@remixmate/cli 0.1.0 → 0.1.2

package/dist/handlers/gen-image.js

@@ -9,10 +9,9 @@
  *   - Async response (generationId + status:'generating'/'pending') is polled
  *     against /model/getImgStatus until completed/failed/timeout
  */
-import { readFileSync, existsSync, statSync } from 'node:fs';
-import path from 'node:path';
 import { mmPost, pollUntil, resolveHttpContext, SkillError } from '../http.js';
 import { emitProgress } from '../progress.js';
+import { isTrue, resolveImageInput, toNumber, toStringArray } from './shared.js';
 const DEFAULT_IMAGE_MODEL = 'doubao/doubao-seedream-5-0-260128';
 // Maps aspect-ratio shorthand → (width, height) for Seedream models.
 const SIZE_PRESETS = {
@@ -23,13 +22,6 @@ const SIZE_PRESETS = {
     '3:4': [1728, 2304],
     '2:3': [1664, 2496],
 };
-const MIME_BY_EXT = {
-    '.png': 'image/png',
-    '.jpg': 'image/jpeg',
-    '.jpeg': 'image/jpeg',
-    '.webp': 'image/webp',
-    '.gif': 'image/gif',
-};
 // Per-model reference-image caps (matches ab-api validation).
 const SEEDREAM_REF_LIMIT = 14;
 const GEMINI_REF_LIMIT = 4;
@@ -37,38 +29,6 @@ function isSeedreamModel(model) {
     const lower = model.toLowerCase();
     return lower.includes('seedream') || lower.includes('seedance');
 }
-function resolveImageInput(pathOrUrl) {
-    const trimmed = pathOrUrl.trim();
-    if (!trimmed)
-        return '';
-    const lower = trimmed.toLowerCase();
-    if (lower.startsWith('http://') || lower.startsWith('https://') || lower.startsWith('data:')) {
-        return trimmed;
-    }
-    if (!existsSync(trimmed) || !statSync(trimmed).isFile()) {
-        throw new SkillError(`❌ image path does not exist or is not a file: ${trimmed}`);
-    }
-    const ext = path.extname(trimmed).toLowerCase();
-    const mime = MIME_BY_EXT[ext] ?? 'image/jpeg';
-    const b64 = readFileSync(trimmed).toString('base64');
-    return `data:${mime};base64,${b64}`;
-}
-/** Coerce a CLI flag value (string|string[]|boolean) into an array of strings. */
-function toStringArray(value) {
-    if (value === undefined || value === null)
-        return [];
-    if (Array.isArray(value))
-        return value.map(String);
-    return [String(value)];
-}
-function toNumber(value, name) {
-    if (value === undefined || value === null)
-        return undefined;
-    const n = Number(value);
-    if (!Number.isFinite(n))
-        throw new SkillError(`❌ ${name} must be a number, received: ${value}`);
-    return n;
-}
 /** Resolve {width, height} for Seedream from --size (preset key or WxH). */
 function resolveSeedreamSize(size) {
     if (SIZE_PRESETS[size]) {
@@ -153,6 +113,10 @@ export async function genImage(input, ctxIn) {
     if (imageUrls.length === 0) {
         throw new SkillError(`❌ API response did not contain any image URL: ${JSON.stringify(data)}`);
     }
+    if (isTrue(input.json_output)) {
+        process.stdout.write(JSON.stringify({ urls: imageUrls }) + '\n');
+        return;
+    }
     process.stdout.write(`\n📸 Generated ${imageUrls.length} image(s)\n`);
     process.stdout.write(`\n🔗 Image URLs (renders directly, no download needed):\n`);
     for (const url of imageUrls)

package/dist/handlers/gen-video.d.ts

@@ -0,0 +1,18 @@
+/**
+ * gen-video handler — POST ab-api /model/genVideo, then poll /model/getVideoStatus.
+ *
+ * Ported from the previous `scripts/gen_video.py` (entry.type python →
+ * entry.type http) so it shares auth / base-URL / business-code / polling with
+ * the other atomic generators via http.ts instead of re-implementing them.
+ *
+ * Model families (aligned with the ab-web AI_VIDEO_MODELS dropdown):
+ *   - Seedance 1.5 Pro (Volcano): audio, first/last frame, fixed camera,
+ *     adaptive ratio, duration 4–12s
+ *   - Veo 3.1 / Veo 3.1 Fast (Google): first/last frame, reference images,
+ *     negative prompt, 4K, duration 4/6/8s
+ *
+ * Local image paths for first/last/reference frames are read and base64-encoded
+ * into a data: URI; https / data: URIs pass through unchanged.
+ */
+import type { HandlerContext, HandlerInput } from './index.js';
+export declare function genVideo(input: HandlerInput, ctxIn: HandlerContext): Promise<void>;

package/dist/handlers/gen-video.js

@@ -0,0 +1,153 @@
+/**
+ * gen-video handler — POST ab-api /model/genVideo, then poll /model/getVideoStatus.
+ *
+ * Ported from the previous `scripts/gen_video.py` (entry.type python →
+ * entry.type http) so it shares auth / base-URL / business-code / polling with
+ * the other atomic generators via http.ts instead of re-implementing them.
+ *
+ * Model families (aligned with the ab-web AI_VIDEO_MODELS dropdown):
+ *   - Seedance 1.5 Pro (Volcano): audio, first/last frame, fixed camera,
+ *     adaptive ratio, duration 4–12s
+ *   - Veo 3.1 / Veo 3.1 Fast (Google): first/last frame, reference images,
+ *     negative prompt, 4K, duration 4/6/8s
+ *
+ * Local image paths for first/last/reference frames are read and base64-encoded
+ * into a data: URI; https / data: URIs pass through unchanged.
+ */
+import { mmPost, pollUntil, resolveHttpContext, SkillError } from '../http.js';
+import { emitProgress } from '../progress.js';
+import { isTrue, resolveImageInput, toNumber, toStringArray } from './shared.js';
+const DEFAULT_VIDEO_MODEL = 'doubao-seedance-1-5-pro-251215';
+// CLI shorthand → full LiteLLM / gateway model id (matches the frontend dropdown).
+const MODEL_ALIASES = {
+    seedance: 'doubao-seedance-1-5-pro-251215',
+    'seedance-1.5': 'doubao-seedance-1-5-pro-251215',
+    'seedance-1.5-pro': 'doubao-seedance-1-5-pro-251215',
+    veo: 'veo-3.1-generate-001',
+    'veo-3.1': 'veo-3.1-generate-001',
+    'veo-3.1-generate': 'veo-3.1-generate-001',
+    'veo-fast': 'veo-3.1-fast-generate-001',
+    'veo-3.1-fast': 'veo-3.1-fast-generate-001',
+    'veo-3.1-fast-generate': 'veo-3.1-fast-generate-001',
+};
+const SEEDANCE_RATIOS = new Set(['adaptive', '16:9', '4:3', '1:1', '3:4', '9:16', '21:9']);
+const SEEDANCE_RESOLUTIONS = new Set(['480p', '720p', '1080p']);
+const VEO_RATIOS = new Set(['16:9', '9:16']);
+const VEO_RESOLUTIONS = new Set(['720p', '1080p', '4k']);
+const VEO_DURATIONS = new Set([4, 6, 8]);
+const VEO_MAX_REFERENCE_IMAGES = 3;
+function normalizeModelId(model) {
+    const key = model.trim().toLowerCase();
+    return MODEL_ALIASES[key] ?? model.trim();
+}
+function isSeedanceModel(model) {
+    return model.toLowerCase().includes('seedance');
+}
+function isVeoModel(model) {
+    return model.toLowerCase().includes('veo');
+}
+function validateParamsForModel(model, duration, ratio, resolution, refCount) {
+    if (isSeedanceModel(model)) {
+        if (duration < 4 || duration > 12) {
+            throw new SkillError(`❌ Seedance duration must be 4–12 s, got: ${duration}`);
+        }
+        if (!SEEDANCE_RATIOS.has(ratio)) {
+            throw new SkillError(`❌ Seedance aspect ratio must be one of: ${[...SEEDANCE_RATIOS].join(', ')}; got: ${ratio}`);
+        }
+        if (!SEEDANCE_RESOLUTIONS.has(resolution)) {
+            throw new SkillError(`❌ Seedance resolution must be one of: ${[...SEEDANCE_RESOLUTIONS].join(', ')}; got: ${resolution}`);
+        }
+        if (refCount > 0) {
+            throw new SkillError('❌ Seedance 1.5 Pro does not support a reference-image array (first / last frame only); use --first-frame / --last-frame instead');
+        }
+    }
+    else if (isVeoModel(model)) {
+        if (!VEO_DURATIONS.has(duration)) {
+            throw new SkillError(`❌ Veo duration must be one of: ${[...VEO_DURATIONS].join(', ')} s; got: ${duration}`);
+        }
+        if (!VEO_RATIOS.has(ratio)) {
+            throw new SkillError(`❌ Veo aspect ratio must be one of: ${[...VEO_RATIOS].join(', ')}; got: ${ratio}`);
+        }
+        if (!VEO_RESOLUTIONS.has(resolution)) {
+            throw new SkillError(`❌ Veo resolution must be one of: ${[...VEO_RESOLUTIONS].join(', ')}; got: ${resolution}`);
+        }
+        if (refCount > VEO_MAX_REFERENCE_IMAGES) {
+            throw new SkillError(`❌ Veo accepts at most ${VEO_MAX_REFERENCE_IMAGES} reference images; got: ${refCount}`);
+        }
+    }
+}
+async function pollVideoStatus(ctx, taskId) {
+    return pollUntil(async () => {
+        const data = await mmPost(ctx, '/model/getVideoStatus', { taskId }, { timeoutMs: 30_000 });
+        const status = data.status ?? 'unknown';
+        emitProgress({ phase: 'gen-video:poll', taskId, status });
+        if (status === 'succeeded' || status === 'completed') {
+            return data.videoUrl ?? '';
+        }
+        if (status === 'failed') {
+            const err = data.errorMsg ?? data.error ?? data.message ?? 'unknown error';
+            throw new SkillError(`❌ video generation failed: ${err}`);
+        }
+        return null;
+    }, { intervalMs: 10_000, timeoutMs: 300_000 });
+}
+export async function genVideo(input, ctxIn) {
+    const model = normalizeModelId(input.model ?? process.env.MM_VIDEO_MODEL ?? DEFAULT_VIDEO_MODEL);
+    const firstFrame = input.first_frame ? resolveImageInput(String(input.first_frame)) : '';
+    const lastFrame = input.last_frame ? resolveImageInput(String(input.last_frame)) : '';
+    const references = toStringArray(input.reference).map(resolveImageInput).filter(Boolean);
+    const prompt = (input.prompt ?? '').trim();
+    const hasVisual = Boolean(firstFrame || lastFrame || references.length > 0);
+    if (!prompt && !hasVisual) {
+        throw new SkillError('❌ Provide at least one of: a non-empty --prompt, or --first-frame / --last-frame / --reference');
+    }
+    const ratio = input.ratio ?? '16:9';
+    const resolution = input.resolution ?? '720p';
+    const duration = toNumber(input.duration, 'duration') ?? (isVeoModel(model) ? 8 : 5);
+    const seed = toNumber(input.seed, 'seed');
+    validateParamsForModel(model, duration, ratio, resolution, references.length);
+    const ctx = resolveHttpContext(ctxIn.skillName, {
+        apiBaseUrl: input.api_base_url,
+        privateToken: input.priv_token,
+    });
+    const payload = {
+        model,
+        duration,
+        ratio,
+        resolution,
+        generateAudio: isTrue(input.generate_audio),
+        cameraFixed: isTrue(input.camera_fixed),
+    };
+    if (prompt)
+        payload.prompt = prompt;
+    if (firstFrame)
+        payload.firstFrameImage = firstFrame;
+    if (lastFrame)
+        payload.lastFrameImage = lastFrame;
+    if (references.length > 0)
+        payload.referenceImages = references;
+    if (input.negative_prompt && String(input.negative_prompt).trim()) {
+        payload.negativePrompt = String(input.negative_prompt).trim();
+    }
+    if (seed !== undefined)
+        payload.seed = seed;
+    if (input.person_generation)
+        payload.personGeneration = String(input.person_generation);
+    emitProgress({ phase: 'gen-video:request', model, ratio, resolution, duration });
+    const data = await mmPost(ctx, '/model/genVideo', payload, { timeoutMs: 120_000 });
+    const taskId = data.taskId;
+    if (!taskId) {
+        throw new SkillError(`❌ could not parse the task id; API returned: ${JSON.stringify(data)}`);
+    }
+    emitProgress({ phase: 'gen-video:async', taskId });
+    const videoUrl = await pollVideoStatus(ctx, taskId);
+    if (!videoUrl) {
+        throw new SkillError(`❌ video generated but no URL was returned, task id: ${taskId}`);
+    }
+    if (isTrue(input.json_output)) {
+        process.stdout.write(JSON.stringify({ url: videoUrl }) + '\n');
+        return;
+    }
+    process.stdout.write(`\n🔗 Video URL (renders directly, no download needed):\n${videoUrl}\n`);
+    process.stdout.write(`\n🎉 Video generation done!\n`);
+}

package/dist/handlers/gen-voice.js

@@ -14,6 +14,7 @@
  */
 import { mmPost, resolveHttpContext, SkillError } from '../http.js';
 import { emitProgress } from '../progress.js';
+import { isTrue, toNumber } from './shared.js';
 const DEFAULT_VOICE_ID = 'Chinese (Mandarin)_Male_Announcer';
 /**
  * Fallback voice catalog printed when `--list-voices --local` is passed.
@@ -27,17 +28,6 @@ const LOCAL_VOICE_CATALOG = [
     // TODO: replace once gen-voice service confirms an English default voice id.
     { id: 'english-male-friendly-01', language: 'en', name: 'Friendly Male (English, TBD)' },
 ];
-function toNumber(value, name, defaultValue) {
-    if (value === undefined || value === null || value === '')
-        return defaultValue;
-    const n = Number(value);
-    if (!Number.isFinite(n))
-        throw new SkillError(`❌ ${name} must be a number, received: ${value}`);
-    return n;
-}
-function isTrue(value) {
-    return value === true || value === 'true';
-}
 async function listVoicesRemote(ctx) {
     const data = await mmPost(ctx, '/voice/page', { current: 1, pageSize: 100 }, { timeoutMs: 30_000 });
     const voices = data.list ?? [];
@@ -75,7 +65,7 @@ export async function genVoice(input, ctxIn) {
         throw new SkillError('❌ Please provide --text to specify the text to synthesize');
     }
     const voiceId = input.voice_id ?? DEFAULT_VOICE_ID;
-    const speed = toNumber(input.speed, 'speed', 1.0);
+    const speed = toNumber(input.speed, 'speed') ?? 1.0;
     if (speed < 0.5 || speed > 2.0) {
         throw new SkillError('❌ speed must be in the range 0.5~2.0');
     }

package/dist/handlers/index.js

@@ -6,9 +6,11 @@
  */
 import { genImage } from './gen-image.js';
 import { genVoice } from './gen-voice.js';
-import { templateBind } from './template-bind.js';
+import { genVideo } from './gen-video.js';
+import { genDigitalHuman } from './gen-digital-human.js';
 export const HANDLERS = {
     'gen-image': genImage,
     'gen-voice': genVoice,
-    'template-bind': templateBind,
+    'gen-video': genVideo,
+    'gen-digital-human': genDigitalHuman,
 };

package/dist/handlers/shared.d.ts

@@ -0,0 +1,26 @@
+/**
+ * Shared helpers for TS skill handlers.
+ *
+ * These were previously copy-pasted across gen-image / gen-video /
+ * gen-digital-human / gen-voice. Centralizing them keeps argument coercion and
+ * image-input resolution identical across every atomic generator.
+ *
+ * Lives under handlers/ (not src/ root) so the compiled `dist/handlers/shared.js`
+ * travels with the rest of the handler bundle when ab-agent copies `dist/`.
+ */
+/** Truthy check tolerant of CLI string values (`--flag` → true, `--flag true` → 'true'). */
+export declare function isTrue(value: unknown): boolean;
+/**
+ * Coerce a flag value into a finite number, or `undefined` when absent/empty.
+ * Throws SkillError on a non-numeric value so callers fail fast with a clear msg.
+ */
+export declare function toNumber(value: unknown, name: string): number | undefined;
+/** Coerce a repeatable flag (string | string[] | undefined) into a string array. */
+export declare function toStringArray(value: unknown): string[];
+/**
+ * Resolve an image reference into a value the backend accepts:
+ *   - http(s) URLs and data: URIs pass through unchanged
+ *   - local file paths are read and base64-encoded into a `data:<mime>;base64,…` URI
+ * Throws SkillError when a local path does not exist or is not a file.
+ */
+export declare function resolveImageInput(pathOrUrl: string): string;

package/dist/handlers/shared.js

@@ -0,0 +1,66 @@
+/**
+ * Shared helpers for TS skill handlers.
+ *
+ * These were previously copy-pasted across gen-image / gen-video /
+ * gen-digital-human / gen-voice. Centralizing them keeps argument coercion and
+ * image-input resolution identical across every atomic generator.
+ *
+ * Lives under handlers/ (not src/ root) so the compiled `dist/handlers/shared.js`
+ * travels with the rest of the handler bundle when ab-agent copies `dist/`.
+ */
+import { readFileSync, existsSync, statSync } from 'node:fs';
+import path from 'node:path';
+import { SkillError } from '../http.js';
+const MIME_BY_EXT = {
+    '.png': 'image/png',
+    '.jpg': 'image/jpeg',
+    '.jpeg': 'image/jpeg',
+    '.webp': 'image/webp',
+    '.gif': 'image/gif',
+};
+/** Truthy check tolerant of CLI string values (`--flag` → true, `--flag true` → 'true'). */
+export function isTrue(value) {
+    return value === true || value === 'true';
+}
+/**
+ * Coerce a flag value into a finite number, or `undefined` when absent/empty.
+ * Throws SkillError on a non-numeric value so callers fail fast with a clear msg.
+ */
+export function toNumber(value, name) {
+    if (value === undefined || value === null || value === '')
+        return undefined;
+    const n = Number(value);
+    if (!Number.isFinite(n))
+        throw new SkillError(`❌ ${name} must be a number, received: ${value}`);
+    return n;
+}
+/** Coerce a repeatable flag (string | string[] | undefined) into a string array. */
+export function toStringArray(value) {
+    if (value === undefined || value === null)
+        return [];
+    if (Array.isArray(value))
+        return value.map(String);
+    return [String(value)];
+}
+/**
+ * Resolve an image reference into a value the backend accepts:
+ *   - http(s) URLs and data: URIs pass through unchanged
+ *   - local file paths are read and base64-encoded into a `data:<mime>;base64,…` URI
+ * Throws SkillError when a local path does not exist or is not a file.
+ */
+export function resolveImageInput(pathOrUrl) {
+    const trimmed = pathOrUrl.trim();
+    if (!trimmed)
+        return '';
+    const lower = trimmed.toLowerCase();
+    if (lower.startsWith('http://') || lower.startsWith('https://') || lower.startsWith('data:')) {
+        return trimmed;
+    }
+    if (!existsSync(trimmed) || !statSync(trimmed).isFile()) {
+        throw new SkillError(`❌ image path does not exist or is not a file: ${trimmed}`);
+    }
+    const ext = path.extname(trimmed).toLowerCase();
+    const mime = MIME_BY_EXT[ext] ?? 'image/jpeg';
+    const b64 = readFileSync(trimmed).toString('base64');
+    return `data:${mime};base64,${b64}`;
+}

package/dist/http.d.ts

@@ -43,8 +43,10 @@ export declare function mmPost<T = unknown>(ctx: HttpContext, pathOrUrl: string,
     timeoutMs?: number;
 }): Promise<T>;
 /**
- * Generic polling helper. Repeatedly invokes `probe()` every `intervalMs`
- * until it returns a non-null value or `timeoutMs` elapses.
+ * Generic polling helper. Invokes `probe()` immediately, then every
+ * `intervalMs`, until it returns a non-null value or `timeoutMs` elapses.
+ * Probing first (rather than sleeping first) avoids a wasted initial interval
+ * when the backend job is already complete.
  */
 export declare function pollUntil<T>(probe: () => Promise<T | null>, opts?: {
     intervalMs?: number;

package/dist/http.js

@@ -93,18 +93,20 @@ export async function mmPost(ctx, pathOrUrl, body, opts = {}) {
     return parsed.data ?? undefined;
 }
 /**
- * Generic polling helper. Repeatedly invokes `probe()` every `intervalMs`
- * until it returns a non-null value or `timeoutMs` elapses.
+ * Generic polling helper. Invokes `probe()` immediately, then every
+ * `intervalMs`, until it returns a non-null value or `timeoutMs` elapses.
+ * Probing first (rather than sleeping first) avoids a wasted initial interval
+ * when the backend job is already complete.
  */
 export async function pollUntil(probe, opts = {}) {
     const intervalMs = opts.intervalMs ?? 5_000;
     const timeoutMs = opts.timeoutMs ?? 120_000;
     const start = Date.now();
     while (Date.now() - start < timeoutMs) {
-        await new Promise((r) => setTimeout(r, intervalMs));
         const result = await probe();
         if (result !== null)
             return result;
+        await new Promise((r) => setTimeout(r, intervalMs));
     }
     throw new SkillError(`❌ polling timed out (${Math.round(timeoutMs / 1000)}s)`);
 }

package/dist/manifest.json

@@ -1,11 +1,12 @@
 {
   "schemaVersion": 1,
-  "version": "0.1.0",
-  "generatedAt": "2026-06-06T15:02:44.292Z",
+  "version": "0.1.2",
+  "generatedAt": "2026-06-08T08:37:23.147Z",
   "skills": [
     {
       "id": "export-jianying",
       "tier": "orchestration",
+      "category": "consuming",
       "title": "Jianying (CapCut) Draft Export",
       "summary": "Jianying (CapCut) draft export: package asset URLs into a draft ZIP that Jianying can import. Supports automatic conversion from a RenderPlan.",
       "triggers": [
@@ -28,6 +29,7 @@
     {
       "id": "gen-digital-human",
       "tier": "atomic",
+      "category": "asset",
       "title": "Digital-Human Talking-Head",
       "summary": "Digital-human video: list available avatars; produce a talking-head video from text via TTS, or drive an avatar from an existing audio URL.",
       "triggers": [
@@ -37,10 +39,10 @@
         "Use Jimeng or HiFly to drive an avatar"
       ],
       "entry": {
-        "type": "python",
-        "scriptPath": "scripts/gen_digital_human_video.py"
+        "type": "http",
+        "handler": "gen-digital-human"
       },
-      "runtime": "python",
+      "runtime": "node",
       "envVars": [
         "PRIV_TOKEN",
         "MM_API_BASE_URL",
@@ -50,6 +52,7 @@
     {
       "id": "gen-image",
       "tier": "atomic",
+      "category": "asset",
       "title": "AI Image Generation",
       "summary": "AI image generation: produce an image from a text prompt. Supports Seedream and Gemini models, plus image-to-image with reference images.",
       "triggers": [
@@ -73,6 +76,7 @@
     {
       "id": "gen-script",
       "tier": "orchestration",
+      "category": "authoring",
       "title": "Video Script Generation",
       "summary": "Video script generation: turn a topic into a structured Video DSL (JSON) that describes the full video — scene structure, asset requirements, and narrative flow.",
       "triggers": [
@@ -95,6 +99,7 @@
     {
       "id": "gen-video",
       "tier": "atomic",
+      "category": "asset",
       "title": "AI Video Generation",
       "summary": "AI video generation: produce a short video clip from a text prompt. Supports Seedance and Veo models, plus first/last frame and reference images.",
       "triggers": [
@@ -103,10 +108,10 @@
         "Image-to-video, first-frame / last-frame, reference-image-to-video"
       ],
       "entry": {
-        "type": "python",
-        "scriptPath": "scripts/gen_video.py"
+        "type": "http",
+        "handler": "gen-video"
       },
-      "runtime": "python",
+      "runtime": "node",
       "envVars": [
         "PRIV_TOKEN",
         "MM_API_BASE_URL",
@@ -117,6 +122,7 @@
     {
       "id": "gen-voice",
       "tier": "atomic",
+      "category": "asset",
       "title": "Text-to-Speech (Minimax)",
       "summary": "Text-to-speech (TTS): synthesize narration audio from text via the Minimax TTS model. Returns the persisted audio URL — no download needed.",
       "triggers": [
@@ -138,6 +144,7 @@
     {
       "id": "prepare-video-assets",
       "tier": "orchestration",
+      "category": "authoring",
       "title": "Video Asset Preparation",
       "summary": "Resolves and generates every asset (image / audio / video) referenced by a Video DSL, persists a RenderPlan to the database, and returns a job_id for the subsequent render_video call. This is Phase 1 of the two-phase video pipeline; Phase 3 (Remotion render) lives in render_video.",
       "triggers": [
@@ -165,6 +172,7 @@
     {
       "id": "render-video",
       "tier": "orchestration",
+      "category": "authoring",
       "title": "Remotion Video Renderer",
       "summary": "Loads a persisted RenderPlan by job_id and drives the Remotion engine to produce the final video. Assets must already be generated via prepare_video_assets — this skill never resolves or regenerates assets.",
       "triggers": [
@@ -190,20 +198,19 @@
       ]
     },
     {
-      "id": "template-bind",
+      "id": "template-registry",
       "tier": "orchestration",
-      "title": "Template Registry (List-only)",
+      "category": "authoring",
+      "title": "Video Template Registry",
       "summary": "List all available video templates (templateId / name / aspect ratio / style tags). Template-to-DSL binding is no longer exposed as a separate step — once prepare_video_assets receives a template_id it runs the full DSL→RenderPlan pipeline internally.",
       "triggers": [
-        "View available templates / list every template",
-        "Bind a template to a DSL (a template-id must be supplied)",
-        "Agent has a Video DSL and needs a TemplateBinding"
+        "View available templates / list every template"
       ],
       "entry": {
-        "type": "builtin",
-        "handler": "template-bind"
+        "type": "python",
+        "scriptPath": "scripts/list_templates.py"
       },
-      "runtime": "node",
+      "runtime": "python",
       "envVars": [
         "PRIV_TOKEN",
         "VIDEO_TEMPLATE_REGISTRY",
@@ -214,6 +221,7 @@
     {
       "id": "video-parser",
       "tier": "tool",
+      "category": "consuming",
       "title": "Video Deconstruction",
       "summary": "Video deconstruction: split a video into reusable assets (audio extraction, ASR transcript, keyframe extraction, scene segmentation).",
       "triggers": [
@@ -227,25 +235,31 @@
       },
       "runtime": "python",
       "envVars": [
+        "RENDER_API_URL",
         "PRIV_TOKEN",
-        "MM_API_BASE_URL",
-        "AGENT_NAME",
-        "RENDER_API_URL"
+        "CONVERSATION_ID"
       ]
     },
     {
       "id": "web-capture",
       "tier": "tool",
+      "category": "consuming",
       "title": "Web Page Capture",
-      "summary": "Drive a headless browser (Playwright) to capture a web page: viewport/full-page screenshots, or a smooth top-to-bottom scrolling video. Produces PNG / MP4 assets that feed the screen-walkthrough template.",
+      "summary": "Drive a headless browser (Playwright Python) to capture any URL to local files: full-page / element / region screenshots, fixed-duration / condition-triggered / auto-scroll recordings, and storyboard videos (highlight / focus / zoom / scroll / virtual-cursor click / caption / title-card / arrow / numbered sequence / redact / code-line highlight). Two entry scripts: scripts/screenshot.py (stills) and scripts/record.py (webm video, storyboards, templates).",
       "triggers": [
-        "Screenshot a web page / website / a specific URL, full-page screenshot",
-        "Record a page scrolling top-to-bottom into a video, \"scroll video\" of a site",
-        "Produce screen-recording / walkthrough footage of a GitHub repo, landing page, or product page"
+        "网页截图、网页截屏、整页截图、长截图、full page screenshot",
+        "截某个元素 / 区域、局部截屏、focus 某个区域",
+        "录屏、网页录制、录制视频、录一段操作、生成 webm",
+        "滚动录屏、页面从头划到尾的视频",
+        "多场景视频、分镜视频、storyboard、按时间线编排",
+        "演示点击 / 输入 / 悬停（虚拟鼠标 + 涟漪 / 打字机 / 触发 tooltip）",
+        "加箭头、加字幕条、加标题卡 / 片头片尾、按顺序编号高亮多个区域",
+        "截图前隐藏元素 / 涂盖打码、给静态截图加注释（红框 / 箭头 / 标签）",
+        "移动端 / 设备模拟、带 cookie / 登录态截屏 / 录屏"
       ],
       "entry": {
         "type": "python",
-        "scriptPath": "scripts/web_capture.py"
+        "scriptPath": "scripts/screenshot.py"
       },
       "runtime": "python",
       "envVars": [

package/dist/registry.d.ts

@@ -1,24 +1,13 @@
 /**
  * Skill registry — scan skills/ and load skill.json definitions.
  *
- * skill.json schema:
- *   { name, toolName, description, parameters, entry?, scriptPath? }
- *
- * `entry` is the dispatcher hint. Missing entry falls back to the legacy
- * top-level scriptPath, treated as { type:'python', scriptPath } so older
- * skills keep working without a touch.
+ * Validation and entry-normalization live in `skill-schema.ts` (shared with
+ * the build-time manifest generator and smoke test). A record that fails
+ * validation is skipped with a warning rather than crashing the whole CLI.
  */
+import { type SkillEntry } from './skill-schema.js';
 export declare const SKILLS_DIR: string;
-export type SkillEntry = {
-    type: 'python';
-    scriptPath: string;
-} | {
-    type: 'http';
-    handler: string;
-} | {
-    type: 'builtin';
-    handler: string;
-};
+export type { SkillEntry } from './skill-schema.js';
 export interface SkillDef {
     name: string;
     toolName: string;