@pixeldojo/mcp 0.5.1 → 0.8.0

package/dist/cli.js

@@ -181,6 +181,85 @@ var init_upload_helpers = __esm({
     };
   }
 });
+
+// src/skills/schema-snap.ts
+function getFieldConstraints(apiId, client) {
+  const cached = schemaCache.get(apiId);
+  if (cached) return cached;
+  const pending = client.getModelSchema(apiId).then(
+    (res) => res.schema?.properties ?? {}
+  ).catch(() => {
+    schemaCache.delete(apiId);
+    return null;
+  });
+  schemaCache.set(apiId, pending);
+  return pending;
+}
+function parseAspectRatio(ar) {
+  const m = /^\s*(\d+)\s*:\s*(\d+)\s*$/.exec(ar);
+  if (!m) return null;
+  const w = Number(m[1]);
+  const h = Number(m[2]);
+  return h > 0 ? w / h : null;
+}
+function nearestAspectFromEnum(requested, allowed) {
+  if (allowed.includes(requested)) return requested;
+  const target = parseAspectRatio(requested);
+  if (target === null) return requested;
+  let best = requested;
+  let bestDiff = Infinity;
+  for (const candidate of allowed) {
+    const r = parseAspectRatio(candidate);
+    if (r === null) continue;
+    const diff = Math.abs(r - target);
+    if (diff < bestDiff) {
+      bestDiff = diff;
+      best = candidate;
+    }
+  }
+  return best;
+}
+async function snapAspectRatio(apiId, requested, client, opts = {}) {
+  if (!requested) return void 0;
+  const props = await getFieldConstraints(apiId, client);
+  if (props === null) {
+    return opts.onUnknown === "omit" ? void 0 : requested;
+  }
+  const field = props.aspect_ratio;
+  if (!field) return void 0;
+  if (!field.enum || field.enum.length === 0) return requested;
+  return nearestAspectFromEnum(requested, field.enum);
+}
+async function snapDuration(apiId, requested, client, opts = {}) {
+  if (requested === void 0) return void 0;
+  const props = await getFieldConstraints(apiId, client);
+  if (props === null) {
+    return opts.onUnknown === "omit" ? void 0 : requested;
+  }
+  const field = props.duration;
+  if (!field) return void 0;
+  if (field.enum && field.enum.length > 0) {
+    const allowed = field.enum.map(Number).filter((n) => Number.isFinite(n));
+    if (allowed.length === 0) return requested;
+    if (allowed.includes(requested)) return requested;
+    return allowed.reduce(
+      (best, candidate) => Math.abs(candidate - requested) < Math.abs(best - requested) ? candidate : best
+    );
+  }
+  if (typeof field.minimum === "number" || typeof field.maximum === "number") {
+    let v = requested;
+    if (typeof field.minimum === "number") v = Math.max(field.minimum, v);
+    if (typeof field.maximum === "number") v = Math.min(field.maximum, v);
+    return v;
+  }
+  return requested;
+}
+var schemaCache;
+var init_schema_snap = __esm({
+  "src/skills/schema-snap.ts"() {
+    schemaCache = /* @__PURE__ */ new Map();
+  }
+});
 function buildPayload(apiId, input) {
   const body = { prompt: input.prompt };
   if (input.aspect_ratio) body.aspect_ratio = input.aspect_ratio;
@@ -233,6 +312,13 @@ async function run(input, ctx) {
       };
     }
   }
+  if (!VIDEO_INPUT_MODELS.has(apiId)) {
+    const [aspect_ratio, duration] = await Promise.all([
+      snapAspectRatio(apiId, resolved.aspect_ratio, ctx.client),
+      snapDuration(apiId, resolved.duration, ctx.client)
+    ]);
+    resolved = { ...resolved, aspect_ratio, duration };
+  }
   const body = buildPayload(apiId, resolved);
   const submitted = await ctx.client.run(apiId, body);
   const result = await pollJobUntilDoneOrHandoff(
@@ -249,16 +335,17 @@ async function run(input, ctx) {
     (urls) => urls.length === 0 ? `Job ${submitted.jobId} completed but returned no assets.` : urls.length === 1 ? `Generated. Asset: ${urls[0]}` : `Generated ${urls.length} assets. URLs: ${urls.join(", ")}`
   );
 }
-var InputSchema, DEFAULT_MODEL, InputJsonSchema, generateSkill;
+var InputSchema, DEFAULT_MODEL, VIDEO_INPUT_MODELS, InputJsonSchema, generateSkill;
 var init_generate = __esm({
   "src/skills/generate.ts"() {
     init_poll();
     init_types();
     init_upload_helpers();
+    init_schema_snap();
     InputSchema = z.object({
       prompt: z.string().min(1).max(8e3).describe("Prompt describing what to generate."),
       model: z.string().optional().describe(
-        'Override the model. Any apiId from /api/v1/models (e.g. "flux", "google-nano-banana", "veo-3.1", "kling-video"). Omit to use the default.'
+        'Override the model. Any apiId from /api/v1/models (e.g. "flux", "google-nano-banana", "veo-3.1-fast", "kling-video-v3-standard-text"). Omit to use the default.'
       ),
       preset: z.string().optional().describe(
         'Preset slug to seed the params (e.g. "cinematic-portrait", "product-floating-clean", "magazine-cover-text"). Browse the catalog at https://pixeldojo.ai/presets. Explicit fields override preset values.'
@@ -271,13 +358,14 @@ var init_generate = __esm({
       )
     });
     DEFAULT_MODEL = "google-nano-banana";
+    VIDEO_INPUT_MODELS = /* @__PURE__ */ new Set(["grok-video-extend", "wan-video-edit", "seedance-video-edit"]);
     InputJsonSchema = {
       type: "object",
       properties: {
         prompt: { type: "string", description: "Prompt describing what to generate." },
         model: {
           type: "string",
-          description: 'Override the model. Any apiId from /api/v1/models (e.g. "flux", "google-nano-banana", "veo-3.1", "kling-video"). Omit to use the default.'
+          description: 'Override the model. Any apiId from /api/v1/models (e.g. "flux", "google-nano-banana", "veo-3.1-fast", "kling-video-v3-standard-text"). Omit to use the default.'
         },
         preset: {
           type: "string",
@@ -368,6 +456,10 @@ async function run2(input, ctx) {
     };
   }
   const apiId = resolved.model ?? DEFAULT_MODEL2;
+  resolved = {
+    ...resolved,
+    aspect_ratio: await snapAspectRatio(apiId, resolved.aspect_ratio, ctx.client)
+  };
   const body = buildPayload2(apiId, resolved);
   const submitted = await ctx.client.run(apiId, body);
   const result = await pollJobUntilDoneOrHandoff(
@@ -390,6 +482,7 @@ var init_edit = __esm({
     init_poll();
     init_types();
     init_upload_helpers();
+    init_schema_snap();
     DEFAULT_MODEL2 = "wan-2.7-image-edit";
     ALLOWED_MODELS = [
       "wan-2.7-image-edit",
@@ -471,18 +564,24 @@ async function run3(input, ctx) {
     };
   }
   const apiId = requested;
-  if ((apiId === "flux-edit" || apiId === "gpt-image-2-edit") && !input.reference_image_url) {
+  if (REQUIRES_REFERENCE.has(apiId) && !input.reference_image_url) {
     return {
       kind: "text",
-      text: `pixeldojo:character with model="${apiId}" requires reference_image_url (the source image of the character to edit). For generation from a prompt alone, use model="ideogram-character".`
+      text: `pixeldojo:character with model="${apiId}" requires reference_image_url (the source image of the character to edit). For generation from a prompt alone, use model="${DEFAULT_MODEL3}".`
     };
   }
   const body = { prompt: input.prompt };
-  if (input.aspect_ratio) body.aspect_ratio = input.aspect_ratio;
   if (input.reference_image_url) {
-    body.image_url = input.reference_image_url;
-    body.reference_image_url = input.reference_image_url;
+    if (apiId === "flux-edit") {
+      body.input_image = input.reference_image_url;
+    } else {
+      body.image_urls = [input.reference_image_url];
+    }
   }
+  const snappedAspect = await snapAspectRatio(apiId, input.aspect_ratio, ctx.client, {
+    onUnknown: "omit"
+  });
+  if (snappedAspect !== void 0) body.aspect_ratio = snappedAspect;
   const submitted = await ctx.client.run(apiId, body);
   const result = await pollJobUntilDoneOrHandoff(
     submitted.jobId,
@@ -498,20 +597,22 @@ async function run3(input, ctx) {
     (urls) => urls.length === 0 ? `Job ${submitted.jobId} completed but returned no assets.` : `Character shot ready. Asset: ${urls[0]}`
   );
 }
-var ALLOWED, DEFAULT_MODEL3, InputSchema3, InputJsonSchema3, characterSkill;
+var ALLOWED, DEFAULT_MODEL3, REQUIRES_REFERENCE, InputSchema3, InputJsonSchema3, characterSkill;
 var init_character = __esm({
   "src/skills/character.ts"() {
     init_poll();
     init_types();
-    ALLOWED = ["ideogram-character", "flux-edit", "gpt-image-2-edit"];
-    DEFAULT_MODEL3 = "ideogram-character";
+    init_schema_snap();
+    ALLOWED = ["seedream-5", "seedream-4", "flux-edit", "gpt-image-2-edit"];
+    DEFAULT_MODEL3 = "seedream-5";
+    REQUIRES_REFERENCE = /* @__PURE__ */ new Set(["flux-edit", "gpt-image-2-edit"]);
     InputSchema3 = z.object({
       prompt: z.string().min(1).max(8e3),
       model: z.string().optional().describe(`One of: ${ALLOWED.join(", ")}. Defaults to ${DEFAULT_MODEL3}.`),
       reference_image_url: z.string().url().optional().describe(
-        "Reference image of the character. Required by flux-edit and gpt-image-2-edit; optional for ideogram-character."
+        "Reference image of the character. Required by flux-edit and gpt-image-2-edit; optional for seedream-5/seedream-4."
       ),
-      aspect_ratio: z.string().optional()
+      aspect_ratio: z.string().optional().describe('Only honored for model="flux-edit".')
     });
     InputJsonSchema3 = {
       type: "object",
@@ -529,7 +630,7 @@ var init_character = __esm({
     };
     characterSkill = {
       name: "pixeldojo:character",
-      description: "Generate or edit consistent characters across shots. Pass a reference_image_url for edit-style models.",
+      description: "Generate or edit consistent characters across shots. Pass a reference_image_url for identity locking (seedream-5 by default).",
       inputZod: InputSchema3,
       inputJsonSchema: InputJsonSchema3,
       allowedModels: [...ALLOWED],
@@ -547,9 +648,13 @@ async function run4(input, ctx) {
     };
   }
   const apiId = requested;
+  const [snappedAspect, snappedDuration] = await Promise.all([
+    snapAspectRatio(apiId, input.aspect_ratio, ctx.client),
+    snapDuration(apiId, input.duration, ctx.client)
+  ]);
   const body = { prompt: input.brief };
-  if (input.aspect_ratio) body.aspect_ratio = input.aspect_ratio;
-  if (input.duration !== void 0) body.duration = input.duration;
+  if (snappedAspect !== void 0) body.aspect_ratio = snappedAspect;
+  if (snappedDuration !== void 0) body.duration = snappedDuration;
   const submitted = await ctx.client.run(apiId, body);
   const result = await pollJobUntilDoneOrHandoff(
     submitted.jobId,
@@ -570,8 +675,14 @@ var init_storyboard = __esm({
   "src/skills/storyboard.ts"() {
     init_poll();
     init_types();
-    ALLOWED2 = ["veo-3.1", "kling-video", "seedance-2"];
-    DEFAULT_MODEL4 = "veo-3.1";
+    init_schema_snap();
+    ALLOWED2 = [
+      "veo-3.1-fast",
+      "veo-3.1-standard",
+      "kling-video-v3-standard-text",
+      "seedance-2-high"
+    ];
+    DEFAULT_MODEL4 = "veo-3.1-fast";
     InputSchema4 = z.object({
       brief: z.string().min(1).max(8e3).describe("Storyboard brief: scene, shots, mood, action."),
       model: z.string().optional().describe(`One of: ${ALLOWED2.join(", ")}. Defaults to ${DEFAULT_MODEL4}.`),
@@ -617,11 +728,20 @@ async function run5(input, ctx) {
   const apiId = requested;
   const body = {
     prompt: "",
-    image_url: input.image_url
+    image: input.image_url
   };
   if (apiId === "magnific-upscaler") {
-    body.scale_factor = input.scale_factor ?? 2;
+    const allowed = [2, 4, 8, 16];
+    const requestedScale = input.scale_factor ?? 2;
+    const nearest = allowed.reduce(
+      (a, b) => Math.abs(b - requestedScale) < Math.abs(a - requestedScale) ? b : a
+    );
+    body.scale_factor = `${nearest}x`;
     if (input.creativity !== void 0) body.creativity = input.creativity;
+  } else if (apiId === "creative-upscale") {
+    body.scale_factor = Math.min(10, Math.max(1, input.scale_factor ?? 2));
+  } else if (apiId === "face-enhance") {
+    body.scale_factor = Math.min(8, Math.max(1, input.scale_factor ?? 4));
   }
   const submitted = await ctx.client.run(apiId, body);
   const result = await pollJobUntilDoneOrHandoff(
@@ -648,7 +768,9 @@ var init_upscale = __esm({
     InputSchema5 = z.object({
       image_url: z.string().url().describe("URL of the source image to upscale."),
       model: z.string().optional().describe(`One of: ${ALLOWED3.join(", ")}. Defaults to ${DEFAULT_MODEL5}.`),
-      scale_factor: z.number().int().min(1).max(4).optional().describe("Magnific only: 1, 2, or 4. Defaults to 2."),
+      scale_factor: z.number().min(1).max(16).optional().describe(
+        "Upscale multiplier. Magnific snaps to the nearest of 2/4/8/16 (default 2); creative-upscale accepts 1-10 (default 2); face-enhance accepts 1-8 (default 4)."
+      ),
       creativity: z.number().min(0).max(1).optional().describe("Magnific only: 0 (preserve) \u2192 1 (creative). Defaults to 0.")
     });
     InputJsonSchema5 = {
@@ -656,7 +778,7 @@ var init_upscale = __esm({
       properties: {
         image_url: { type: "string", format: "uri" },
         model: { type: "string", enum: ALLOWED3 },
-        scale_factor: { type: "integer", minimum: 1, maximum: 4 },
+        scale_factor: { type: "number", minimum: 1, maximum: 16 },
         creativity: { type: "number", minimum: 0, maximum: 1 }
       },
       required: ["image_url"]
@@ -812,6 +934,146 @@ var init_from_url = __esm({
   }
 });
 async function run8(input, ctx) {
+  const productImages = [
+    ...input.product_image_urls ?? [],
+    ...await Promise.all((input.product_image_paths ?? []).map((p) => uploadLocalFile(p, ctx.client)))
+  ];
+  const characterImages = [
+    ...input.character_image_urls ?? [],
+    ...await Promise.all((input.character_image_paths ?? []).map((p) => uploadLocalFile(p, ctx.client)))
+  ];
+  const body = {
+    style: input.style,
+    product_url: input.product_url,
+    product: !input.product_url && input.product_name ? {
+      name: input.product_name,
+      description: input.product_description,
+      images: productImages
+    } : void 0,
+    character_images: characterImages.length > 0 ? characterImages : void 0,
+    location: input.location_description || input.location_image_url ? { description: input.location_description, image_url: input.location_image_url } : void 0,
+    hook: input.hook,
+    script: input.script,
+    quality: input.quality,
+    resolution: input.resolution,
+    aspect_ratio: input.aspect_ratio,
+    duration: input.duration
+  };
+  const client = ctx.client;
+  const submitted = await client.runAd(body);
+  const result = await pollJobUntilDoneOrHandoff(
+    submitted.jobId,
+    (id) => ctx.client.getJob(id),
+    ctx.buildStatusUrl,
+    {
+      now: ctx.pollOverrides?.now,
+      sleep: ctx.pollOverrides?.sleep
+    }
+  );
+  const styleLabel = submitted.ad?.style ?? input.style ?? "ugc-testimonial";
+  return pollResultToSkillResult(
+    result,
+    (urls) => urls.length === 0 ? `Ad job ${submitted.jobId} completed but returned no assets.` : `Ad rendered (style: ${styleLabel}${submitted.ad?.product?.name ? `, product: ${submitted.ad.product.name}` : ""}). Video: ${urls[0]}`
+  );
+}
+var InputSchema8, InputJsonSchema8, adSkill;
+var init_ad = __esm({
+  "src/skills/ad.ts"() {
+    init_types();
+    init_poll();
+    init_upload_helpers();
+    InputSchema8 = z.object({
+      style: z.string().max(80).optional().describe(
+        "Ad style slug (default ugc-testimonial). Catalog: GET /api/v1/marketing/presets \u2014 e.g. ugc-testimonial, unboxing-reveal, unboxing-asmr, virtual-try-on, street-style, product-showcase, macro-details, crush-test, hyper-motion, before-after, day-in-the-life, the-gift, tv-spot."
+      ),
+      product_url: z.string().url().optional().describe("Product page URL. Name, description, and images are extracted automatically."),
+      product_name: z.string().max(200).optional(),
+      product_description: z.string().max(2e3).optional(),
+      product_image_urls: z.array(z.string().url()).max(5).optional(),
+      product_image_paths: z.array(z.string()).max(5).optional().describe("Absolute local paths; uploaded to 24h temp storage automatically."),
+      character_image_urls: z.array(z.string().url()).max(4).optional().describe("Presenter reference images for a consistent on-camera character."),
+      character_image_paths: z.array(z.string()).max(4).optional(),
+      location_description: z.string().max(1500).optional().describe('Where the ad takes place ("a sunlit Parisian cafe"). Text is enough; no image needed.'),
+      location_image_url: z.string().url().optional(),
+      hook: z.string().max(1e3).optional().describe("The opening visual beat (first 1-2 seconds). Omit to use the style default."),
+      script: z.string().max(1e3).optional().describe("The exact spoken line, voiced word for word. Omit and the presenter improvises."),
+      quality: z.enum(["mini", "fast", "high"]).optional(),
+      resolution: z.enum(["480p", "720p", "1080p", "4k"]).optional().describe('1080p/4k require quality "high".'),
+      aspect_ratio: z.enum(["16:9", "4:3", "1:1", "3:4", "9:16", "21:9"]).optional(),
+      duration: z.number().int().min(4).max(15).optional()
+    }).refine(
+      (v) => v.product_url || (v.product_image_urls?.length || v.product_image_paths?.length) && v.product_name,
+      {
+        message: "Pass product_url, OR product_name plus product_image_urls/product_image_paths.",
+        path: ["product_url"]
+      }
+    );
+    InputJsonSchema8 = {
+      type: "object",
+      properties: {
+        style: {
+          type: "string",
+          description: "Ad style slug (default ugc-testimonial). Catalog: GET /api/v1/marketing/presets \u2014 e.g. ugc-testimonial, unboxing-reveal, unboxing-asmr, virtual-try-on, street-style, mirror-fit-check, pro-try-on, product-showcase, macro-details, secret-hack-reveal, crush-test, hyper-motion, before-after, day-in-the-life, the-gift, tv-spot, gadget-saved-me, selfie-testimonial, direct-to-camera, pack-an-order, mystery-box."
+        },
+        product_url: {
+          type: "string",
+          format: "uri",
+          description: "Product page URL. Name, description, and images are extracted automatically."
+        },
+        product_name: { type: "string" },
+        product_description: { type: "string" },
+        product_image_urls: { type: "array", items: { type: "string", format: "uri" }, maxItems: 5 },
+        product_image_paths: {
+          type: "array",
+          items: { type: "string" },
+          maxItems: 5,
+          description: "Absolute local file paths; uploaded to 24h temp storage automatically."
+        },
+        character_image_urls: {
+          type: "array",
+          items: { type: "string", format: "uri" },
+          maxItems: 4,
+          description: "Presenter reference images for a consistent on-camera character."
+        },
+        character_image_paths: { type: "array", items: { type: "string" }, maxItems: 4 },
+        location_description: {
+          type: "string",
+          description: 'Where the ad takes place ("a sunlit Parisian cafe"). Text is enough.'
+        },
+        location_image_url: { type: "string", format: "uri" },
+        hook: {
+          type: "string",
+          description: "The opening visual beat (first 1-2 seconds). Omit to use the style default."
+        },
+        script: {
+          type: "string",
+          description: "The exact spoken line, voiced word for word. Omit and the presenter improvises."
+        },
+        quality: { type: "string", enum: ["mini", "fast", "high"] },
+        resolution: {
+          type: "string",
+          enum: ["480p", "720p", "1080p", "4k"],
+          description: '1080p/4k require quality "high".'
+        },
+        aspect_ratio: { type: "string", enum: ["16:9", "4:3", "1:1", "3:4", "9:16", "21:9"] },
+        duration: { type: "integer", minimum: 4, maximum: 15 }
+      }
+      // "product_url OR product_name+images" is enforced by the server and the
+      // Zod refine; no root-level oneOf (Grok rejects non-object root unions).
+    };
+    adSkill = {
+      name: "pixeldojo:ad",
+      description: "Marketing Studio: turn any product into a video ad. Pass a product URL (auto-extracted) or product images, pick an ad style (UGC testimonial, unboxing, try-on, showcase, TV spot, and more via GET /api/v1/marketing/presets), optionally add presenter reference images, a location, a hook, and the exact spoken script. Renders with voiceover; returns the video or a statusUrl handoff to poll via pixeldojo:status. Ideal for automated generate-and-post workflows.",
+      inputZod: InputSchema8,
+      inputJsonSchema: InputJsonSchema8,
+      // Composition skill — the server routes to the ad engine internally.
+      allowedModels: [],
+      defaultModel: "",
+      run: run8
+    };
+  }
+});
+async function run9(input, ctx) {
   const client = ctx.client;
   const result = await client.runCampaign(input);
   if (result.kind === "campaign") {
@@ -838,10 +1100,10 @@ async function run8(input, ctx) {
     statusUrl: result.statusUrl ?? ctx.buildStatusUrl(result.campaignId)
   };
 }
-var InputSchema8, InputJsonSchema8, campaignSkill;
+var InputSchema9, InputJsonSchema9, campaignSkill;
 var init_campaign = __esm({
   "src/skills/campaign.ts"() {
-    InputSchema8 = z.object({
+    InputSchema9 = z.object({
       productUrl: z.string().url().optional().describe("Product page URL. We extract name + description + images automatically."),
       product: z.object({
         name: z.string().min(1).max(200),
@@ -857,7 +1119,7 @@ var init_campaign = __esm({
       message: "productUrl or product is required",
       path: ["productUrl"]
     });
-    InputJsonSchema8 = {
+    InputJsonSchema9 = {
       type: "object",
       properties: {
         productUrl: {
@@ -888,17 +1150,17 @@ var init_campaign = __esm({
     campaignSkill = {
       name: "pixeldojo:campaign",
       description: "Generate a coherent product launch package (hero image + N lifestyle variants + optional vertical video) from one URL or product profile. Composes hero / lifestyle / video models server-side; returns assets in one call or a handoff with campaignId to poll via pixeldojo:campaign_status.",
-      inputZod: InputSchema8,
-      inputJsonSchema: InputJsonSchema8,
+      inputZod: InputSchema9,
+      inputJsonSchema: InputJsonSchema9,
       // Composition skill — internal sub-jobs route to per-tool models, no
       // single allowed-models list applies here.
       allowedModels: [],
       defaultModel: "",
-      run: run8
+      run: run9
     };
   }
 });
-async function run9(input, ctx) {
+async function run10(input, ctx) {
   const client = ctx.client;
   const result = await client.getCampaign(input.campaignId);
   if (result.kind === "campaign") {
@@ -928,13 +1190,13 @@ async function run9(input, ctx) {
     statusUrl: result.statusUrl ?? ctx.buildStatusUrl(result.campaignId)
   };
 }
-var InputSchema9, InputJsonSchema9, campaignStatusSkill;
+var InputSchema10, InputJsonSchema10, campaignStatusSkill;
 var init_campaign_status = __esm({
   "src/skills/campaign-status.ts"() {
-    InputSchema9 = z.object({
+    InputSchema10 = z.object({
       campaignId: z.string().min(1).max(200).describe("Campaign id returned from a prior pixeldojo:campaign handoff.")
     });
-    InputJsonSchema9 = {
+    InputJsonSchema10 = {
       type: "object",
       properties: {
         campaignId: {
@@ -947,15 +1209,15 @@ var init_campaign_status = __esm({
     campaignStatusSkill = {
       name: "pixeldojo:campaign_status",
       description: "Poll a campaign by campaignId returned from pixeldojo:campaign. Returns assets when every sub-job is done, or a handoff describing what is still in flight.",
-      inputZod: InputSchema9,
-      inputJsonSchema: InputJsonSchema9,
+      inputZod: InputSchema10,
+      inputJsonSchema: InputJsonSchema10,
       allowedModels: [],
       defaultModel: "",
-      run: run9
+      run: run10
     };
   }
 });
-async function run10(input, ctx) {
+async function run11(input, ctx) {
   let url;
   try {
     url = await uploadLocalFile(input.path, ctx.client);
@@ -971,16 +1233,16 @@ async function run10(input, ctx) {
 (temp storage; auto-deletes in ~24h \u2014 pass this URL as image_url to :generate or :edit)`
   };
 }
-var InputSchema10, InputJsonSchema10, uploadSkill;
+var InputSchema11, InputJsonSchema11, uploadSkill;
 var init_upload = __esm({
   "src/skills/upload.ts"() {
     init_upload_helpers();
-    InputSchema10 = z.object({
+    InputSchema11 = z.object({
       path: z.string().min(1).max(4096).describe(
         "Absolute local file path to upload. JPEG, PNG, WebP, GIF (\u226450MB) or MP4, WebM, MOV (\u2264200MB). Returns a public URL on the temp CDN; storage auto-expires after 24 hours."
       )
     });
-    InputJsonSchema10 = {
+    InputJsonSchema11 = {
       type: "object",
       properties: {
         path: {
@@ -993,15 +1255,15 @@ var init_upload = __esm({
     uploadSkill = {
       name: "pixeldojo:upload",
       description: "Upload a local image or video file and get back a public URL you can pass to :generate or :edit as image_url. Use when the user has a reference image on disk (drag-dropped into Claude Code, screenshot, etc.) and you need a URL. Storage auto-expires after 24 hours.",
-      inputZod: InputSchema10,
-      inputJsonSchema: InputJsonSchema10,
+      inputZod: InputSchema11,
+      inputJsonSchema: InputJsonSchema11,
       allowedModels: [],
       defaultModel: "",
-      run: run10
+      run: run11
     };
   }
 });
-async function run11(input, ctx) {
+async function run12(input, ctx) {
   const requested = input.model ?? DEFAULT_MODEL6;
   if (!ALLOWED4.includes(requested)) {
     return {
@@ -1050,7 +1312,7 @@ async function run11(input, ctx) {
     (urls) => urls.length === 0 ? `Job ${submitted.jobId} completed but returned no assets.` : `Audio ready. Asset: ${urls[0]}`
   );
 }
-var ALLOWED4, DEFAULT_MODEL6, InputSchema11, InputJsonSchema11, audioSkill;
+var ALLOWED4, DEFAULT_MODEL6, InputSchema12, InputJsonSchema12, audioSkill;
 var init_audio = __esm({
   "src/skills/audio.ts"() {
     init_poll();
@@ -1058,7 +1320,7 @@ var init_audio = __esm({
     init_types();
     ALLOWED4 = ["seed-audio", "text-to-speech"];
     DEFAULT_MODEL6 = "seed-audio";
-    InputSchema11 = z.object({
+    InputSchema12 = z.object({
       prompt: z.string().min(1).describe(
         "A script for speech, or a scene description for cinematic audio (dialogue + effects + music). Reference uploaded clips as @Audio1..@Audio3."
       ),
@@ -1069,7 +1331,7 @@ var init_audio = __esm({
       output_format: z.string().optional().describe("seed-audio: mp3 (default), wav, pcm, ogg_opus. text-to-speech: mp3, wav, flac, pcm."),
       speed: z.number().min(0.5).max(2).optional().describe("Speech speed multiplier (0.5\u20132).")
     });
-    InputJsonSchema11 = {
+    InputJsonSchema12 = {
       type: "object",
       properties: {
         prompt: { type: "string" },
@@ -1085,15 +1347,15 @@ var init_audio = __esm({
     audioSkill = {
       name: "pixeldojo:audio",
       description: "Generate audio: speech, multi-character dialogue, sound effects, and background music from one prompt (Seed Audio 1.0 by default; plain TTS available).",
-      inputZod: InputSchema11,
-      inputJsonSchema: InputJsonSchema11,
+      inputZod: InputSchema12,
+      inputJsonSchema: InputJsonSchema12,
       allowedModels: [...ALLOWED4],
       defaultModel: DEFAULT_MODEL6,
-      run: run11
+      run: run12
     };
   }
 });
-async function run12(input, ctx) {
+async function run13(input, ctx) {
   const res = await ctx.client.listLibrary({
     q: input.q,
     model: input.model,
@@ -1124,16 +1386,16 @@ ${lines.join("\n")}
 Pass any URL back into :generate, :edit, :character, or :upscale as a reference input.`
   };
 }
-var InputSchema12, InputJsonSchema12, librarySkill;
+var InputSchema13, InputJsonSchema13, librarySkill;
 var init_library = __esm({
   "src/skills/library.ts"() {
-    InputSchema12 = z.object({
+    InputSchema13 = z.object({
       q: z.string().optional().describe("Substring match against the prompt that produced the asset."),
       model: z.string().optional().describe("Restrict to a single model (apiId)."),
       modality: z.enum(["image", "video", "audio"]).optional().describe("Restrict to one output type."),
       limit: z.number().int().min(1).max(50).optional().describe("Max results (default 10).")
     });
-    InputJsonSchema12 = {
+    InputJsonSchema13 = {
       type: "object",
       properties: {
         q: { type: "string" },
@@ -1146,15 +1408,15 @@ var init_library = __esm({
     librarySkill = {
       name: "pixeldojo:library",
       description: "Search your media library: saved My Media items (permanent) and recent API/MCP generations (expire after 24h), newest first. Returns asset URLs to reuse as reference inputs. Keep a result forever with pixeldojo:save.",
-      inputZod: InputSchema12,
-      inputJsonSchema: InputJsonSchema12,
+      inputZod: InputSchema13,
+      inputJsonSchema: InputJsonSchema13,
       allowedModels: [],
       defaultModel: "",
-      run: run12
+      run: run13
     };
   }
 });
-async function run13(input, ctx) {
+async function run14(input, ctx) {
   const res = await ctx.client.saveMedia({
     url: input.url,
     prompt: input.prompt,
@@ -1168,17 +1430,17 @@ async function run13(input, ctx) {
 It will appear at https://pixeldojo.ai/my-images and in pixeldojo:library searches.`
   };
 }
-var InputSchema13, InputJsonSchema13, saveSkill;
+var InputSchema14, InputJsonSchema14, saveSkill;
 var init_save = __esm({
   "src/skills/save.ts"() {
-    InputSchema13 = z.object({
+    InputSchema14 = z.object({
       url: z.string().url().describe("The asset URL to save (image or video)."),
       prompt: z.string().optional().describe("The prompt that produced it \u2014 makes it findable in the library later."),
       tool_name: z.string().optional().describe("Tool/model slug that produced it."),
       modality: z.enum(["image", "video", "audio"]).optional(),
       job_id: z.string().optional().describe("The jobId that produced this asset, when known.")
     });
-    InputJsonSchema13 = {
+    InputJsonSchema14 = {
       type: "object",
       properties: {
         url: { type: "string", format: "uri" },
@@ -1192,11 +1454,11 @@ var init_save = __esm({
     saveSkill = {
       name: "pixeldojo:save",
       description: "Save a generated image or video into the user's My Media library permanently. Generation outputs otherwise expire after 24 hours \u2014 call this when the user wants to keep a result.",
-      inputZod: InputSchema13,
-      inputJsonSchema: InputJsonSchema13,
+      inputZod: InputSchema14,
+      inputJsonSchema: InputJsonSchema14,
       allowedModels: [],
       defaultModel: "",
-      run: run13
+      run: run14
     };
   }
 });
@@ -1285,7 +1547,24 @@ async function runWorkflow(input, ctx) {
         text: `Step ${i + 1} (${step.skill}) has invalid args: ${issue?.path.join(".") || "input"} ${issue?.message || "invalid"}.`
       };
     }
-    const result = await skill.run(parsed.data, ctx);
+    let result = await skill.run(parsed.data, ctx);
+    if (result.kind === "handoff") {
+      const handoffJobId = result.jobId;
+      const polled = await pollJobUntilDoneOrHandoff(
+        handoffJobId,
+        (id) => ctx.client.getJob(id),
+        ctx.buildStatusUrl,
+        {
+          syncBudgetMs: WORKFLOW_STEP_EXTRA_BUDGET_MS,
+          now: ctx.pollOverrides?.now,
+          sleep: ctx.pollOverrides?.sleep
+        }
+      );
+      result = pollResultToSkillResult(
+        polled,
+        (urls) => urls.length === 0 ? `Job ${handoffJobId} completed but returned no assets.` : `Done. Asset: ${urls[0]}`
+      );
+    }
     const url = firstUrl(result);
     if (url) {
       prevUrl = url;
@@ -1295,6 +1574,11 @@ async function runWorkflow(input, ctx) {
         kind: "text",
         text: `Workflow "${workflow.name}" stopped at step ${i + 1} (${step.skill}): ${result.text}`
       };
+    } else if (result.kind === "handoff") {
+      return {
+        kind: "text",
+        text: `Workflow "${workflow.name}" is still on step ${i + 1} (${step.skill}) after an extended wait. Poll with pixeldojo:status { jobId: "${result.jobId}" }, then re-run the workflow once it completes.`
+      };
     } else {
       outputs.push(`step ${i + 1} (${step.skill}): done (no asset URL)`);
     }
@@ -1305,9 +1589,12 @@ async function runWorkflow(input, ctx) {
 ${outputs.join("\n")}`
   };
 }
-var StepSchema, SaveInputSchema, RunInputSchema, ListInputSchema, stepJson, saveWorkflowSkill, runWorkflowSkill, listWorkflowsSkill;
+var WORKFLOW_STEP_EXTRA_BUDGET_MS, StepSchema, SaveInputSchema, RunInputSchema, ListInputSchema, stepJson, saveWorkflowSkill, runWorkflowSkill, listWorkflowsSkill;
 var init_workflows = __esm({
   "src/skills/workflows.ts"() {
+    init_poll();
+    init_types();
+    WORKFLOW_STEP_EXTRA_BUDGET_MS = 6e4;
     StepSchema = z.object({
       skill: z.string().regex(/^pixeldojo:(generate|edit|character|storyboard|upscale|audio|upload|from_url)$/, {
         message: "Runnable skills only: generate, edit, character, storyboard, upscale, audio, upload, from_url."
@@ -1383,6 +1670,7 @@ var init_workflows = __esm({
 var skills_exports = {};
 __export(skills_exports, {
   ALL_SKILLS: () => ALL_SKILLS,
+  adSkill: () => adSkill,
   audioSkill: () => audioSkill,
   campaignSkill: () => campaignSkill,
   campaignStatusSkill: () => campaignStatusSkill,
@@ -1414,6 +1702,7 @@ var init_skills = __esm({
     init_upscale();
     init_status();
     init_from_url();
+    init_ad();
     init_campaign();
     init_campaign_status();
     init_upload();
@@ -1436,6 +1725,7 @@ var init_skills = __esm({
       statusSkill,
       fromUrlSkill,
       uploadSkill,
+      adSkill,
       campaignSkill,
       campaignStatusSkill
     ];
@@ -1492,7 +1782,7 @@ function makeRedactedLogger(target = console) {
 }
 
 // src/server.ts
-var PACKAGE_VERSION = "0.5.1" ;
+var PACKAGE_VERSION = "0.8.0" ;
 function buildServer(options) {
   if (!options.apiKey) {
     throw new Error(