@axiom-lattice/core 3.0.7 → 3.1.0

package/dist/index.d.mts

@@ -5494,12 +5494,13 @@ interface EvalRunService {
      *   executes in. Comes from the CALLER's runConfig, NOT from the eval
      *   project's targetServerConfig. Test cases are environment-agnostic;
      *   the run environment is decided at run time.
+     * @param taskId - Optional training task (round) this run belongs to.
      * @returns The newly created run ID
      */
     startRun(tenantId: string, projectId: string, suiteIds?: string[], caseIds?: string[], runConfig?: {
         workspaceId?: string;
         projectId?: string;
-    }): Promise<string>;
+    }, taskId?: string): Promise<string>;
     /**
      * Abort a running evaluation.
      *

package/dist/index.d.ts

@@ -5494,12 +5494,13 @@ interface EvalRunService {
      *   executes in. Comes from the CALLER's runConfig, NOT from the eval
      *   project's targetServerConfig. Test cases are environment-agnostic;
      *   the run environment is decided at run time.
+     * @param taskId - Optional training task (round) this run belongs to.
      * @returns The newly created run ID
      */
     startRun(tenantId: string, projectId: string, suiteIds?: string[], caseIds?: string[], runConfig?: {
         workspaceId?: string;
         projectId?: string;
-    }): Promise<string>;
+    }, taskId?: string): Promise<string>;
     /**
      * Abort a running evaluation.
      *

package/dist/index.js

@@ -4501,6 +4501,7 @@ var InMemoryTaskStore = class {
       projectId: params.projectId,
       dueDate: params.dueDate,
       metadata: params.metadata,
+      files: params.files,
       parentId: params.parentId,
       sourceId: params.sourceId,
       context: params.context,
@@ -4558,6 +4559,7 @@ var InMemoryTaskStore = class {
     const updated = {
       ...existing,
       ...updates,
+      files: updates.files !== void 0 ? updates.files : existing.files,
       updatedAt: /* @__PURE__ */ new Date()
     };
     tenantTasks.set(id, updated);
@@ -16043,6 +16045,11 @@ var manageTaskSchema = import_zod43.z.object({
   dependencies: import_zod43.z.array(import_zod43.z.string()).optional().describe("List of task IDs that must be completed before this task can start"),
   result: import_zod43.z.string().optional().describe("Result summary when task is completed"),
   failureReason: import_zod43.z.string().optional().describe("Reason for failure (use when status='failed')"),
+  files: import_zod43.z.array(import_zod43.z.object({
+    uri: import_zod43.z.string().describe("Uniquely locates the resource: http(s):// URL, /s/:token share, or sandbox path"),
+    name: import_zod43.z.string().optional().describe("Display name"),
+    addedBy: import_zod43.z.enum(["user", "agent"]).optional().describe("Who attached the file")
+  })).optional().describe("File references attached to this task"),
   summary: import_zod43.z.string().optional().describe("Brief summary of the operation")
 });
 function buildReviewMarkdown(task) {
@@ -16101,7 +16108,8 @@ function createTaskMiddleware() {
           requireReview: input.requireReview,
           dependencies: input.dependencies,
           workspaceId,
-          projectId
+          projectId,
+          files: input.files
         });
         return JSON.stringify({ success: true, data: task });
       }
@@ -16180,7 +16188,8 @@ function createTaskMiddleware() {
           "result",
           "failureReason",
           "requireReview",
-          "dependencies"
+          "dependencies",
+          "files"
         ];
         for (const field of settableFields) {
           if (input[field] !== void 0) {
@@ -20012,11 +20021,11 @@ ${BASE_PROMPT}` : BASE_PROMPT;
       defaultInterruptOn: interruptOn,
       subagents,
       generalPurposeAgent: true
-    }),
-    // Enables Anthropic prompt caching for improved performance and reduced costs
-    (0, import_langchain54.anthropicPromptCachingMiddleware)({
-      unsupportedModelBehavior: "ignore"
     })
+    // Enables Anthropic prompt caching for improved performance and reduced costs
+    // anthropicPromptCachingMiddleware({
+    //   unsupportedModelBehavior: "ignore",
+    // })
   ];
   if (interruptOn) {
     middleware.push((0, import_langchain54.humanInTheLoopMiddleware)({ interruptOn }));
@@ -23704,6 +23713,27 @@ registerToolLattice(
     }
   }
 );
+registerToolLattice(
+  "list_models",
+  {
+    name: "list_models",
+    description: "List all registered models. Returns each model's key (use this string as modelKey when creating eval projects via manage_eval create_project, or as an agent's modelKey in create_agent/update_agent) and its display name.",
+    schema: import_zod47.default.object({})
+  },
+  async (_input) => {
+    try {
+      const lattices = modelLatticeManager.getAllLattices();
+      return JSON.stringify({
+        models: lattices.map((l) => ({
+          key: l.key,
+          name: l.client.name ?? l.key
+        }))
+      });
+    } catch (error) {
+      return JSON.stringify({ error: `Failed to list models: ${error.message}` });
+    }
+  }
+);
 registerToolLattice(
   "invoke_agent",
   {
@@ -23835,6 +23865,22 @@ The skills document WHY and HOW; these gates are the unskippable
 minimum. If you cannot satisfy a gate (e.g. user says skip), record it
 and proceed only on the user's explicit instruction.
 
+LEARNING ROUND KICKOFF \u2014 a message that names an existing target agent
+id AND an existing tracking task id (a "learning round"). This protocol
+OVERRIDES the defaults above:
+- The target agent ALREADY EXISTS (an empty placeholder). Build and
+  refine it via update_agent on that exact id. NEVER call create_agent \u2014
+  a new agent would disconnect the round's tracking.
+- The parent task ALREADY EXISTS \u2014 your create-a-task-first duty is
+  satisfied by it. Create subtasks with manage_task under its id
+  (parentId); NEVER create a new parent task for the round.
+- The round is pre-approved \u2014 skip the DESIGN\u2192CONFIRM gates: show the
+  design in your reply, then build directly.
+- Every update_agent call must be mirrored by a manage_task work item
+  whose summary names the changed keys (e.g. "update_agent: prompt,
+  modelKey") \u2014 the round feed highlights these so the user can see what
+  changed between iterations.
+
 Your sub-skills (accessible via the MOC or direct loading):
 - [[learn-capability]] \u2014 Learn from any source material \u2192 skills + agents
 - [[agent-build]] \u2014 Design and build single agents (REACT/DEEP_AGENT)
@@ -23949,6 +23995,7 @@ var agentArchitectConfig = {
   tools: [
     "list_agents",
     "list_tools",
+    "list_models",
     "list_middleware_types",
     "list_connections",
     "get_agent",
@@ -30832,6 +30879,7 @@ function createManageEvalTool() {
     description: import_zod64.z.string().optional(),
     judgeModelKey: import_zod64.z.string().optional(),
     concurrency: import_zod64.z.number().optional(),
+    targetAgentId: import_zod64.z.string().optional().describe("Optional for create_project \u2014 the agent this eval project verifies. Recorded in targetServerConfig so the agent's detail page can find its eval data without relying on project naming."),
     suiteId: import_zod64.z.string().optional(),
     caseId: import_zod64.z.string().optional(),
     inputMessage: import_zod64.z.string().optional(),
@@ -30863,7 +30911,11 @@ function createManageEvalTool() {
               judgeModelConfig: { modelKey: input.judgeModelKey },
               targetServerConfig: {
                 workspace_id: ctx.workspaceId,
-                project_id: ctx.projectId
+                project_id: ctx.projectId,
+                // The project↔agent association lives here (not in the project
+                // name): an eval project is agent-agnostic by design, this
+                // pointer is merely the agent's designated verifier.
+                ...input.targetAgentId ? { targetAgentId: input.targetAgentId } : {}
               },
               concurrency: input.concurrency ?? 3
             });
@@ -30930,7 +30982,7 @@ function createManageEvalTool() {
       name: "manage_eval",
       description: `Create, update, delete evaluation projects, suites, and test cases.
  
-Project: create_project(name, description?, judgeModelKey?, concurrency?) | update_project | delete_project
+Project: create_project(name, description?, judgeModelKey?, concurrency?, targetAgentId?) | update_project | delete_project
   judgeModelKey defaults to first available model. concurrency defaults to 3.
   delete_project rejected if active runs exist.
   **When creating a project from within a workspace, the workspace/project context is
@@ -30954,6 +31006,7 @@ function createRunEvalTool() {
     suiteIds: import_zod64.z.array(import_zod64.z.string()).optional().describe("Optional for start \u2014 only run these suites (e.g. dev set only). Omit to run all."),
     caseIds: import_zod64.z.array(import_zod64.z.string()).optional().describe("Optional for start \u2014 only run these cases across the selected suites. Omit to run all cases in those suites."),
     runId: import_zod64.z.string().optional().describe("Required for status, resume, abort"),
+    taskId: import_zod64.z.string().optional().describe("Optional for start \u2014 training task ID this run belongs to (round association)"),
     sleepMs: import_zod64.z.number().int().min(0).max(12e4).optional().describe("Optional for status \u2014 sleep this many ms BEFORE checking the run, to pace polling (e.g. 15000 \u2192 30000 \u2192 60000 \u2192 120000). Omit to check immediately."),
     wait: import_zod64.z.boolean().optional().describe("Optional for start \u2014 defaults to true: block synchronously (up to ~150s) and return final results in one call. Set false to return the runId immediately and poll.")
   });
@@ -30971,7 +31024,7 @@ function createRunEvalTool() {
           switch (input.action) {
             case "start": {
               const ctx = workspaceContext(exeConfig);
-              const runId = await svc.startRun(tid, input.projectId, input.suiteIds, input.caseIds, ctx);
+              const runId = await svc.startRun(tid, input.projectId, input.suiteIds, input.caseIds, ctx, input.taskId);
               if (input.wait === false) {
                 data = sanitize({ runId, message: "Run started. Poll with run_eval status (backoff: 15s\u219230s\u219260s\u2192max 120s)." });
                 break;