@vibeframe/cli 0.109.1 → 0.110.0

package/dist/index.js

@@ -29613,6 +29613,7 @@ function isComposerProvider(value) {
   return value === "claude" || value === "openai" || value === "gemini";
 }
 function resolveComposer(explicit) {
+  loadEnv();
   if (explicit) {
     const key = readKey(explicit);
     if (!key) {
@@ -29637,6 +29638,7 @@ var ENV_VAR, PROVIDER_LABEL, ComposerResolveError;
 var init_composer_resolve = __esm({
   "src/commands/_shared/composer-resolve.ts"() {
     "use strict";
+    init_api_key();
     ENV_VAR = {
       claude: "ANTHROPIC_API_KEY",
       openai: "OPENAI_API_KEY",
@@ -67836,6 +67838,82 @@ var init_long_poll = __esm({
   }
 });
 
+// src/tools/_shared/elicit.ts
+function planBuildElicitation(args) {
+  const assetsWillRun = args.stage === void 0 || args.stage === "all" || args.stage === "assets";
+  if (!assetsWillRun) return null;
+  const properties = {};
+  const required = [];
+  const narrationUndecided = !args.skipNarration && (args.ttsProvider === void 0 || args.ttsProvider === "auto");
+  if (narrationUndecided) {
+    properties.narration = {
+      type: "string",
+      title: "Narration voice-over",
+      description: "Which TTS engine narrates the video?",
+      enum: [...NARRATION_CHOICES],
+      enumNames: [
+        "Kokoro \u2014 free, runs locally on this machine",
+        "ElevenLabs \u2014 cloud TTS, uses your ElevenLabs credits"
+      ]
+    };
+    required.push("narration");
+  }
+  if (args.skipBackdrop === void 0 && args.imageProvider === void 0) {
+    properties.backdrop_images = {
+      type: "string",
+      title: "Backdrop images",
+      description: "Generate AI background images for each scene?",
+      enum: [...BACKDROP_CHOICES],
+      enumNames: [
+        "Skip \u2014 free, pure typographic scenes",
+        "OpenAI \u2014 paid image generation (roughly $1-3 per video)"
+      ]
+    };
+    required.push("backdrop_images");
+  }
+  if (args.maxCostUsd === void 0) {
+    properties.max_cost_usd = {
+      type: "number",
+      title: "Max provider spend (USD)",
+      description: "Abort before any paid provider call when the estimate exceeds this cap. Leave empty for no cap.",
+      minimum: 0
+    };
+  }
+  if (Object.keys(properties).length === 0) return null;
+  return {
+    message: "VibeFrame is about to build this video. Choose how the assets are generated:",
+    requestedSchema: { type: "object", properties, required }
+  };
+}
+function applyElicitationAnswers(args, content) {
+  if (!content) return args;
+  const next = { ...args };
+  const narration = content.narration;
+  if (typeof narration === "string" && NARRATION_CHOICES.includes(narration)) {
+    next.ttsProvider = narration;
+  }
+  const backdrop = content.backdrop_images;
+  if (backdrop === "skip") {
+    next.skipBackdrop = true;
+  } else if (backdrop === "openai") {
+    next.skipBackdrop = false;
+    next.imageProvider = "openai";
+  }
+  const cap = content.max_cost_usd;
+  if (typeof cap === "number" && Number.isFinite(cap) && cap >= 0) {
+    next.maxCostUsd = cap;
+  }
+  return next;
+}
+var NARRATION_CHOICES, BACKDROP_CHOICES;
+var init_elicit = __esm({
+  "src/tools/_shared/elicit.ts"() {
+    "use strict";
+    NARRATION_CHOICES = ["kokoro", "elevenlabs"];
+    BACKDROP_CHOICES = ["skip", "openai"];
+  }
+});
+
 // src/tools/manifest/scene.ts
 import { resolve as resolve75, relative as relative12 } from "node:path";
 function summariseLint(result) {
@@ -67974,6 +68052,7 @@ var init_scene2 = __esm({
     init_zod();
     init_define_tool();
     init_long_poll();
+    init_elicit();
     init_visual_styles();
     init_scene_project();
     init_scene();
@@ -68307,9 +68386,32 @@ var init_scene2 = __esm({
       name: "build",
       category: "scene",
       cost: "high",
-      description: "v0.60 one-shot orchestrator: read STORYBOARD.md per-beat YAML cues (narration / backdrop / duration), dispatch TTS + image generation per beat, compose scene HTML via the compose-scenes-with-skills pipeline, then render to MP4. Use this instead of chaining init + scene_add + render manually. Caches by SHA256 of (DESIGN.md + cue body) so re-runs are idempotent and cheap. Long runs: over MCP, calls exceeding ~45s return immediately with { promoted: true, jobId, status: 'running' } \u2014 poll status_job every 15-30s until completed/failed instead of re-invoking build. Emits MCP progress notifications when the client supplies a progressToken.",
+      description: "v0.60 one-shot orchestrator: read STORYBOARD.md per-beat YAML cues (narration / backdrop / duration), dispatch TTS + image generation per beat, compose scene HTML via the compose-scenes-with-skills pipeline, then render to MP4. Use this instead of chaining init + scene_add + render manually. Caches by SHA256 of (DESIGN.md + cue body) so re-runs are idempotent and cheap. Long runs: over MCP, calls exceeding ~45s return immediately with { promoted: true, jobId, status: 'running' } \u2014 poll status_job every 15-30s until completed/failed instead of re-invoking build. Emits MCP progress notifications when the client supplies a progressToken. When the MCP host supports elicitation, unspecified asset choices (narration provider, backdrop images, cost cap) are confirmed with the user via a form before the build starts.",
       schema: sceneBuildSchema,
       async execute(args, ctx) {
+        let warnings;
+        if (ctx.surface === "mcp" && ctx.elicit) {
+          const form = planBuildElicitation(args);
+          if (form) {
+            const outcome = await ctx.elicit(form).catch(() => null);
+            if (outcome === null) {
+              warnings = ["Asset-choice elicitation failed or timed out; proceeding with defaults."];
+            } else if (outcome.action === "accept") {
+              args = applyElicitationAnswers(args, outcome.content);
+            } else {
+              return {
+                success: false,
+                error: "Build cancelled \u2014 the user declined the asset choices.",
+                data: {
+                  cancelled: true,
+                  retryWith: [
+                    `build { projectDir: "${args.projectDir ?? "."}", ttsProvider: "kokoro", skipBackdrop: true }`
+                  ]
+                }
+              };
+            }
+          }
+        }
         const projectDir = args.projectDir ? resolve75(ctx.workingDirectory, args.projectDir) : ctx.workingDirectory;
         const runBuild = (report) => executeSceneBuild({
           projectDir,
@@ -68337,7 +68439,9 @@ var init_scene2 = __esm({
             ctx.onProgress?.({ message: update.message ?? update.stage });
             report(update);
           }
-        }).then(mapBuildResultToToolResult);
+        }).then(mapBuildResultToToolResult).then(
+          (result) => warnings === void 0 ? result : { ...result, data: { ...result.data, elicitationWarnings: warnings } }
+        );
         if (ctx.surface !== "mcp") {
           return runBuild(() => void 0);
         }
@@ -72909,7 +73013,7 @@ init_source();
 // package.json
 var package_default = {
   name: "@vibeframe/cli",
-  version: "0.109.1",
+  version: "0.110.0",
   description: "VibeFrame CLI - agentic video workflow runtime",
   private: false,
   license: "MIT",

package/dist/tools/_shared/elicit.d.ts

@@ -0,0 +1,31 @@
+import type { ElicitForm } from "../adapters/mcp.js";
+/**
+ * @module tools/_shared/elicit
+ *
+ * Pre-flight elicitation for the `build` tool: when an MCP client supports
+ * elicitation, ask the user the choices the host agent would otherwise pick
+ * silently (narration provider, paid backdrop generation, cost ceiling).
+ * Pure functions — the MCP wiring lives in the tool's execute and the
+ * server's makeElicitFn; CLI/Agent surfaces never reach this code path.
+ */
+/** The subset of build args that drive the question set. */
+export interface BuildChoiceArgs {
+    stage?: "assets" | "compose" | "sync" | "render" | "all";
+    ttsProvider?: "auto" | "elevenlabs" | "kokoro";
+    skipNarration?: boolean;
+    skipBackdrop?: boolean;
+    imageProvider?: string;
+    maxCostUsd?: number;
+}
+/**
+ * Builds the elicitation form for a build call, asking ONLY about choices the
+ * caller left unspecified. Returns null when there is nothing to ask — every
+ * choice explicit, or the requested stage doesn't generate assets.
+ */
+export declare function planBuildElicitation(args: BuildChoiceArgs): ElicitForm | null;
+/**
+ * Maps accepted form answers back onto build args. Unknown or invalid values
+ * are ignored so a misbehaving client can never produce a worse state than
+ * the defaults the user would have gotten without elicitation.
+ */
+export declare function applyElicitationAnswers<T extends BuildChoiceArgs>(args: T, content: Record<string, string | number | boolean | string[]> | undefined): T;

package/dist/tools/_shared/elicit.test.d.ts

@@ -0,0 +1 @@
+export {};

package/dist/tools/adapters/mcp.d.ts

@@ -22,6 +22,25 @@ export interface McpTool {
     description: string;
     inputSchema: McpInputSchema;
 }
+/**
+ * Structural mirror of the MCP elicitation form request/result (the CLI
+ * package deliberately has no dependency on @modelcontextprotocol/sdk).
+ * `requestedSchema.properties` values follow the spec's flat primitive
+ * schema: string (optionally with enum/enumNames), number, boolean.
+ */
+export interface ElicitForm {
+    message: string;
+    requestedSchema: {
+        type: "object";
+        properties: Record<string, unknown>;
+        required?: string[];
+    };
+}
+export interface ElicitOutcome {
+    action: "accept" | "decline" | "cancel";
+    content?: Record<string, string | number | boolean | string[]>;
+}
+export type ElicitFn = (form: ElicitForm) => Promise<ElicitOutcome>;
 /** Per-call extras forwarded by the MCP server request handler. */
 export interface McpCallExtra {
     /**
@@ -34,6 +53,12 @@ export interface McpCallExtra {
         total?: number;
         message?: string;
     }) => void;
+    /**
+     * Sends an `elicitation/create` form to the client and resolves with the
+     * user's answer. Only present when the connected client advertises the
+     * elicitation capability.
+     */
+    elicit?: ElicitFn;
 }
 export type McpDispatcher = (name: string, args: Record<string, unknown>, extra?: McpCallExtra) => Promise<{
     content: Array<{

package/dist/tools/adapters/mcp.js

@@ -245,7 +245,8 @@ function buildMcpDispatcher(manifest) {
         () => tool.execute(parsed.data, {
           workingDirectory: process.cwd(),
           surface: "mcp",
-          ...extra?.onProgress ? { onProgress: sink.send } : {}
+          ...extra?.onProgress ? { onProgress: sink.send } : {},
+          ...extra?.elicit ? { elicit: extra.elicit } : {}
         })
       );
       const text = result.success ? JSON.stringify({ success: true, ...result.data }) : result.data ? JSON.stringify({

package/dist/tools/define-tool.d.ts

@@ -49,6 +49,23 @@ export interface ExecuteContext {
         total?: number;
         message?: string;
     }) => void;
+    /**
+     * Optional MCP elicitation channel. Present only when the connected MCP
+     * client advertises the elicitation capability; CLI/Agent surfaces leave
+     * it undefined, so CLI behavior never changes. See adapters/mcp.ts for
+     * the structural types.
+     */
+    elicit?: (form: {
+        message: string;
+        requestedSchema: {
+            type: "object";
+            properties: Record<string, unknown>;
+            required?: string[];
+        };
+    }) => Promise<{
+        action: "accept" | "decline" | "cancel";
+        content?: Record<string, string | number | boolean | string[]>;
+    }>;
 }
 export interface ToolExecuteResult {
     success: boolean;

package/dist/tools/manifest/index.js

@@ -30319,6 +30319,7 @@ function isComposerProvider(value) {
   return value === "claude" || value === "openai" || value === "gemini";
 }
 function resolveComposer(explicit) {
+  loadEnv();
   if (explicit) {
     const key = readKey(explicit);
     if (!key) {
@@ -30343,6 +30344,7 @@ var ENV_VAR, PROVIDER_LABEL, ComposerResolveError;
 var init_composer_resolve = __esm({
   "src/commands/_shared/composer-resolve.ts"() {
     "use strict";
+    init_api_key();
     ENV_VAR = {
       claude: "ANTHROPIC_API_KEY",
       openai: "OPENAI_API_KEY",
@@ -51298,6 +51300,76 @@ function pickOutputPath(data) {
   return typeof candidate === "string" ? candidate : void 0;
 }
 
+// src/tools/_shared/elicit.ts
+var NARRATION_CHOICES = ["kokoro", "elevenlabs"];
+var BACKDROP_CHOICES = ["skip", "openai"];
+function planBuildElicitation(args) {
+  const assetsWillRun = args.stage === void 0 || args.stage === "all" || args.stage === "assets";
+  if (!assetsWillRun) return null;
+  const properties = {};
+  const required = [];
+  const narrationUndecided = !args.skipNarration && (args.ttsProvider === void 0 || args.ttsProvider === "auto");
+  if (narrationUndecided) {
+    properties.narration = {
+      type: "string",
+      title: "Narration voice-over",
+      description: "Which TTS engine narrates the video?",
+      enum: [...NARRATION_CHOICES],
+      enumNames: [
+        "Kokoro \u2014 free, runs locally on this machine",
+        "ElevenLabs \u2014 cloud TTS, uses your ElevenLabs credits"
+      ]
+    };
+    required.push("narration");
+  }
+  if (args.skipBackdrop === void 0 && args.imageProvider === void 0) {
+    properties.backdrop_images = {
+      type: "string",
+      title: "Backdrop images",
+      description: "Generate AI background images for each scene?",
+      enum: [...BACKDROP_CHOICES],
+      enumNames: [
+        "Skip \u2014 free, pure typographic scenes",
+        "OpenAI \u2014 paid image generation (roughly $1-3 per video)"
+      ]
+    };
+    required.push("backdrop_images");
+  }
+  if (args.maxCostUsd === void 0) {
+    properties.max_cost_usd = {
+      type: "number",
+      title: "Max provider spend (USD)",
+      description: "Abort before any paid provider call when the estimate exceeds this cap. Leave empty for no cap.",
+      minimum: 0
+    };
+  }
+  if (Object.keys(properties).length === 0) return null;
+  return {
+    message: "VibeFrame is about to build this video. Choose how the assets are generated:",
+    requestedSchema: { type: "object", properties, required }
+  };
+}
+function applyElicitationAnswers(args, content) {
+  if (!content) return args;
+  const next = { ...args };
+  const narration = content.narration;
+  if (typeof narration === "string" && NARRATION_CHOICES.includes(narration)) {
+    next.ttsProvider = narration;
+  }
+  const backdrop = content.backdrop_images;
+  if (backdrop === "skip") {
+    next.skipBackdrop = true;
+  } else if (backdrop === "openai") {
+    next.skipBackdrop = false;
+    next.imageProvider = "openai";
+  }
+  const cap = content.max_cost_usd;
+  if (typeof cap === "number" && Number.isFinite(cap) && cap >= 0) {
+    next.maxCostUsd = cap;
+  }
+  return next;
+}
+
 // src/commands/_shared/visual-styles.ts
 var STYLES = [
   {
@@ -53536,9 +53608,32 @@ var sceneBuildTool = defineTool({
   name: "build",
   category: "scene",
   cost: "high",
-  description: "v0.60 one-shot orchestrator: read STORYBOARD.md per-beat YAML cues (narration / backdrop / duration), dispatch TTS + image generation per beat, compose scene HTML via the compose-scenes-with-skills pipeline, then render to MP4. Use this instead of chaining init + scene_add + render manually. Caches by SHA256 of (DESIGN.md + cue body) so re-runs are idempotent and cheap. Long runs: over MCP, calls exceeding ~45s return immediately with { promoted: true, jobId, status: 'running' } \u2014 poll status_job every 15-30s until completed/failed instead of re-invoking build. Emits MCP progress notifications when the client supplies a progressToken.",
+  description: "v0.60 one-shot orchestrator: read STORYBOARD.md per-beat YAML cues (narration / backdrop / duration), dispatch TTS + image generation per beat, compose scene HTML via the compose-scenes-with-skills pipeline, then render to MP4. Use this instead of chaining init + scene_add + render manually. Caches by SHA256 of (DESIGN.md + cue body) so re-runs are idempotent and cheap. Long runs: over MCP, calls exceeding ~45s return immediately with { promoted: true, jobId, status: 'running' } \u2014 poll status_job every 15-30s until completed/failed instead of re-invoking build. Emits MCP progress notifications when the client supplies a progressToken. When the MCP host supports elicitation, unspecified asset choices (narration provider, backdrop images, cost cap) are confirmed with the user via a form before the build starts.",
   schema: sceneBuildSchema,
   async execute(args, ctx) {
+    let warnings;
+    if (ctx.surface === "mcp" && ctx.elicit) {
+      const form = planBuildElicitation(args);
+      if (form) {
+        const outcome = await ctx.elicit(form).catch(() => null);
+        if (outcome === null) {
+          warnings = ["Asset-choice elicitation failed or timed out; proceeding with defaults."];
+        } else if (outcome.action === "accept") {
+          args = applyElicitationAnswers(args, outcome.content);
+        } else {
+          return {
+            success: false,
+            error: "Build cancelled \u2014 the user declined the asset choices.",
+            data: {
+              cancelled: true,
+              retryWith: [
+                `build { projectDir: "${args.projectDir ?? "."}", ttsProvider: "kokoro", skipBackdrop: true }`
+              ]
+            }
+          };
+        }
+      }
+    }
     const projectDir = args.projectDir ? resolve26(ctx.workingDirectory, args.projectDir) : ctx.workingDirectory;
     const runBuild = (report) => executeSceneBuild({
       projectDir,
@@ -53566,7 +53661,9 @@ var sceneBuildTool = defineTool({
         ctx.onProgress?.({ message: update.message ?? update.stage });
         report(update);
       }
-    }).then(mapBuildResultToToolResult);
+    }).then(mapBuildResultToToolResult).then(
+      (result) => warnings === void 0 ? result : { ...result, data: { ...result.data, elicitationWarnings: warnings } }
+    );
     if (ctx.surface !== "mcp") {
       return runBuild(() => void 0);
     }

package/dist/tools/manifest/scene.d.ts

@@ -167,9 +167,9 @@ export declare const sceneBuildTool: import("../define-tool.js").ToolDefinition<
     skipBackdrop?: boolean | undefined;
     skipRender?: boolean | undefined;
     imageProvider?: "openai" | undefined;
+    ttsProvider?: "elevenlabs" | "auto" | "kokoro" | undefined;
     skipVideo?: boolean | undefined;
     skipMusic?: boolean | undefined;
-    ttsProvider?: "elevenlabs" | "auto" | "kokoro" | undefined;
     videoProvider?: "grok" | "seedance" | "kling" | "runway" | "veo" | undefined;
     musicProvider?: "elevenlabs" | "replicate" | undefined;
     imageQuality?: "standard" | "hd" | undefined;
@@ -188,9 +188,9 @@ export declare const sceneBuildTool: import("../define-tool.js").ToolDefinition<
     skipBackdrop?: boolean | undefined;
     skipRender?: boolean | undefined;
     imageProvider?: "openai" | undefined;
+    ttsProvider?: "elevenlabs" | "auto" | "kokoro" | undefined;
     skipVideo?: boolean | undefined;
     skipMusic?: boolean | undefined;
-    ttsProvider?: "elevenlabs" | "auto" | "kokoro" | undefined;
     videoProvider?: "grok" | "seedance" | "kling" | "runway" | "veo" | undefined;
     musicProvider?: "elevenlabs" | "replicate" | undefined;
     imageQuality?: "standard" | "hd" | undefined;

package/dist/tools/manifest/storyboard.d.ts

@@ -107,9 +107,9 @@ export declare const planTool: import("../define-tool.js").ToolDefinition<z.ZodO
     skipNarration?: boolean | undefined;
     skipBackdrop?: boolean | undefined;
     imageProvider?: string | undefined;
+    ttsProvider?: "elevenlabs" | "auto" | "kokoro" | undefined;
     skipVideo?: boolean | undefined;
     skipMusic?: boolean | undefined;
-    ttsProvider?: "elevenlabs" | "auto" | "kokoro" | undefined;
     videoProvider?: "grok" | "seedance" | "kling" | "runway" | "veo" | undefined;
     musicProvider?: "elevenlabs" | "replicate" | undefined;
     imageQuality?: "standard" | "hd" | undefined;
@@ -125,9 +125,9 @@ export declare const planTool: import("../define-tool.js").ToolDefinition<z.ZodO
     skipNarration?: boolean | undefined;
     skipBackdrop?: boolean | undefined;
     imageProvider?: string | undefined;
+    ttsProvider?: "elevenlabs" | "auto" | "kokoro" | undefined;
     skipVideo?: boolean | undefined;
     skipMusic?: boolean | undefined;
-    ttsProvider?: "elevenlabs" | "auto" | "kokoro" | undefined;
     videoProvider?: "grok" | "seedance" | "kling" | "runway" | "veo" | undefined;
     musicProvider?: "elevenlabs" | "replicate" | undefined;
     imageQuality?: "standard" | "hd" | undefined;

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@vibeframe/cli",
-  "version": "0.109.1",
+  "version": "0.110.0",
   "description": "VibeFrame CLI - agentic video workflow runtime",
   "private": false,
   "license": "MIT",
@@ -83,8 +83,8 @@
     "tsx": "^4.21.0",
     "typescript": "^5.3.3",
     "vitest": "^1.2.2",
-    "@vibeframe/ai-providers": "0.109.1",
-    "@vibeframe/core": "0.109.1"
+    "@vibeframe/ai-providers": "0.110.0",
+    "@vibeframe/core": "0.110.0"
   },
   "engines": {
     "node": ">=20"