical form if applicable */
 declare function catalogIdsForModel(model: ModelInfo): string[];
 
-export { type ChatModelSource, type ChatModelValidationFailure, type ChatModelValidationResult, type ChatModelValidationSuccess, type LoadModels, type ModelInfo, type ResolveChatModelInput, type ResolvedChatModel, type ValidateChatModelIdInput, catalogIdsForModel, cleanModelId, isWellFormedModelId, resolveChatModel, validateChatModelId };
+export { type ChatModelSource, type ChatModelValidationFailure, type ChatModelValidationResult, type ChatModelValidationSuccess, type LoadModels, type ModelFailoverAttempt, ModelFailoverExhaustedError, type ModelFailoverResult, type ModelInfo, type ResolveChatModelInput, type ResolvedChatModel, type RunWithModelFailoverInput, UPSTREAM_UNAVAILABLE_CODES, UPSTREAM_UNAVAILABLE_STATUSES, type ValidateChatModelIdInput, buildModelChain, catalogIdsForModel, cleanModelId, isUpstreamUnavailable, isWellFormedModelId, resolveChatModel, runWithModelFailover, validateChatModelId };