mulmocast-preprocessor 0.1.2 → 0.2.0

package/lib/cli/commands/process.js

@@ -1,6 +1,6 @@
 import { readFileSync, writeFileSync } from "fs";
 import { GraphAILogger } from "graphai";
-import { processScript } from "../../core/process.js";
+import { processScript } from "../../core/preprocessing/process.js";
 /**
  * Process script with profile and output result
  */

package/lib/cli/commands/profiles.js

@@ -1,6 +1,6 @@
 import { readFileSync } from "fs";
 import { GraphAILogger } from "graphai";
-import { listProfiles } from "../../core/profiles.js";
+import { listProfiles } from "../../core/preprocessing/profiles.js";
 /**
  * List available profiles in script
  */

package/lib/cli/commands/query.d.ts

@@ -0,0 +1,15 @@
+import type { LLMProvider } from "../../types/summarize.js";
+interface QueryCommandOptions {
+    provider?: LLMProvider;
+    model?: string;
+    lang?: string;
+    systemPrompt?: string;
+    verbose?: boolean;
+    section?: string;
+    tags?: string[];
+}
+/**
+ * Query command handler - outputs answer to stdout
+ */
+export declare const queryCommand: (scriptPath: string, question: string, options: QueryCommandOptions) => Promise<void>;
+export {};

package/lib/cli/commands/query.js

@@ -0,0 +1,31 @@
+import { GraphAILogger } from "graphai";
+import { queryScript } from "../../core/ai/command/query/index.js";
+import { loadScript } from "../utils.js";
+/**
+ * Query command handler - outputs answer to stdout
+ */
+export const queryCommand = async (scriptPath, question, options) => {
+    try {
+        const script = await loadScript(scriptPath);
+        const result = await queryScript(script, question, {
+            provider: options.provider ?? "openai",
+            model: options.model,
+            lang: options.lang,
+            systemPrompt: options.systemPrompt,
+            verbose: options.verbose ?? false,
+            section: options.section,
+            tags: options.tags,
+        });
+        // Output answer to stdout
+        process.stdout.write(result.answer + "\n");
+    }
+    catch (error) {
+        if (error instanceof Error) {
+            GraphAILogger.error(`Error: ${error.message}`);
+        }
+        else {
+            GraphAILogger.error("Unknown error occurred");
+        }
+        process.exit(1);
+    }
+};

package/lib/cli/commands/summarize.js

@@ -1,40 +1,6 @@
-import { readFileSync } from "fs";
 import { GraphAILogger } from "graphai";
-import { summarizeScript } from "../../core/summarize/index.js";
-/**
- * Check if input is a URL
- */
-const isUrl = (input) => {
-    return input.startsWith("http://") || input.startsWith("https://");
-};
-/**
- * Fetch JSON from URL with timeout
- */
-const fetchJson = async (url) => {
-    const controller = new AbortController();
-    const timeout_ms = 30000;
-    const timeoutId = setTimeout(() => controller.abort(), timeout_ms);
-    try {
-        const response = await fetch(url, { signal: controller.signal });
-        if (!response.ok) {
-            throw new Error(`HTTP error: ${response.status} ${response.statusText}`);
-        }
-        return (await response.json());
-    }
-    finally {
-        clearTimeout(timeoutId);
-    }
-};
-/**
- * Load script from file path or URL
- */
-const loadScript = async (input) => {
-    if (isUrl(input)) {
-        return fetchJson(input);
-    }
-    const content = readFileSync(input, "utf-8");
-    return JSON.parse(content);
-};
+import { summarizeScript } from "../../core/ai/command/summarize/index.js";
+import { loadScript } from "../utils.js";
 /**
  * Summarize command handler - outputs summary to stdout
  */

package/lib/cli/index.js

@@ -4,6 +4,7 @@ import { hideBin } from "yargs/helpers";
 import { processCommand } from "./commands/process.js";
 import { profilesCommand } from "./commands/profiles.js";
 import { summarizeCommand } from "./commands/summarize.js";
+import { queryCommand } from "./commands/query.js";
 yargs(hideBin(process.argv))
     .command("$0 <script>", "Process MulmoScript with profile", (builder) => builder
     .positional("script", {
@@ -109,6 +110,62 @@ yargs(hideBin(process.argv))
         section: argv.section,
         tags,
     });
+})
+    .command("query <script> <question>", "Ask a question about the script content", (builder) => builder
+    .positional("script", {
+    describe: "Path or URL to MulmoScript JSON file",
+    type: "string",
+    demandOption: true,
+})
+    .positional("question", {
+    describe: "Question to ask about the script",
+    type: "string",
+    demandOption: true,
+})
+    .option("provider", {
+    describe: "LLM provider (openai, anthropic, groq, gemini)",
+    type: "string",
+    default: "openai",
+})
+    .option("model", {
+    alias: "m",
+    describe: "Model name",
+    type: "string",
+})
+    .option("lang", {
+    alias: "l",
+    describe: "Output language (e.g., ja, en, zh)",
+    type: "string",
+})
+    .option("system-prompt", {
+    describe: "Custom system prompt",
+    type: "string",
+})
+    .option("verbose", {
+    describe: "Show detailed progress",
+    type: "boolean",
+    default: false,
+})
+    .option("section", {
+    alias: "s",
+    describe: "Filter by section name",
+    type: "string",
+})
+    .option("tags", {
+    alias: "t",
+    describe: "Filter by tags (comma-separated)",
+    type: "string",
+}), (argv) => {
+    const tags = argv.tags ? argv.tags.split(",").map((t) => t.trim()) : undefined;
+    queryCommand(argv.script, argv.question, {
+        provider: argv.provider,
+        model: argv.model,
+        lang: argv.lang,
+        systemPrompt: argv.systemPrompt,
+        verbose: argv.verbose,
+        section: argv.section,
+        tags,
+    });
 })
     .example("$0 script.json --profile summary -o summary.json", "Apply summary profile and save to file")
     .example("$0 script.json -p teaser", "Apply teaser profile and output to stdout")
@@ -120,6 +177,8 @@ yargs(hideBin(process.argv))
     .example("$0 summarize script.json --format markdown", "Generate markdown summary")
     .example("$0 summarize script.json -l ja", "Output summary in Japanese")
     .example("$0 summarize https://example.com/script.json", "Summarize from URL")
+    .example('$0 query script.json "What is the main topic?"', "Ask a question about the script")
+    .example('$0 query script.json "登場人物は？" -l ja', "Query in Japanese")
     .help()
     .alias("h", "help")
     .version()

package/lib/cli/utils.d.ts

@@ -0,0 +1,9 @@
+import type { ExtendedScript } from "../types/index.js";
+/**
+ * Check if input is a URL
+ */
+export declare const isUrl: (input: string) => boolean;
+/**
+ * Load script from file path or URL
+ */
+export declare const loadScript: (input: string) => Promise<ExtendedScript>;

package/lib/cli/utils.js

@@ -0,0 +1,35 @@
+import { readFileSync } from "fs";
+/**
+ * Check if input is a URL
+ */
+export const isUrl = (input) => {
+    return input.startsWith("http://") || input.startsWith("https://");
+};
+/**
+ * Fetch JSON from URL with timeout
+ */
+const fetchJson = async (url) => {
+    const controller = new AbortController();
+    const timeout_ms = 30000;
+    const timeoutId = setTimeout(() => controller.abort(), timeout_ms);
+    try {
+        const response = await fetch(url, { signal: controller.signal });
+        if (!response.ok) {
+            throw new Error(`HTTP error: ${response.status} ${response.statusText}`);
+        }
+        return (await response.json());
+    }
+    finally {
+        clearTimeout(timeoutId);
+    }
+};
+/**
+ * Load script from file path or URL
+ */
+export const loadScript = async (input) => {
+    if (isUrl(input)) {
+        return fetchJson(input);
+    }
+    const content = readFileSync(input, "utf-8");
+    return JSON.parse(content);
+};

package/lib/core/ai/command/query/index.d.ts

@@ -0,0 +1,8 @@
+import type { ExtendedScript } from "../../../../types/index.js";
+import type { QueryOptions, QueryResult } from "../../../../types/query.js";
+/**
+ * Main query function - answers a question based on script content
+ */
+export declare const queryScript: (script: ExtendedScript, question: string, options?: Partial<QueryOptions>) => Promise<QueryResult>;
+export type { QueryOptions, QueryResult } from "../../../../types/query.js";
+export { queryOptionsSchema } from "../../../../types/query.js";

package/lib/core/ai/command/query/index.js

@@ -0,0 +1,33 @@
+import { queryOptionsSchema } from "../../../../types/query.js";
+import { executeLLM, filterScript } from "../../llm.js";
+import { buildUserPrompt, getSystemPrompt } from "./prompts.js";
+/**
+ * Main query function - answers a question based on script content
+ */
+export const queryScript = async (script, question, options = {}) => {
+    // Validate and apply defaults
+    const validatedOptions = queryOptionsSchema.parse(options);
+    // Filter script if section/tags specified
+    const filteredScript = filterScript(script, validatedOptions);
+    const scriptTitle = script.title || "Untitled";
+    if (filteredScript.beats.length === 0) {
+        return {
+            answer: "No content available to answer the question.",
+            question,
+            scriptTitle,
+            beatCount: 0,
+        };
+    }
+    // Build prompts
+    const systemPrompt = getSystemPrompt(validatedOptions);
+    const userPrompt = buildUserPrompt(filteredScript, question);
+    // Execute LLM
+    const answer = await executeLLM(systemPrompt, userPrompt, validatedOptions, `Querying script "${script.title}" with ${validatedOptions.provider}... Beats: ${filteredScript.beats.length}, Question: ${question}`);
+    return {
+        answer,
+        question,
+        scriptTitle,
+        beatCount: filteredScript.beats.length,
+    };
+};
+export { queryOptionsSchema } from "../../../../types/query.js";

package/lib/core/ai/command/query/prompts.d.ts

@@ -0,0 +1,14 @@
+import type { QueryOptions } from "../../../../types/query.js";
+import type { ExtendedScript } from "../../../../types/index.js";
+/**
+ * Default system prompt for query
+ */
+export declare const DEFAULT_SYSTEM_PROMPT = "You are answering questions based on the content provided.\n- Answer based ONLY on the information in the provided content\n- If the answer cannot be found in the content, say so clearly\n- Be concise and direct in your answers\n- Do not make up information that is not in the content";
+/**
+ * Get system prompt based on options
+ */
+export declare const getSystemPrompt: (options: QueryOptions) => string;
+/**
+ * Build user prompt from script and question
+ */
+export declare const buildUserPrompt: (script: ExtendedScript, question: string) => string;

package/lib/core/ai/command/query/prompts.js

@@ -0,0 +1,59 @@
+import { getLanguageName } from "../../llm.js";
+/**
+ * Default system prompt for query
+ */
+export const DEFAULT_SYSTEM_PROMPT = `You are answering questions based on the content provided.
+- Answer based ONLY on the information in the provided content
+- If the answer cannot be found in the content, say so clearly
+- Be concise and direct in your answers
+- Do not make up information that is not in the content`;
+/**
+ * Get system prompt based on options
+ */
+export const getSystemPrompt = (options) => {
+    if (options.systemPrompt) {
+        return options.systemPrompt;
+    }
+    const basePrompt = DEFAULT_SYSTEM_PROMPT;
+    // Add language instruction if specified
+    if (options.lang) {
+        const langName = getLanguageName(options.lang);
+        return `${basePrompt}\n- IMPORTANT: Write the answer in ${langName}`;
+    }
+    return basePrompt;
+};
+/**
+ * Build user prompt from script and question
+ */
+export const buildUserPrompt = (script, question) => {
+    const parts = [];
+    // Add script metadata
+    parts.push(`# Script: ${script.title}`);
+    parts.push(`Language: ${script.lang}`);
+    parts.push("");
+    // Collect all text from beats
+    const sections = new Map();
+    script.beats.forEach((beat, index) => {
+        const text = beat.text || "";
+        if (!text.trim())
+            return;
+        const section = beat.meta?.section || "main";
+        if (!sections.has(section)) {
+            sections.set(section, []);
+        }
+        sections.get(section).push(`[${index}] ${text}`);
+    });
+    // Output by section
+    sections.forEach((texts, section) => {
+        parts.push(`## Section: ${section}`);
+        texts.forEach((t) => parts.push(t));
+        parts.push("");
+    });
+    parts.push("");
+    parts.push("---");
+    parts.push("");
+    parts.push(`Question: ${question}`);
+    parts.push("");
+    parts.push("Answer:");
+    return parts.join("\n");
+};

package/lib/core/ai/command/summarize/index.d.ts

@@ -0,0 +1,8 @@
+import type { ExtendedScript } from "../../../../types/index.js";
+import type { SummarizeOptions, SummarizeResult } from "../../../../types/summarize.js";
+/**
+ * Main summarize function - generates a summary of the entire script
+ */
+export declare const summarizeScript: (script: ExtendedScript, options?: Partial<SummarizeOptions>) => Promise<SummarizeResult>;
+export type { SummarizeOptions, SummarizeResult, LLMProvider, SummarizeFormat } from "../../../../types/summarize.js";
+export { summarizeOptionsSchema, llmProviderSchema, summarizeFormatSchema } from "../../../../types/summarize.js";

package/lib/core/ai/command/summarize/index.js

@@ -0,0 +1,33 @@
+import { summarizeOptionsSchema } from "../../../../types/summarize.js";
+import { executeLLM, filterScript } from "../../llm.js";
+import { buildUserPrompt, getSystemPrompt } from "./prompts.js";
+/**
+ * Main summarize function - generates a summary of the entire script
+ */
+export const summarizeScript = async (script, options = {}) => {
+    // Validate and apply defaults
+    const validatedOptions = summarizeOptionsSchema.parse(options);
+    // Filter script if section/tags specified
+    const filteredScript = filterScript(script, validatedOptions);
+    const scriptTitle = script.title || "Untitled";
+    if (filteredScript.beats.length === 0) {
+        return {
+            summary: "No content to summarize.",
+            format: validatedOptions.format,
+            scriptTitle,
+            beatCount: 0,
+        };
+    }
+    // Build prompts
+    const systemPrompt = getSystemPrompt(validatedOptions);
+    const userPrompt = buildUserPrompt(filteredScript, validatedOptions);
+    // Execute LLM
+    const summary = await executeLLM(systemPrompt, userPrompt, validatedOptions, `Summarizing script "${script.title}" with ${validatedOptions.provider}... Beats: ${filteredScript.beats.length}, Format: ${validatedOptions.format}`);
+    return {
+        summary,
+        format: validatedOptions.format,
+        scriptTitle,
+        beatCount: filteredScript.beats.length,
+    };
+};
+export { summarizeOptionsSchema, llmProviderSchema, summarizeFormatSchema } from "../../../../types/summarize.js";

package/lib/core/ai/command/summarize/prompts.d.ts

@@ -0,0 +1,18 @@
+import type { SummarizeOptions } from "../../../../types/summarize.js";
+import type { ExtendedScript } from "../../../../types/index.js";
+/**
+ * Default system prompt for text summary
+ */
+export declare const DEFAULT_SYSTEM_PROMPT_TEXT = "You are creating a summary based on the content provided.\n- Extract and explain the actual information and knowledge from the content\n- Do NOT describe what the presentation/script is about (avoid phrases like \"this presentation explains...\" or \"the script describes...\")\n- Write as if you are directly explaining the topic to the reader\n- Be concise and informative\n- Output plain text only";
+/**
+ * Default system prompt for markdown summary
+ */
+export declare const DEFAULT_SYSTEM_PROMPT_MARKDOWN = "You are creating a summary based on the content provided.\n- Extract and explain the actual information and knowledge from the content\n- Do NOT describe what the presentation/script is about (avoid phrases like \"this presentation explains...\" or \"the script describes...\")\n- Write as if you are directly explaining the topic to the reader\n- Use markdown formatting (headers, bullet points, etc.)\n- Include a title, key points, and conclusion\n- Output well-formatted markdown";
+/**
+ * Build user prompt from entire script
+ */
+export declare const buildUserPrompt: (script: ExtendedScript, options: SummarizeOptions) => string;
+/**
+ * Get system prompt based on format and language
+ */
+export declare const getSystemPrompt: (options: SummarizeOptions) => string;

package/lib/core/ai/command/summarize/prompts.js

@@ -0,0 +1,70 @@
+import { getLanguageName } from "../../llm.js";
+/**
+ * Default system prompt for text summary
+ */
+export const DEFAULT_SYSTEM_PROMPT_TEXT = `You are creating a summary based on the content provided.
+- Extract and explain the actual information and knowledge from the content
+- Do NOT describe what the presentation/script is about (avoid phrases like "this presentation explains..." or "the script describes...")
+- Write as if you are directly explaining the topic to the reader
+- Be concise and informative
+- Output plain text only`;
+/**
+ * Default system prompt for markdown summary
+ */
+export const DEFAULT_SYSTEM_PROMPT_MARKDOWN = `You are creating a summary based on the content provided.
+- Extract and explain the actual information and knowledge from the content
+- Do NOT describe what the presentation/script is about (avoid phrases like "this presentation explains..." or "the script describes...")
+- Write as if you are directly explaining the topic to the reader
+- Use markdown formatting (headers, bullet points, etc.)
+- Include a title, key points, and conclusion
+- Output well-formatted markdown`;
+/**
+ * Build user prompt from entire script
+ */
+export const buildUserPrompt = (script, options) => {
+    const parts = [];
+    // Add script metadata
+    parts.push(`# Script: ${script.title}`);
+    parts.push(`Language: ${script.lang}`);
+    parts.push("");
+    // Collect all text from beats
+    const sections = new Map();
+    script.beats.forEach((beat, index) => {
+        const text = beat.text || "";
+        if (!text.trim())
+            return;
+        const section = beat.meta?.section || "main";
+        if (!sections.has(section)) {
+            sections.set(section, []);
+        }
+        sections.get(section).push(`[${index}] ${text}`);
+    });
+    // Output by section
+    sections.forEach((texts, section) => {
+        parts.push(`## Section: ${section}`);
+        texts.forEach((t) => parts.push(t));
+        parts.push("");
+    });
+    // Add target length if specified
+    if (options.targetLengthChars) {
+        parts.push(`Target summary length: approximately ${options.targetLengthChars} characters`);
+    }
+    parts.push("");
+    parts.push("Based on the above content, explain the topic directly to the reader:");
+    return parts.join("\n");
+};
+/**
+ * Get system prompt based on format and language
+ */
+export const getSystemPrompt = (options) => {
+    if (options.systemPrompt) {
+        return options.systemPrompt;
+    }
+    const basePrompt = options.format === "markdown" ? DEFAULT_SYSTEM_PROMPT_MARKDOWN : DEFAULT_SYSTEM_PROMPT_TEXT;
+    // Add language instruction if specified
+    if (options.lang) {
+        const langName = getLanguageName(options.lang);
+        return `${basePrompt}\n- IMPORTANT: Write the output in ${langName}`;
+    }
+    return basePrompt;
+};

package/lib/core/ai/llm.d.ts

@@ -0,0 +1,45 @@
+import type { ExtendedScript } from "../../types/index.js";
+import type { LLMProvider } from "../../types/summarize.js";
+/**
+ * Base options for LLM operations
+ */
+export interface BaseLLMOptions {
+    provider: LLMProvider;
+    model?: string;
+    temperature?: number;
+    maxTokens?: number;
+    lang?: string;
+    systemPrompt?: string;
+    verbose?: boolean;
+    section?: string;
+    tags?: string[];
+}
+/**
+ * Provider configuration
+ */
+export interface ProviderConfig {
+    agentName: string;
+    defaultModel: string;
+    keyName: string;
+    maxTokens?: number;
+}
+/**
+ * Get provider configuration
+ */
+export declare const getProviderConfig: (provider: LLMProvider) => ProviderConfig;
+/**
+ * Get API key for provider
+ */
+export declare const getProviderApiKey: (provider: LLMProvider) => string | undefined;
+/**
+ * Filter script based on options (section, tags)
+ */
+export declare const filterScript: (script: ExtendedScript, options: BaseLLMOptions) => ExtendedScript;
+/**
+ * Get language name from code
+ */
+export declare const getLanguageName: (langCode: string) => string;
+/**
+ * Execute LLM call with GraphAI
+ */
+export declare const executeLLM: (systemPrompt: string, userPrompt: string, options: BaseLLMOptions, verboseMessage?: string) => Promise<string>;