@phuetz/code-buddy 1.6.1 → 2.1.0

package/dist/tools/registry/multimodal-tools.js

@@ -7,9 +7,16 @@
  * Each adapter lazy-loads the underlying tool and dispatches based on
  * the `operation` parameter from the OpenAI function-calling schema.
  */
+import fs from 'fs/promises';
+import path from 'path';
 import { synthesizeTextToSpeech, } from '../text-to-speech-tool.js';
-import { generateImage, generateVideo, } from '../media-generation-tool.js';
+import { editImage, generateImage, generateVideo, } from '../media-generation-tool.js';
 import { analyzeVideoWithModel, } from '../video-analysis-tool.js';
+import { understandVideo, isUnderstandOk, } from '../video/video-understanding.js';
+import { assembleFilm, } from '../video/film-assemble.js';
+import { gpuMediaWorkerFromEnv, parseAvatarVideoPayload, } from '../gpu-media-worker.js';
+import { scheduleAvatarDelivery } from '../gpu-avatar-delivery.js';
+import { createVideoStudioTools } from './video-studio-tools.js';
 // ============================================================================
 // Lazy-loaded tool instances
 // ============================================================================
@@ -221,8 +228,8 @@ export class TextToSpeechTool {
                     },
                     provider: {
                         type: 'string',
-                        enum: ['auto', 'system', 'edge-tts', 'espeak', 'say', 'audioreader'],
-                        description: 'TTS provider. auto detects a local provider; system uses Windows SAPI.',
+                        enum: ['auto', 'system', 'edge-tts', 'espeak', 'say', 'audioreader', 'piper'],
+                        description: 'TTS provider. auto detects a local provider (piper first when CODEBUDDY_TTS_VOICE is set); system uses Windows SAPI.',
                     },
                     voice: {
                         type: 'string',
@@ -350,6 +357,121 @@ export class ImageGenerateTool {
     isAvailable() { return true; }
 }
 // ============================================================================
+// ImageEditTool
+// ============================================================================
+export class ImageEditTool {
+    options;
+    name = 'image_edit';
+    description = 'Create a new edited image from a workspace image, with an optional explicit PNG mask and normalized marked regions.';
+    constructor(options = {}) {
+        this.options = options;
+    }
+    async execute(input, context) {
+        try {
+            const cwd = path.resolve(context?.cwd ?? this.options.rootDir ?? process.cwd());
+            const imageUrl = await loadBoundedWorkspaceImage(requiredString(input, 'image_path'), cwd, false);
+            const maskPath = optionalString(input, 'mask_path');
+            const maskUrl = maskPath ? await loadBoundedWorkspaceImage(maskPath, cwd, true) : undefined;
+            const result = await editImage({
+                prompt: requiredString(input, 'prompt'),
+                imageUrl,
+                ...(maskUrl ? { maskUrl } : {}),
+                ...(Array.isArray(input.selections) ? { selections: parseImageEditSelections(input.selections) } : {}),
+            }, {
+                ...this.options,
+                rootDir: this.options.rootDir ?? cwd,
+                signal: this.options.signal ?? context?.abortSignal,
+            });
+            return { success: true, output: JSON.stringify(result, null, 2), data: result };
+        }
+        catch (error) {
+            return { success: false, error: error instanceof Error ? error.message : String(error) };
+        }
+    }
+    getSchema() {
+        return {
+            name: this.name,
+            description: this.description,
+            parameters: {
+                type: 'object',
+                properties: {
+                    prompt: { type: 'string', description: 'The exact visual change to make.' },
+                    image_path: { type: 'string', description: 'Source PNG/JPEG/WebP inside the current workspace. The original is never overwritten.' },
+                    mask_path: { type: 'string', description: 'Optional PNG alpha mask inside the current workspace.' },
+                    selections: {
+                        type: 'array',
+                        description: 'Optional normalized edit rectangles, each in the 0..1 range.',
+                        items: {
+                            type: 'object',
+                            properties: {
+                                x: { type: 'number' },
+                                y: { type: 'number' },
+                                width: { type: 'number' },
+                                height: { type: 'number' },
+                            },
+                            required: ['x', 'y', 'width', 'height'],
+                        },
+                    },
+                },
+                required: ['prompt', 'image_path'],
+            },
+        };
+    }
+    validate(input) {
+        if (!input || typeof input !== 'object')
+            return { valid: false, errors: ['Input must be an object'] };
+        const data = input;
+        const errors = [];
+        if (typeof data.prompt !== 'string' || !data.prompt.trim())
+            errors.push('prompt is required');
+        if (typeof data.image_path !== 'string' || !data.image_path.trim())
+            errors.push('image_path is required');
+        return { valid: errors.length === 0, ...(errors.length ? { errors } : {}) };
+    }
+    getMetadata() {
+        return {
+            name: this.name,
+            description: this.description,
+            category: 'media',
+            keywords: ['image', 'edit', 'inpaint', 'mask', 'mark', 'design', 'openai', 'xai', 'comfyui'],
+            priority: 8,
+            requiresConfirmation: true,
+            modifiesFiles: true,
+            makesNetworkRequests: true,
+        };
+    }
+    isAvailable() { return true; }
+}
+async function loadBoundedWorkspaceImage(value, workspace, requirePng) {
+    const resolved = path.resolve(workspace, value);
+    const [root, realPath] = await Promise.all([fs.realpath(workspace), fs.realpath(resolved)]);
+    if (realPath !== root && !realPath.startsWith(`${root}${path.sep}`)) {
+        throw new Error('image path escapes the current workspace');
+    }
+    const extension = path.extname(realPath).toLowerCase();
+    const mime = extension === '.png' ? 'image/png'
+        : extension === '.jpg' || extension === '.jpeg' ? 'image/jpeg'
+            : extension === '.webp' ? 'image/webp' : undefined;
+    if (!mime || (requirePng && mime !== 'image/png')) {
+        throw new Error(requirePng ? 'mask_path must be a PNG file' : 'image_path must be PNG, JPEG, or WebP');
+    }
+    const metadata = await fs.stat(realPath);
+    if (!metadata.isFile() || metadata.size <= 0 || metadata.size > 50 * 1024 * 1024) {
+        throw new Error('image must be a file smaller than 50 MB');
+    }
+    return `data:${mime};base64,${(await fs.readFile(realPath)).toString('base64')}`;
+}
+function parseImageEditSelections(input) {
+    return input.slice(0, 20).flatMap((value) => {
+        if (!value || typeof value !== 'object')
+            return [];
+        const item = value;
+        if (![item.x, item.y, item.width, item.height].every((entry) => typeof entry === 'number' && Number.isFinite(entry)))
+            return [];
+        return [{ x: item.x, y: item.y, width: item.width, height: item.height }];
+    });
+}
+// ============================================================================
 // VideoAnalyzeTool
 // ============================================================================
 export class VideoAnalyzeTool {
@@ -425,6 +547,206 @@ export class VideoAnalyzeTool {
     isAvailable() { return true; }
 }
 // ============================================================================
+// UnderstandVideoTool
+// ============================================================================
+export class UnderstandVideoTool {
+    name = 'understand_video';
+    description = 'Understand a video (YouTube/URL/local file): timestamped transcript, research card, unverified experiment backlog, and optional visual analysis. Local-first and $0.';
+    async execute(input, context) {
+        try {
+            const result = await understandVideo({
+                source: requiredString(input, 'source'),
+                question: optionalString(input, 'question'),
+                language: optionalString(input, 'language'),
+                ...(optionalBoolean(input.visual) !== undefined ? { visual: optionalBoolean(input.visual) } : {}),
+                ...(optionalBoolean(input.ocr) !== undefined ? { ocr: optionalBoolean(input.ocr) } : {}),
+                ...(optionalBoolean(input.cloud) !== undefined ? { cloud: optionalBoolean(input.cloud) } : {}),
+            }, {
+                ...(context?.cwd ? { cwd: context.cwd } : {}),
+            });
+            if (!isUnderstandOk(result)) {
+                return { success: false, error: result.error };
+            }
+            return {
+                success: true,
+                output: result.output,
+                data: {
+                    segments: result.segments,
+                    transcriptPath: result.transcriptPath,
+                    ...(result.researchCardPath ? { researchCardPath: result.researchCardPath } : {}),
+                    ...(result.experimentBacklogPath ? { experimentBacklogPath: result.experimentBacklogPath } : {}),
+                    source: result.source,
+                    method: result.method,
+                    ...(result.cloud ? { cloud: result.cloud } : {}),
+                },
+            };
+        }
+        catch (error) {
+            return {
+                success: false,
+                error: error instanceof Error ? error.message : String(error),
+            };
+        }
+    }
+    getSchema() {
+        return {
+            name: this.name,
+            description: this.description,
+            parameters: {
+                type: 'object',
+                properties: {
+                    source: {
+                        type: 'string',
+                        description: 'YouTube URL, direct media URL, or local video/audio file path.',
+                    },
+                    question: {
+                        type: 'string',
+                        description: 'Optional question to answer about the video (recorded with the transcript).',
+                    },
+                    language: {
+                        type: 'string',
+                        description: "Optional preferred language code (e.g. 'en', 'fr').",
+                    },
+                    visual: {
+                        type: 'boolean',
+                        description: 'Also analyze what is SHOWN on screen (frames → local vision model). Default false.',
+                    },
+                    ocr: {
+                        type: 'boolean',
+                        description: 'With visual:true, also OCR each keyframe (best for on-screen code). Default false.',
+                    },
+                    cloud: {
+                        type: 'boolean',
+                        description: 'OPT-IN: also send the video/URL to a cloud model (Gemini) for a joint audio+visual, timestamped answer. Sends data to Google — public/non-sensitive videos only. Needs GEMINI_API_KEY; degrades to the local transcript on any failure. Default false.',
+                    },
+                },
+                required: ['source'],
+            },
+        };
+    }
+    validate(input) {
+        if (typeof input !== 'object' || input === null)
+            return { valid: false, errors: ['Input must be an object'] };
+        const data = input;
+        if (typeof data.source !== 'string' || !data.source.trim())
+            return { valid: false, errors: ['source is required'] };
+        return { valid: true };
+    }
+    getMetadata() {
+        return {
+            name: this.name,
+            description: this.description,
+            category: 'media',
+            keywords: ['video', 'youtube', 'transcribe', 'transcript', 'captions', 'subtitles', 'summarize', 'watch', 'visual', 'screencast', 'frames', 'ocr', 'on-screen', 'cloud', 'gemini', 'research', 'experiment', 'backlog', 'innovation'],
+            priority: 8,
+            modifiesFiles: true,
+            makesNetworkRequests: true,
+        };
+    }
+    isAvailable() { return true; }
+}
+// ============================================================================
+// GpuMediaJobTool — isolated GPU node jobs (PanoWorld / LongCat)
+// ============================================================================
+export class GpuMediaJobTool {
+    name = 'gpu_media_job';
+    description = 'Inspect, submit, monitor, or cancel isolated GPU media jobs on the configured GPU node worker. Supports bounded PanoWorld reconstruction and asynchronous LongCat avatar video rendering.';
+    async execute(input) {
+        try {
+            const client = gpuMediaWorkerFromEnv();
+            const operation = requiredString(input, 'operation');
+            let result;
+            switch (operation) {
+                case 'capabilities':
+                    result = await client.capabilities();
+                    break;
+                case 'submit': {
+                    const kind = requiredString(input, 'job_kind');
+                    if (kind !== 'panoworld_reconstruct' && kind !== 'avatar_video_render') {
+                        throw new Error('job_kind must be panoworld_reconstruct or avatar_video_render');
+                    }
+                    const avatarPayload = kind === 'avatar_video_render' ? parseAvatarVideoPayload(input.payload) : undefined;
+                    result = await client.submit(kind, avatarPayload ?? input.payload);
+                    if (avatarPayload?.channelTarget) {
+                        scheduleAvatarDelivery(client, result.id, avatarPayload);
+                    }
+                    break;
+                }
+                case 'status':
+                    result = await client.status(requiredString(input, 'job_id'));
+                    break;
+                case 'cancel':
+                    result = await client.cancel(requiredString(input, 'job_id'));
+                    break;
+                default:
+                    throw new Error('operation must be capabilities, submit, status, or cancel');
+            }
+            return { success: true, output: JSON.stringify(result, null, 2), data: result };
+        }
+        catch (error) {
+            return { success: false, error: error instanceof Error ? error.message : String(error) };
+        }
+    }
+    getSchema() {
+        return {
+            name: this.name,
+            description: this.description,
+            parameters: {
+                type: 'object',
+                properties: {
+                    operation: {
+                        type: 'string',
+                        enum: ['capabilities', 'submit', 'status', 'cancel'],
+                        description: 'Worker operation.',
+                    },
+                    job_kind: {
+                        type: 'string',
+                        enum: ['panoworld_reconstruct', 'avatar_video_render'],
+                        description: 'Required for submit.',
+                    },
+                    payload: {
+                        type: 'object',
+                        description: 'Validated job payload. PanoWorld accepts single-2048 (1 view) or multi-1024 (max 5); avatar rendering is restricted to 480p.',
+                    },
+                    job_id: { type: 'string', description: 'Required for status or cancel.' },
+                },
+                required: ['operation'],
+            },
+        };
+    }
+    validate(input) {
+        if (!input || typeof input !== 'object' || Array.isArray(input)) {
+            return { valid: false, errors: ['Input must be an object'] };
+        }
+        const data = input;
+        if (!['capabilities', 'submit', 'status', 'cancel'].includes(String(data.operation))) {
+            return { valid: false, errors: ['operation is invalid'] };
+        }
+        if (data.operation === 'submit' && (typeof data.job_kind !== 'string' || !data.payload)) {
+            return { valid: false, errors: ['job_kind and payload are required for submit'] };
+        }
+        if ((data.operation === 'status' || data.operation === 'cancel') && typeof data.job_id !== 'string') {
+            return { valid: false, errors: ['job_id is required'] };
+        }
+        return { valid: true };
+    }
+    getMetadata() {
+        return {
+            name: this.name,
+            description: this.description,
+            category: 'media',
+            keywords: ['gpuNode', 'gpu', 'panoworld', 'longcat', 'avatar', '3dgs', 'world model'],
+            priority: 8,
+            requiresConfirmation: true,
+            modifiesFiles: true,
+            makesNetworkRequests: true,
+        };
+    }
+    isAvailable() {
+        return Boolean(process.env.CODEBUDDY_GPU_WORKER_URL);
+    }
+}
+// ============================================================================
 // VideoGenerateTool
 // ============================================================================
 export class VideoGenerateTool {
@@ -508,6 +830,158 @@ export class VideoGenerateTool {
     isAvailable() { return true; }
 }
 // ============================================================================
+// MarkdownConvertTool — any document → Markdown for the model, via MarkItDown
+export class MarkdownConvertTool {
+    name = 'markdown_convert';
+    description = 'Convert a document (PDF, Word, Excel, PowerPoint, HTML, CSV, JSON, XML, ZIP, EPub, image, audio, or a URL) to structured Markdown for the model — keeps headings, lists and TABLES instead of flattening them. Requires MarkItDown (pip install markitdown[all]).';
+    getSchema() {
+        return {
+            name: this.name,
+            description: this.description,
+            parameters: {
+                type: 'object',
+                properties: {
+                    source: {
+                        type: 'string',
+                        description: 'Local file path, or an http(s)/YouTube URL. Handles PDF, DOCX, XLSX, PPTX, HTML, CSV, JSON, XML, ZIP, EPub, images (OCR) and audio (transcription).',
+                    },
+                    output_path: {
+                        type: 'string',
+                        description: 'Write the Markdown to this file instead of returning it inline. Use it for long documents.',
+                    },
+                    max_chars: {
+                        type: 'number',
+                        description: 'Inline truncation budget (default 60000 characters).',
+                    },
+                },
+                required: ['source'],
+            },
+        };
+    }
+    async execute(input) {
+        const { MarkdownConvertTool: Impl } = await import('../markdown-convert.js');
+        const source = optionalString(input, 'source') ?? optionalString(input, 'path') ?? '';
+        const outputPath = optionalString(input, 'output_path');
+        const maxChars = optionalNumber(input, 'max_chars');
+        return new Impl().convert({
+            source,
+            ...(outputPath ? { outputPath } : {}),
+            ...(maxChars !== undefined ? { maxChars } : {}),
+        });
+    }
+}
+// VideoStitchTool — chain clips into a longer film with transitions
+// ============================================================================
+/** Parse the `transitions`/`transition` inputs into what assembleFilm expects. */
+function parseStitchTransitions(input, defaultDuration) {
+    const arr = input.transitions;
+    if (Array.isArray(arr) && arr.length > 0) {
+        const specs = arr.map((t) => {
+            if (typeof t === 'string')
+                return { type: t.trim() || 'fade', duration: defaultDuration };
+            if (t && typeof t === 'object') {
+                const o = t;
+                const type = typeof o.type === 'string' && o.type.trim() ? o.type.trim() : 'fade';
+                const duration = typeof o.duration === 'number' && Number.isFinite(o.duration) ? o.duration : defaultDuration;
+                return { type, duration };
+            }
+            return { type: 'fade', duration: defaultDuration };
+        });
+        return specs;
+    }
+    return optionalString(input, 'transition');
+}
+export class VideoStitchTool {
+    options;
+    name = 'video_stitch';
+    description = 'Chain multiple video clips into one longer film with transitions, optional music (ducked) and voiceover. Requires ffmpeg.';
+    constructor(options = {}) {
+        this.options = options;
+    }
+    async execute(input, context) {
+        try {
+            const engineRaw = optionalString(input, 'engine');
+            const engine = engineRaw === 'gl' ? 'gl' : engineRaw === 'xfade' ? 'xfade' : undefined;
+            const defaultDuration = optionalNumber(input, 'transition_duration') ?? 1;
+            const result = await assembleFilm({
+                clips: optionalStringList(input.clips) ?? [],
+                transitions: parseStitchTransitions(input, defaultDuration),
+                transitionDuration: optionalNumber(input, 'transition_duration'),
+                engine,
+                resolution: optionalString(input, 'resolution'),
+                aspectRatio: optionalString(input, 'aspect_ratio'),
+                fps: optionalNumber(input, 'fps'),
+                music: optionalString(input, 'music'),
+                musicVolume: optionalNumber(input, 'music_volume'),
+                ducking: optionalBoolean(input.ducking),
+                voiceover: optionalString(input, 'voiceover'),
+                name: optionalString(input, 'name'),
+                output: optionalString(input, 'output'),
+                rootDir: this.options.rootDir ?? context?.cwd,
+            }, this.options.deps ?? {});
+            return {
+                success: result.success,
+                output: JSON.stringify(result, null, 2),
+                data: result,
+                error: result.success ? undefined : result.error,
+            };
+        }
+        catch (error) {
+            return {
+                success: false,
+                error: error instanceof Error ? error.message : String(error),
+            };
+        }
+    }
+    getSchema() {
+        return {
+            name: this.name,
+            description: this.description,
+            parameters: {
+                type: 'object',
+                properties: {
+                    clips: { type: 'array', items: { type: 'string' }, description: 'Ordered local video paths to weld together.' },
+                    transition: { type: 'string', description: "Single transition applied at every boundary (fade, wipeleft, dissolve, circleopen, 'cut'…). Default 'fade'." },
+                    transitions: { type: 'array', items: { type: 'object' }, description: 'Optional per-boundary transitions ({type, duration}), length clips.length−1.' },
+                    transition_duration: { type: 'number', description: 'Default transition duration in seconds (default 1).' },
+                    engine: { type: 'string', enum: ['xfade', 'gl'], description: "Transition engine ('xfade' default, 'gl' falls back to xfade)." },
+                    resolution: { type: 'string', description: "Preset ('720p','1080p','4k'…) or 'WxH'. Defaults to the first clip." },
+                    aspect_ratio: { type: 'string', enum: ['16:9', '9:16', '1:1', '4:3', '3:4', '3:2', '2:3'], description: 'Aspect ratio for a resolution preset. Default 16:9.' },
+                    fps: { type: 'number', description: 'Output frame rate. Defaults to the first clip, else 30.' },
+                    music: { type: 'string', description: 'Optional background music path (looped, ducked).' },
+                    music_volume: { type: 'number', description: 'Music volume 0..1 (default 0.25).' },
+                    ducking: { type: 'boolean', description: 'Duck music under dialogue/voiceover (default true).' },
+                    voiceover: { type: 'string', description: 'Optional full-length narration audio path.' },
+                    name: { type: 'string', description: 'Optional film name (output filename + sidecar).' },
+                    output: { type: 'string', description: 'Optional explicit output path.' },
+                },
+                required: ['clips'],
+            },
+        };
+    }
+    validate(input) {
+        if (typeof input !== 'object' || input === null)
+            return { valid: false, errors: ['Input must be an object'] };
+        const data = input;
+        const clips = optionalStringList(data.clips);
+        if (!clips || clips.length === 0)
+            return { valid: false, errors: ['clips (a non-empty array of video paths) is required'] };
+        return { valid: true };
+    }
+    getMetadata() {
+        return {
+            name: this.name,
+            description: this.description,
+            category: 'media',
+            keywords: ['video', 'stitch', 'montage', 'film', 'concatenate', 'transition', 'xfade', 'enchainer', 'assembler', 'media'],
+            priority: 8,
+            modifiesFiles: true,
+            makesNetworkRequests: false,
+        };
+    }
+    isAvailable() { return true; }
+}
+// ============================================================================
 // VideoExecuteTool
 // ============================================================================
 export class VideoExecuteTool {
@@ -1054,13 +1528,116 @@ export class ArchiveExecuteTool {
 /**
  * Create all multimodal tool instances
  */
+/** Generate a portrait of Lisa (LoRA trigger) and optionally send it on Telegram. */
+export class LisaSelfieTool {
+    options;
+    name = 'lisa_selfie';
+    description = 'Generate a photo/selfie of Lisa (companion character, Krea LoRA when installed) and optionally send it on Telegram. Use when the user asks for a picture of Lisa / photo of you / selfie.';
+    constructor(options = {}) {
+        this.options = options;
+    }
+    async execute(input, context) {
+        try {
+            const { createAndMaybeSendLisaSelfie } = await import('../../companion/lisa-selfie.js');
+            const moodRaw = optionalString(input, 'mood') ?? 'portrait';
+            const mood = ['tender', 'playful', 'bold', 'sparkly', 'calm', 'mika', 'portrait'].includes(moodRaw)
+                ? moodRaw
+                : 'portrait';
+            const sendTelegram = input.send_telegram !== false && input.sendTelegram !== false;
+            const contentTierRaw = optionalString(input, 'content_tier')
+                ?? optionalString(input, 'contentTier')
+                ?? 'safe';
+            const contentTier = ['safe', 'sensual', 'explicit'].includes(contentTierRaw)
+                ? contentTierRaw
+                : 'safe';
+            const result = await createAndMaybeSendLisaSelfie({
+                mood,
+                ...(optionalString(input, 'scene') ? { scene: optionalString(input, 'scene') } : {}),
+                sendTelegram,
+                contentTier,
+                aspectRatio: optionalString(input, 'aspect_ratio') || 'portrait',
+                rootDir: this.options.rootDir ?? context?.cwd ?? process.cwd(),
+                force: input.force === true,
+            });
+            return {
+                success: result.success,
+                output: JSON.stringify(result, null, 2),
+                data: result,
+                ...(result.error && !result.success ? { error: result.error } : {}),
+            };
+        }
+        catch (error) {
+            return {
+                success: false,
+                error: error instanceof Error ? error.message : String(error),
+            };
+        }
+    }
+    getSchema() {
+        return {
+            name: this.name,
+            description: this.description,
+            parameters: {
+                type: 'object',
+                properties: {
+                    mood: {
+                        type: 'string',
+                        enum: ['tender', 'playful', 'bold', 'sparkly', 'calm', 'mika', 'portrait'],
+                        description: 'Expressive register for the portrait.',
+                    },
+                    scene: {
+                        type: 'string',
+                        description: 'Optional short scene description (no paths).',
+                    },
+                    send_telegram: {
+                        type: 'boolean',
+                        description: 'Send the photo on Telegram after generation (default true when configured).',
+                    },
+                    aspect_ratio: {
+                        type: 'string',
+                        enum: ['landscape', 'square', 'portrait'],
+                        description: 'Default portrait.',
+                    },
+                    content_tier: {
+                        type: 'string',
+                        enum: ['safe', 'sensual', 'explicit'],
+                        description: 'Presentation tier. Use sensual for sexy/glamour requests. Explicit remains protected by the verified adult gate.',
+                    },
+                },
+                required: [],
+            },
+        };
+    }
+    getMetadata() {
+        return {
+            name: this.name,
+            description: this.description,
+            category: 'media',
+            keywords: ['lisa', 'selfie', 'photo', 'telegram', 'portrait', 'companion', 'lora'],
+            priority: 8,
+            requiresConfirmation: false,
+            modifiesFiles: true,
+            makesNetworkRequests: true,
+        };
+    }
+    isAvailable() {
+        return true;
+    }
+}
 export function createMultimodalTools() {
     return [
         new AudioExecuteTool(),
         new TextToSpeechTool(),
         new ImageGenerateTool(),
+        new LisaSelfieTool(),
+        new ImageEditTool(),
         new VideoAnalyzeTool(),
+        new UnderstandVideoTool(),
+        new GpuMediaJobTool(),
         new VideoGenerateTool(),
+        new VideoStitchTool(),
+        new MarkdownConvertTool(),
+        ...createVideoStudioTools(),
         new VideoExecuteTool(),
         new PDFExecuteTool(),
         new OCRExecuteTool(),
@@ -1104,7 +1681,7 @@ function optionalProvider(value) {
     if (typeof value !== 'string' || !value.trim())
         return undefined;
     const provider = value.trim();
-    if (['auto', 'system', 'edge-tts', 'espeak', 'say', 'audioreader'].includes(provider)) {
+    if (['auto', 'system', 'edge-tts', 'espeak', 'say', 'audioreader', 'piper'].includes(provider)) {
         return provider;
     }
     throw new Error(`Unsupported text_to_speech provider: ${provider}`);

package/dist/tools/registry/process-tools.d.ts

@@ -16,5 +16,21 @@ export declare class ProcessOperationTool implements ITool {
     isAvailable(): boolean;
     dispose(): void;
 }
+/**
+ * Managed dev-server lifecycle (Manus-style sessioned shell for the
+ * develop → launch → browse → verify loop): start spawns the server in the
+ * background, waits for the loopback URL to answer, and registers the origin
+ * as browsable for exactly as long as the process lives.
+ */
+export declare class AppServerExecuteTool implements ITool {
+    readonly name = "app_server";
+    readonly description = "Start/stop a managed local dev server (e.g. \"npm run dev\") and make its loopback URL browsable so the browser tool can test the app. Actions: start (command + readiness url), stop (pid), status, logs (pid).";
+    execute(input: Record<string, unknown>): Promise<ToolResult>;
+    getSchema(): ToolSchema;
+    validate(input: unknown): IValidationResult;
+    getMetadata(): IToolMetadata;
+    isAvailable(): boolean;
+    dispose(): void;
+}
 export declare function createProcessTools(): ITool[];
 export declare function resetProcessInstance(): Promise<void>;