@nodetool-ai/agents 0.7.0-rc.23 → 0.8.2

package/dist/evals/surfaces/memory.d.ts

@@ -0,0 +1,34 @@
+/**
+ * Headless bridge for the memory tool-loop eval.
+ *
+ * Unlike the editor-surface bridges (which reimplement browser `ui_*` effects),
+ * this bridge drives the REAL backend tools — `memory_save/list` and
+ * `asset_search` — against an in-memory SQLite DB, plus a stub `generate_image`
+ * that persists a real asset the way a generation node would. So the eval
+ * exercises the actual DB writes, resource validation, and asset resolution a
+ * chat turn would, scoring a model's ability to run the creative loop:
+ * generate media → remember it (referencing the asset) → recall it later.
+ */
+import { type MemoryResource } from "@nodetool-ai/models";
+import type { ProcessingContext } from "@nodetool-ai/runtime";
+import type { HeadlessSurfaceBridge, ToolLoopEvalCase } from "../tool-loop-eval.js";
+/** Snapshot the case predicates run against. */
+export interface MemoryBridgeFinalState {
+    memories: Array<{
+        kind: string;
+        content: string;
+        resources: MemoryResource[];
+    }>;
+    assets: Array<{
+        id: string;
+        name: string;
+        content_type: string;
+    }>;
+}
+/**
+ * Build a fresh headless bridge over a clean in-memory DB. `seed` can pre-load
+ * an asset + memory (used by the recall case).
+ */
+export declare function createMemoryToolBridge(seed?: (ctx: ProcessingContext) => Promise<void>): HeadlessSurfaceBridge<MemoryBridgeFinalState>;
+export declare const MEMORY_TOOL_LOOP_CASES: ToolLoopEvalCase<MemoryBridgeFinalState>[];
+//# sourceMappingURL=memory.d.ts.map

package/dist/evals/surfaces/memory.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"memory.d.ts","sourceRoot":"","sources":["../../../src/evals/surfaces/memory.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;GAUG;AAGH,OAAO,EAIL,KAAK,cAAc,EACpB,MAAM,qBAAqB,CAAC;AAC7B,OAAO,KAAK,EAAE,iBAAiB,EAAE,MAAM,sBAAsB,CAAC;AAG9D,OAAO,KAAK,EACV,qBAAqB,EACrB,gBAAgB,EAEjB,MAAM,sBAAsB,CAAC;AAM9B,gDAAgD;AAChD,MAAM,WAAW,sBAAsB;IACrC,QAAQ,EAAE,KAAK,CAAC;QACd,IAAI,EAAE,MAAM,CAAC;QACb,OAAO,EAAE,MAAM,CAAC;QAChB,SAAS,EAAE,cAAc,EAAE,CAAC;KAC7B,CAAC,CAAC;IACH,MAAM,EAAE,KAAK,CAAC;QAAE,EAAE,EAAE,MAAM,CAAC;QAAC,IAAI,EAAE,MAAM,CAAC;QAAC,YAAY,EAAE,MAAM,CAAA;KAAE,CAAC,CAAC;CACnE;AAmBD;;;GAGG;AACH,wBAAgB,sBAAsB,CACpC,IAAI,CAAC,EAAE,CAAC,GAAG,EAAE,iBAAiB,KAAK,OAAO,CAAC,IAAI,CAAC,GAC/C,qBAAqB,CAAC,sBAAsB,CAAC,CAwI/C;AA6CD,eAAO,MAAM,sBAAsB,EAAE,gBAAgB,CAAC,sBAAsB,CAAC,EA6E1E,CAAC"}

package/dist/evals/surfaces/memory.js

@@ -0,0 +1,264 @@
+/**
+ * Headless bridge for the memory tool-loop eval.
+ *
+ * Unlike the editor-surface bridges (which reimplement browser `ui_*` effects),
+ * this bridge drives the REAL backend tools — `memory_save/list` and
+ * `asset_search` — against an in-memory SQLite DB, plus a stub `generate_image`
+ * that persists a real asset the way a generation node would. So the eval
+ * exercises the actual DB writes, resource validation, and asset resolution a
+ * chat turn would, scoring a model's ability to run the creative loop:
+ * generate media → remember it (referencing the asset) → recall it later.
+ */
+import { z } from "zod";
+import { Asset, Memory, initTestDb } from "@nodetool-ai/models";
+import { toolForCapabilityName } from "../../capabilities/lazy-tool.js";
+import { isString } from "../../utils/type-guards.js";
+const EVAL_USER = "eval-user";
+const EVAL_THREAD = "eval-thread";
+const resourceParam = z.object({
+    type: z.string(),
+    id: z.string(),
+    uri: z.string().optional(),
+    label: z.string().optional()
+});
+function slug(text) {
+    return (text
+        .toLowerCase()
+        .replace(/[^a-z0-9]+/g, "-")
+        .replace(/^-+|-+$/g, "")
+        .slice(0, 40) || "image");
+}
+/**
+ * Build a fresh headless bridge over a clean in-memory DB. `seed` can pre-load
+ * an asset + memory (used by the recall case).
+ */
+export function createMemoryToolBridge(seed) {
+    initTestDb();
+    const partialCtx = {
+        userId: EVAL_USER,
+        threadId: EVAL_THREAD
+    };
+    // SAFETY: the memory tools read only the user and thread ids off the
+    // context; nothing in this bridge reaches for another member.
+    const ctx = partialCtx;
+    const saveTool = toolForCapabilityName("memory_save");
+    const listTool = toolForCapabilityName("memory_list");
+    const memorySearchTool = toolForCapabilityName("memory_search");
+    const searchTool = toolForCapabilityName("asset_search");
+    let state = { memories: [], assets: [] };
+    const refresh = async () => {
+        // User-scoped, like the store: a case that seeds another thread must see
+        // its memory in the final state.
+        const rows = await Memory.list(EVAL_USER, { limit: 200 });
+        const [assetRows] = await Asset.searchAssetsGlobal(EVAL_USER, "", {
+            limit: 200
+        });
+        state = {
+            memories: rows.map((r) => ({
+                kind: r.kind,
+                content: r.content,
+                resources: Array.isArray(r.resources) ? r.resources : []
+            })),
+            assets: assetRows
+                .filter((a) => a.content_type !== "folder")
+                .map((a) => ({
+                id: a.id,
+                name: a.name,
+                content_type: a.content_type
+            }))
+        };
+    };
+    // Bridges are created synchronously but seeding is async. Every tool awaits
+    // this readiness promise first, so the seed (and initial snapshot) always
+    // complete before any tool runs — no race with the first tool call.
+    const ready = (async () => {
+        if (seed)
+            await seed(ctx);
+        await refresh();
+    })();
+    const tools = [
+        {
+            name: "generate_image",
+            description: "Generate an image from a text prompt and save it as an asset. Returns " +
+                "the new asset_id and its asset:// uri.",
+            parameters: z.object({
+                prompt: z.string().describe("What to generate.")
+            }),
+            execute: async (args) => {
+                await ready;
+                const prompt = isString(args.prompt) ? args.prompt : "image";
+                const asset = await Asset.create({
+                    user_id: EVAL_USER,
+                    name: `${slug(prompt)}.png`,
+                    content_type: "image/png"
+                });
+                await refresh();
+                return {
+                    success: true,
+                    asset_id: asset.id,
+                    uri: `asset://${asset.id}.png`,
+                    content_type: "image/png"
+                };
+            }
+        },
+        {
+            name: saveTool.name,
+            description: saveTool.description,
+            parameters: z.object({
+                content: z.string(),
+                title: z.string().optional(),
+                kind: z
+                    .enum(["note", "fact", "preference", "decision", "resource"])
+                    .optional(),
+                resources: z.array(resourceParam).optional()
+            }),
+            execute: async (args) => {
+                await ready;
+                const result = await saveTool.process(ctx, args);
+                await refresh();
+                return result;
+            }
+        },
+        {
+            name: listTool.name,
+            description: listTool.description,
+            parameters: z.object({ limit: z.number().optional() }),
+            execute: async (args) => {
+                await ready;
+                const result = await listTool.process(ctx, args);
+                await refresh();
+                return result;
+            }
+        },
+        {
+            name: memorySearchTool.name,
+            description: memorySearchTool.description,
+            parameters: z.object({
+                query: z.string(),
+                thread: z.enum(["all", "current"]).optional(),
+                limit: z.number().optional()
+            }),
+            execute: async (args) => {
+                await ready;
+                const result = await memorySearchTool.process(ctx, args);
+                await refresh();
+                return result;
+            }
+        },
+        {
+            name: searchTool.name,
+            description: searchTool.description,
+            parameters: z.object({
+                query: z.string().optional(),
+                content_type: z.string().optional(),
+                limit: z.number().optional()
+            }),
+            execute: async (args) => {
+                await ready;
+                const result = await searchTool.process(ctx, args);
+                await refresh();
+                return result;
+            }
+        }
+    ];
+    return { tools, finalState: () => state };
+}
+// --- predicates --------------------------------------------------------------
+const memoryReferencesAnAsset = {
+    name: "a saved memory references a generated asset",
+    test: (s) => s.memories.some((m) => m.resources.some((r) => r.type === "asset" &&
+        isString(r.id) &&
+        s.assets.some((a) => a.id === r.id))),
+    detail: "Expected at least one memory whose resources include an asset " +
+        "ref pointing at a generated image."
+};
+const atLeastOneMemory = {
+    name: "at least one memory persisted",
+    test: (s) => s.memories.length >= 1
+};
+/** The seeded palette memory survives the run — recall must not rewrite it. */
+const paletteMemoryIntact = {
+    name: "the memory saved in the other conversation is still there",
+    test: (s) => s.memories.some((m) => /viridian/i.test(m.content)),
+    detail: "Expected the seeded cross-thread memory to be found and left in place."
+};
+const MEMORY_SYSTEM_PROMPT = `You are a creative assistant with durable memory that spans every conversation.
+
+You work on media projects over many turns. Use these tools:
+- generate_image({ prompt }) — create an image; returns { asset_id, uri }.
+- memory_save({ content, kind?, resources? }) — remember project facts and the media you make. Reference assets by passing resources like [{ "type": "asset", "id": "<asset_id>" }] so you can reuse them later.
+- memory_list() — recall everything you have saved, in any conversation.
+- memory_search({ query }) — find memories whose title or content contain every word of the query, across every conversation.
+- asset_search({ query?, content_type? }) — find assets already created.
+
+Call one tool at a time and use each result before the next call. When the objective is fully satisfied, STOP calling tools and give a one-line summary.`;
+export const MEMORY_TOOL_LOOP_CASES = [
+    {
+        id: "generate-and-remember",
+        description: "Generate an image, then save a memory that references the generated asset for reuse.",
+        objective: "Generate an image of a red fox mascot logo, then save it as the approved project logo, referencing the generated image so you can reuse it later.",
+        systemPrompt: MEMORY_SYSTEM_PROMPT,
+        createBridge: () => createMemoryToolBridge(),
+        expect: {
+            requiredTools: ["generate_image", "memory_save"],
+            ordering: [["generate_image", "memory_save"]],
+            finalState: [memoryReferencesAnAsset],
+            minToolCalls: 2,
+            maxToolCalls: 12
+        }
+    },
+    {
+        id: "recall-existing",
+        description: "Recall memories saved earlier in the conversation via memory_list.",
+        objective: "We've worked on this project before. Look at what you've already saved for this conversation and tell me which image was approved as the logo.",
+        systemPrompt: MEMORY_SYSTEM_PROMPT,
+        createBridge: () => createMemoryToolBridge(async () => {
+            const asset = await Asset.create({
+                user_id: EVAL_USER,
+                name: "approved-logo.png",
+                content_type: "image/png"
+            });
+            await Memory.create({
+                user_id: EVAL_USER,
+                thread_id: EVAL_THREAD,
+                kind: "resource",
+                content: "The approved project logo is the red fox mascot.",
+                resources: [
+                    {
+                        type: "asset",
+                        id: asset.id,
+                        uri: `asset://${asset.id}.png`,
+                        label: "approved-logo.png"
+                    }
+                ]
+            });
+        }),
+        expect: {
+            requiredTools: ["memory_list"],
+            finalState: [atLeastOneMemory],
+            maxToolCalls: 10
+        }
+    },
+    {
+        id: "search-across-threads",
+        description: "Find, by keyword, a memory saved in a different conversation.",
+        objective: "In some earlier conversation I settled on a brand colour. Search your memory for it and tell me the colour name.",
+        systemPrompt: MEMORY_SYSTEM_PROMPT,
+        createBridge: () => createMemoryToolBridge(async () => {
+            // Deliberately NOT this thread: only a cross-thread read finds it.
+            await Memory.create({
+                user_id: EVAL_USER,
+                thread_id: "an-earlier-conversation",
+                kind: "decision",
+                title: "Brand colour",
+                content: "We settled on viridian as the brand colour."
+            });
+        }),
+        expect: {
+            requiredTools: ["memory_search"],
+            finalState: [paletteMemoryIntact],
+            maxToolCalls: 10
+        }
+    }
+];
+//# sourceMappingURL=memory.js.map

package/dist/evals/surfaces/memory.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"memory.js","sourceRoot":"","sources":["../../../src/evals/surfaces/memory.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;GAUG;AAEH,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AACxB,OAAO,EACL,KAAK,EACL,MAAM,EACN,UAAU,EAEX,MAAM,qBAAqB,CAAC;AAE7B,OAAO,EAAE,qBAAqB,EAAE,MAAM,iCAAiC,CAAC;AAOxE,OAAO,EAAE,QAAQ,EAAE,MAAM,4BAA4B,CAAC;AAEtD,MAAM,SAAS,GAAG,WAAW,CAAC;AAC9B,MAAM,WAAW,GAAG,aAAa,CAAC;AAYlC,MAAM,aAAa,GAAG,CAAC,CAAC,MAAM,CAAC;IAC7B,IAAI,EAAE,CAAC,CAAC,MAAM,EAAE;IAChB,EAAE,EAAE,CAAC,CAAC,MAAM,EAAE;IACd,GAAG,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE;IAC1B,KAAK,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE;CAC7B,CAAC,CAAC;AAEH,SAAS,IAAI,CAAC,IAAY;IACxB,OAAO,CACL,IAAI;SACD,WAAW,EAAE;SACb,OAAO,CAAC,aAAa,EAAE,GAAG,CAAC;SAC3B,OAAO,CAAC,UAAU,EAAE,EAAE,CAAC;SACvB,KAAK,CAAC,CAAC,EAAE,EAAE,CAAC,IAAI,OAAO,CAC3B,CAAC;AACJ,CAAC;AAED;;;GAGG;AACH,MAAM,UAAU,sBAAsB,CACpC,IAAgD;IAEhD,UAAU,EAAE,CAAC;IACb,MAAM,UAAU,GAAmD;QACjE,MAAM,EAAE,SAAS;QACjB,QAAQ,EAAE,WAAW;KACtB,CAAC;IACF,qEAAqE;IACrE,8DAA8D;IAC9D,MAAM,GAAG,GAAG,UAA+B,CAAC;IAE5C,MAAM,QAAQ,GAAG,qBAAqB,CAAC,aAAa,CAAC,CAAC;IACtD,MAAM,QAAQ,GAAG,qBAAqB,CAAC,aAAa,CAAC,CAAC;IACtD,MAAM,gBAAgB,GAAG,qBAAqB,CAAC,eAAe,CAAC,CAAC;IAChE,MAAM,UAAU,GAAG,qBAAqB,CAAC,cAAc,CAAC,CAAC;IAEzD,IAAI,KAAK,GAA2B,EAAE,QAAQ,EAAE,EAAE,EAAE,MAAM,EAAE,EAAE,EAAE,CAAC;IAEjE,MAAM,OAAO,GAAG,KAAK,IAAmB,EAAE;QACxC,yEAAyE;QACzE,iCAAiC;QACjC,MAAM,IAAI,GAAG,MAAM,MAAM,CAAC,IAAI,CAAC,SAAS,EAAE,EAAE,KAAK,EAAE,GAAG,EAAE,CAAC,CAAC;QAC1D,MAAM,CAAC,SAAS,CAAC,GAAG,MAAM,KAAK,CAAC,kBAAkB,CAAC,SAAS,EAAE,EAAE,EAAE;YAChE,KAAK,EAAE,GAAG;SACX,CAAC,CAAC;QACH,KAAK,GAAG;YACN,QAAQ,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC;gBACzB,IAAI,EAAE,CAAC,CAAC,IAAI;gBACZ,OAAO,EAAE,CAAC,CAAC,OAAO;gBAClB,SAAS,EAAE,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,EAAE;aACzD,CAAC,CAAC;YACH,MAAM,EAAE,SAAS;iBACd,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,YAAY,KAAK,QAAQ,CAAC;iBAC1C,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC;gBACX,EAAE,EAAE,CAAC,CAAC,EAAE;gBACR,IAAI,EAAE,CAAC,CAAC,IAAI;gBACZ,YAAY,EAAE,CAAC,CAAC,YAAY;aAC7B,CAAC,CAAC;SACN,CAAC;IACJ,CAAC,CAAC;IAEF,4EAA4E;IAC5E,0EAA0E;IAC1E,oEAAoE;IACpE,MAAM,KAAK,GAAkB,CAAC,KAAK,IAAI,EAAE;QACvC,IAAI,IAAI;YAAE,MAAM,IAAI,CAAC,GAAG,CAAC,CAAC;QAC1B,MAAM,OAAO,EAAE,CAAC;IAClB,CAAC,CAAC,EAAE,CAAC;IAEL,MAAM,KAAK,GAAmB;QAC5B;YACE,IAAI,EAAE,gBAAgB;YACtB,WAAW,EACT,wEAAwE;gBACxE,wCAAwC;YAC1C,UAAU,EAAE,CAAC,CAAC,MAAM,CAAC;gBACnB,MAAM,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAAC,mBAAmB,CAAC;aACjD,CAAC;YACF,OAAO,EAAE,KAAK,EAAE,IAAI,EAAE,EAAE;gBACtB,MAAM,KAAK,CAAC;gBACZ,MAAM,MAAM,GAAG,QAAQ,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC,CAAC,OAAO,CAAC;gBAC7D,MAAM,KAAK,GAAG,MAAM,KAAK,CAAC,MAAM,CAAQ;oBACtC,OAAO,EAAE,SAAS;oBAClB,IAAI,EAAE,GAAG,IAAI,CAAC,MAAM,CAAC,MAAM;oBAC3B,YAAY,EAAE,WAAW;iBAC1B,CAAC,CAAC;gBACH,MAAM,OAAO,EAAE,CAAC;gBAChB,OAAO;oBACL,OAAO,EAAE,IAAI;oBACb,QAAQ,EAAE,KAAK,CAAC,EAAE;oBAClB,GAAG,EAAE,WAAW,KAAK,CAAC,EAAE,MAAM;oBAC9B,YAAY,EAAE,WAAW;iBAC1B,CAAC;YACJ,CAAC;SACF;QACD;YACE,IAAI,EAAE,QAAQ,CAAC,IAAI;YACnB,WAAW,EAAE,QAAQ,CAAC,WAAW;YACjC,UAAU,EAAE,CAAC,CAAC,MAAM,CAAC;gBACnB,OAAO,EAAE,CAAC,CAAC,MAAM,EAAE;gBACnB,KAAK,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE;gBAC5B,IAAI,EAAE,CAAC;qBACJ,IAAI,CAAC,CAAC,MAAM,EAAE,MAAM,EAAE,YAAY,EAAE,UAAU,EAAE,UAAU,CAAC,CAAC;qBAC5D,QAAQ,EAAE;gBACb,SAAS,EAAE,CAAC,CAAC,KAAK,CAAC,aAAa,CAAC,CAAC,QAAQ,EAAE;aAC7C,CAAC;YACF,OAAO,EAAE,KAAK,EAAE,IAAI,EAAE,EAAE;gBACtB,MAAM,KAAK,CAAC;gBACZ,MAAM,MAAM,GAAG,MAAM,QAAQ,CAAC,OAAO,CAAC,GAAG,EAAE,IAAI,CAAC,CAAC;gBACjD,MAAM,OAAO,EAAE,CAAC;gBAChB,OAAO,MAAM,CAAC;YAChB,CAAC;SACF;QACD;YACE,IAAI,EAAE,QAAQ,CAAC,IAAI;YACnB,WAAW,EAAE,QAAQ,CAAC,WAAW;YACjC,UAAU,EAAE,CAAC,CAAC,MAAM,CAAC,EAAE,KAAK,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,EAAE,CAAC;YACtD,OAAO,EAAE,KAAK,EAAE,IAAI,EAAE,EAAE;gBACtB,MAAM,KAAK,CAAC;gBACZ,MAAM,MAAM,GAAG,MAAM,QAAQ,CAAC,OAAO,CAAC,GAAG,EAAE,IAAI,CAAC,CAAC;gBACjD,MAAM,OAAO,EAAE,CAAC;gBAChB,OAAO,MAAM,CAAC;YAChB,CAAC;SACF;QACD;YACE,IAAI,EAAE,gBAAgB,CAAC,IAAI;YAC3B,WAAW,EAAE,gBAAgB,CAAC,WAAW;YACzC,UAAU,EAAE,CAAC,CAAC,MAAM,CAAC;gBACnB,KAAK,EAAE,CAAC,CAAC,MAAM,EAAE;gBACjB,MAAM,EAAE,CAAC,CAAC,IAAI,CAAC,CAAC,KAAK,EAAE,SAAS,CAAC,CAAC,CAAC,QAAQ,EAAE;gBAC7C,KAAK,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE;aAC7B,CAAC;YACF,OAAO,EAAE,KAAK,EAAE,IAAI,EAAE,EAAE;gBACtB,MAAM,KAAK,CAAC;gBACZ,MAAM,MAAM,GAAG,MAAM,gBAAgB,CAAC,OAAO,CAAC,GAAG,EAAE,IAAI,CAAC,CAAC;gBACzD,MAAM,OAAO,EAAE,CAAC;gBAChB,OAAO,MAAM,CAAC;YAChB,CAAC;SACF;QACD;YACE,IAAI,EAAE,UAAU,CAAC,IAAI;YACrB,WAAW,EAAE,UAAU,CAAC,WAAW;YACnC,UAAU,EAAE,CAAC,CAAC,MAAM,CAAC;gBACnB,KAAK,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE;gBAC5B,YAAY,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE;gBACnC,KAAK,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE;aAC7B,CAAC;YACF,OAAO,EAAE,KAAK,EAAE,IAAI,EAAE,EAAE;gBACtB,MAAM,KAAK,CAAC;gBACZ,MAAM,MAAM,GAAG,MAAM,UAAU,CAAC,OAAO,CAAC,GAAG,EAAE,IAAI,CAAC,CAAC;gBACnD,MAAM,OAAO,EAAE,CAAC;gBAChB,OAAO,MAAM,CAAC;YAChB,CAAC;SACF;KACF,CAAC;IAEF,OAAO,EAAE,KAAK,EAAE,UAAU,EAAE,GAAG,EAAE,CAAC,KAAK,EAAE,CAAC;AAC5C,CAAC;AAED,gFAAgF;AAEhF,MAAM,uBAAuB,GAC3B;IACE,IAAI,EAAE,6CAA6C;IACnD,IAAI,EAAE,CAAC,CAAC,EAAE,EAAE,CACV,CAAC,CAAC,QAAQ,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CACpB,CAAC,CAAC,SAAS,CAAC,IAAI,CACd,CAAC,CAAC,EAAE,EAAE,CACJ,CAAC,CAAC,IAAI,KAAK,OAAO;QAClB,QAAQ,CAAC,CAAC,CAAC,EAAE,CAAC;QACd,CAAC,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,EAAE,KAAK,CAAC,CAAC,EAAE,CAAC,CACtC,CACF;IACH,MAAM,EACJ,gEAAgE;QAChE,oCAAoC;CACvC,CAAC;AAEJ,MAAM,gBAAgB,GAAmD;IACvE,IAAI,EAAE,+BAA+B;IACrC,IAAI,EAAE,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,QAAQ,CAAC,MAAM,IAAI,CAAC;CACpC,CAAC;AAEF,+EAA+E;AAC/E,MAAM,mBAAmB,GAAmD;IAC1E,IAAI,EAAE,2DAA2D;IACjE,IAAI,EAAE,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,QAAQ,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,WAAW,CAAC,IAAI,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC;IAChE,MAAM,EACJ,wEAAwE;CAC3E,CAAC;AAEF,MAAM,oBAAoB,GAAG;;;;;;;;;yJAS4H,CAAC;AAE1J,MAAM,CAAC,MAAM,sBAAsB,GACjC;IACE;QACE,EAAE,EAAE,uBAAuB;QAC3B,WAAW,EACT,sFAAsF;QACxF,SAAS,EACP,mJAAmJ;QACrJ,YAAY,EAAE,oBAAoB;QAClC,YAAY,EAAE,GAAG,EAAE,CAAC,sBAAsB,EAAE;QAC5C,MAAM,EAAE;YACN,aAAa,EAAE,CAAC,gBAAgB,EAAE,aAAa,CAAC;YAChD,QAAQ,EAAE,CAAC,CAAC,gBAAgB,EAAE,aAAa,CAAC,CAAC;YAC7C,UAAU,EAAE,CAAC,uBAAuB,CAAC;YACrC,YAAY,EAAE,CAAC;YACf,YAAY,EAAE,EAAE;SACjB;KACF;IACD;QACE,EAAE,EAAE,iBAAiB;QACrB,WAAW,EACT,oEAAoE;QACtE,SAAS,EACP,gJAAgJ;QAClJ,YAAY,EAAE,oBAAoB;QAClC,YAAY,EAAE,GAAG,EAAE,CACjB,sBAAsB,CAAC,KAAK,IAAI,EAAE;YAChC,MAAM,KAAK,GAAG,MAAM,KAAK,CAAC,MAAM,CAAQ;gBACtC,OAAO,EAAE,SAAS;gBAClB,IAAI,EAAE,mBAAmB;gBACzB,YAAY,EAAE,WAAW;aAC1B,CAAC,CAAC;YACH,MAAM,MAAM,CAAC,MAAM,CAAS;gBAC1B,OAAO,EAAE,SAAS;gBAClB,SAAS,EAAE,WAAW;gBACtB,IAAI,EAAE,UAAU;gBAChB,OAAO,EAAE,kDAAkD;gBAC3D,SAAS,EAAE;oBACT;wBACE,IAAI,EAAE,OAAO;wBACb,EAAE,EAAE,KAAK,CAAC,EAAE;wBACZ,GAAG,EAAE,WAAW,KAAK,CAAC,EAAE,MAAM;wBAC9B,KAAK,EAAE,mBAAmB;qBAC3B;iBACF;aACF,CAAC,CAAC;QACL,CAAC,CAAC;QACJ,MAAM,EAAE;YACN,aAAa,EAAE,CAAC,aAAa,CAAC;YAC9B,UAAU,EAAE,CAAC,gBAAgB,CAAC;YAC9B,YAAY,EAAE,EAAE;SACjB;KACF;IACD;QACE,EAAE,EAAE,uBAAuB;QAC3B,WAAW,EACT,+DAA+D;QACjE,SAAS,EACP,kHAAkH;QACpH,YAAY,EAAE,oBAAoB;QAClC,YAAY,EAAE,GAAG,EAAE,CACjB,sBAAsB,CAAC,KAAK,IAAI,EAAE;YAChC,mEAAmE;YACnE,MAAM,MAAM,CAAC,MAAM,CAAS;gBAC1B,OAAO,EAAE,SAAS;gBAClB,SAAS,EAAE,yBAAyB;gBACpC,IAAI,EAAE,UAAU;gBAChB,KAAK,EAAE,cAAc;gBACrB,OAAO,EAAE,6CAA6C;aACvD,CAAC,CAAC;QACL,CAAC,CAAC;QACJ,MAAM,EAAE;YACN,aAAa,EAAE,CAAC,eAAe,CAAC;YAChC,UAAU,EAAE,CAAC,mBAAmB,CAAC;YACjC,YAAY,EAAE,EAAE;SACjB;KACF;CACF,CAAC"}

package/dist/evals/surfaces/model3d.d.ts

@@ -0,0 +1,45 @@
+/**
+ * Headless bridge for the 3D model editor tool-loop eval.
+ *
+ * The real frontend tools (`web/src/lib/tools/builtin/model3d.ts`) delegate to
+ * a `Model3DToolHandler` the live `Model3DEditor` registers on
+ * `model3DToolBridge` — it mutates a Three.js scene graph and (for
+ * `ui_3d_capture_view`) renders a WebGL frame. None of that can run under
+ * Node. This bridge reimplements the *effects* of the non-rendering tools
+ * against a plain in-memory scene graph, so a model can drive the same
+ * `ui_3d_*` tool surface headlessly.
+ *
+ * What it does NOT fork is the tool *contract*: names, descriptions, and Zod
+ * parameter shapes are copied verbatim from the builtin file — the single
+ * source of truth the browser tools also expose. `ui_3d_capture_view` is
+ * intentionally excluded: it requires a real WebGL render and has no
+ * meaningful headless equivalent.
+ */
+import type { HeadlessSurfaceBridge, ToolLoopEvalCase } from "../tool-loop-eval.js";
+export type Model3DPrimitiveKind = "box" | "sphere" | "plane" | "cylinder" | "torus" | "directionalLight" | "pointLight";
+export interface Model3DBridgeInitialState {
+    objects?: {
+        name: string;
+        kind: Model3DPrimitiveKind;
+    }[];
+}
+export interface Model3DBridgeFinalState {
+    selectedUuid: string | null;
+    objects: {
+        uuid: string;
+        name: string;
+        type: string;
+        visible: boolean;
+        position: [number, number, number];
+        rotation: [number, number, number];
+        scale: [number, number, number];
+        materialColor?: string;
+    }[];
+}
+/**
+ * Build an in-memory 3D-scene bridge whose tools share the `ui_3d_*` contract
+ * but run headlessly against a plain scene-graph array (no Three.js).
+ */
+export declare function createModel3DToolBridge(initial?: Model3DBridgeInitialState): HeadlessSurfaceBridge<Model3DBridgeFinalState>;
+export declare const MODEL3D_TOOL_LOOP_CASES: readonly ToolLoopEvalCase<Model3DBridgeFinalState>[];
+//# sourceMappingURL=model3d.d.ts.map

package/dist/evals/surfaces/model3d.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"model3d.d.ts","sourceRoot":"","sources":["../../../src/evals/surfaces/model3d.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;GAgBG;AAKH,OAAO,KAAK,EACV,qBAAqB,EACrB,gBAAgB,EAEjB,MAAM,sBAAsB,CAAC;AAE9B,MAAM,MAAM,oBAAoB,GAC5B,KAAK,GACL,QAAQ,GACR,OAAO,GACP,UAAU,GACV,OAAO,GACP,kBAAkB,GAClB,YAAY,CAAC;AAiBjB,MAAM,WAAW,yBAAyB;IACxC,OAAO,CAAC,EAAE;QAAE,IAAI,EAAE,MAAM,CAAC;QAAC,IAAI,EAAE,oBAAoB,CAAA;KAAE,EAAE,CAAC;CAC1D;AAED,MAAM,WAAW,uBAAuB;IACtC,YAAY,EAAE,MAAM,GAAG,IAAI,CAAC;IAC5B,OAAO,EAAE;QACP,IAAI,EAAE,MAAM,CAAC;QACb,IAAI,EAAE,MAAM,CAAC;QACb,IAAI,EAAE,MAAM,CAAC;QACb,OAAO,EAAE,OAAO,CAAC;QACjB,QAAQ,EAAE,CAAC,MAAM,EAAE,MAAM,EAAE,MAAM,CAAC,CAAC;QACnC,QAAQ,EAAE,CAAC,MAAM,EAAE,MAAM,EAAE,MAAM,CAAC,CAAC;QACnC,KAAK,EAAE,CAAC,MAAM,EAAE,MAAM,EAAE,MAAM,CAAC,CAAC;QAChC,aAAa,CAAC,EAAE,MAAM,CAAC;KACxB,EAAE,CAAC;CACL;AA6DD;;;GAGG;AACH,wBAAgB,uBAAuB,CACrC,OAAO,GAAE,yBAA8B,GACtC,qBAAqB,CAAC,uBAAuB,CAAC,CAyNhD;AAoBD,eAAO,MAAM,uBAAuB,EAAE,SAAS,gBAAgB,CAAC,uBAAuB,CAAC,EAgHrF,CAAC"}