mulmocast-preprocessor 0.1.2 → 0.2.1

package/README.md

@@ -14,6 +14,8 @@ npm install mulmocast-preprocessor
 - **Section filtering**: Extract beats by section
 - **Tag filtering**: Extract beats by tags
 - **Profile listing**: List available profiles with beat counts
+- **AI Summarization**: Generate summaries using LLM (OpenAI, Anthropic, Groq, Gemini)
+- **AI Query**: Ask questions about script content with interactive mode
 - **CLI tool**: Command-line interface for processing scripts
 
 ## CLI Usage
@@ -36,9 +38,23 @@ mulmocast-preprocessor script.json --profile summary --section chapter1
 
 # List available profiles
 mulmocast-preprocessor profiles script.json
+
+# Summarize script content
+mulmocast-preprocessor summarize script.json
+mulmocast-preprocessor summarize script.json --format markdown
+mulmocast-preprocessor summarize script.json -l ja  # Output in Japanese
+mulmocast-preprocessor summarize https://example.com/script.json  # From URL
+
+# Query script content
+mulmocast-preprocessor query script.json "What is the main topic?"
+mulmocast-preprocessor query script.json "登場人物は？" -l ja
+
+# Interactive query mode
+mulmocast-preprocessor query script.json -i
+mulmocast-preprocessor query script.json  # Omit question for interactive mode
 ```
 
-### CLI Options
+### CLI Options (process command)
 
 | Option | Alias | Description |
 |--------|-------|-------------|
@@ -49,6 +65,41 @@ mulmocast-preprocessor profiles script.json
 | `--help` | `-h` | Show help |
 | `--version` | `-v` | Show version |
 
+### CLI Options (summarize command)
+
+| Option | Alias | Description |
+|--------|-------|-------------|
+| `--provider` | | LLM provider: openai, anthropic, groq, gemini (default: openai) |
+| `--model` | `-m` | Model name |
+| `--format` | `-f` | Output format: text, markdown (default: text) |
+| `--lang` | `-l` | Output language (e.g., ja, en, zh) |
+| `--target-length` | | Target summary length in characters |
+| `--system-prompt` | | Custom system prompt |
+| `--verbose` | | Show detailed progress |
+| `--section` | `-s` | Filter by section name |
+| `--tags` | `-t` | Filter by tags (comma-separated) |
+
+### CLI Options (query command)
+
+| Option | Alias | Description |
+|--------|-------|-------------|
+| `--interactive` | `-i` | Start interactive query mode |
+| `--provider` | | LLM provider: openai, anthropic, groq, gemini (default: openai) |
+| `--model` | `-m` | Model name |
+| `--lang` | `-l` | Output language (e.g., ja, en, zh) |
+| `--system-prompt` | | Custom system prompt |
+| `--verbose` | | Show detailed progress |
+| `--section` | `-s` | Filter by section name |
+| `--tags` | `-t` | Filter by tags (comma-separated) |
+
+### Interactive Query Commands
+
+| Command | Description |
+|---------|-------------|
+| `/clear` | Clear conversation history |
+| `/history` | Show conversation history |
+| `/exit` | Exit interactive mode |
+
 ## Programmatic Usage
 
 ### Basic Example
@@ -132,6 +183,54 @@ Filter beats by section.
 
 Filter beats by tags (extracts beats that have any of the specified tags).
 
+### `summarizeScript(script, options)`
+
+Generate a summary of the script content using LLM.
+
+**Parameters:**
+- `script: ExtendedScript` - Input script
+- `options: SummarizeOptions` - Summarization options
+  - `provider?: LLMProvider` - LLM provider (default: "openai")
+  - `model?: string` - Model name
+  - `format?: "text" | "markdown"` - Output format
+  - `lang?: string` - Output language code
+  - `targetLengthChars?: number` - Target length
+  - `systemPrompt?: string` - Custom system prompt
+
+**Returns:** `Promise<SummarizeResult>` - Summary result with text and metadata
+
+### `queryScript(script, question, options)`
+
+Ask a question about the script content.
+
+**Parameters:**
+- `script: ExtendedScript` - Input script
+- `question: string` - Question to ask
+- `options: QueryOptions` - Query options (same as summarize)
+
+**Returns:** `Promise<QueryResult>` - Answer with question and metadata
+
+### `createInteractiveSession(script, options)`
+
+Create an interactive query session for follow-up questions.
+
+**Parameters:**
+- `script: ExtendedScript` - Input script
+- `options: QueryOptions` - Query options
+
+**Returns:** Session object with `sendInteractiveQuery()` method
+
+## Environment Variables
+
+For AI features (summarize, query), set the API key for your LLM provider:
+
+| Provider | Environment Variable |
+|----------|---------------------|
+| OpenAI | `OPENAI_API_KEY` |
+| Anthropic | `ANTHROPIC_API_KEY` |
+| Groq | `GROQ_API_KEY` |
+| Gemini | `GEMINI_API_KEY` |
+
 ## Extended Schema
 
 ### ExtendedBeat

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 +1,16 @@
+import type { LLMProvider } from "../../types/summarize.js";
+interface QueryCommandOptions {
+    provider?: LLMProvider;
+    model?: string;
+    lang?: string;
+    systemPrompt?: string;
+    verbose?: boolean;
+    section?: string;
+    tags?: string[];
+    interactive?: boolean;
+}
+/**
+ * Query command handler - outputs answer to stdout
+ */
+export declare const queryCommand: (scriptPath: string, question: string | undefined, options: QueryCommandOptions) => Promise<void>;
+export {};

package/lib/cli/commands/query.js

@@ -0,0 +1,119 @@
+import { createInterface } from "node:readline";
+import { GraphAILogger } from "graphai";
+import { queryScript } from "../../core/ai/command/query/index.js";
+import { createInteractiveSession, sendInteractiveQuery, clearHistory } from "../../core/ai/command/query/interactive.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);
+        // Interactive mode
+        if (options.interactive || question === undefined) {
+            await runInteractiveMode(scriptPath, script, options);
+            return;
+        }
+        // Single query mode
+        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: ${error.message}`);
+        }
+        else {
+            GraphAILogger.error("Unknown error occurred");
+        }
+        process.exit(1);
+    }
+};
+/**
+ * Run interactive query mode
+ */
+const runInteractiveMode = async (scriptPath, script, options) => {
+    const { session, filteredScript, validatedOptions } = createInteractiveSession(script, {
+        provider: options.provider ?? "openai",
+        model: options.model,
+        lang: options.lang,
+        systemPrompt: options.systemPrompt,
+        verbose: options.verbose ?? false,
+        section: options.section,
+        tags: options.tags,
+    });
+    if (filteredScript.beats.length === 0) {
+        GraphAILogger.error("No content available to query.");
+        process.exit(1);
+    }
+    const rl = createInterface({
+        input: process.stdin,
+        output: process.stdout,
+    });
+    GraphAILogger.info(`Interactive query mode for "${session.scriptTitle}" (${session.beatCount} beats)`);
+    GraphAILogger.info("Commands: /clear (clear history), /history (show history), /exit or Ctrl+C (quit)");
+    GraphAILogger.info("");
+    const prompt = () => {
+        rl.question("You: ", async (input) => {
+            const trimmedInput = input.trim();
+            if (!trimmedInput) {
+                prompt();
+                return;
+            }
+            // Handle commands
+            if (trimmedInput === "/exit" || trimmedInput === "/quit") {
+                GraphAILogger.info("Goodbye!");
+                rl.close();
+                return;
+            }
+            if (trimmedInput === "/clear") {
+                clearHistory(session);
+                GraphAILogger.info("Conversation history cleared.\n");
+                prompt();
+                return;
+            }
+            if (trimmedInput === "/history") {
+                if (session.history.length === 0) {
+                    GraphAILogger.info("No conversation history.\n");
+                }
+                else {
+                    GraphAILogger.info("Conversation history:");
+                    session.history.forEach((msg) => {
+                        const prefix = msg.role === "user" ? "Q" : "A";
+                        GraphAILogger.info(`${prefix}: ${msg.content}`);
+                    });
+                    GraphAILogger.info("");
+                }
+                prompt();
+                return;
+            }
+            // Send query
+            try {
+                const answer = await sendInteractiveQuery(filteredScript, trimmedInput, session, validatedOptions);
+                GraphAILogger.info(`\nAssistant: ${answer}\n`);
+            }
+            catch (error) {
+                if (error instanceof Error) {
+                    GraphAILogger.error(`Error: ${error.message}\n`);
+                }
+                else {
+                    GraphAILogger.error("Unknown error occurred\n");
+                }
+            }
+            prompt();
+        });
+    };
+    // Handle Ctrl+C
+    rl.on("close", () => {
+        process.exit(0);
+    });
+    prompt();
+};

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 +110,68 @@ 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 (omit for interactive mode)",
+    type: "string",
+})
+    .option("interactive", {
+    alias: "i",
+    describe: "Start interactive query mode",
+    type: "boolean",
+    default: false,
+})
+    .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,
+        interactive: argv.interactive,
+    });
 })
     .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 +183,10 @@ 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")
+    .example("$0 query script.json -i", "Start interactive query mode")
+    .example("$0 query script.json", "Interactive mode (question omitted)")
     .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/interactive.d.ts

@@ -0,0 +1,22 @@
+import type { ExtendedScript } from "../../../../types/index.js";
+import type { QueryOptions, InteractiveQuerySession, ConversationMessage } from "../../../../types/query.js";
+/**
+ * Create an interactive query session
+ */
+export declare const createInteractiveSession: (script: ExtendedScript, options?: Partial<QueryOptions>) => {
+    session: InteractiveQuerySession;
+    filteredScript: ExtendedScript;
+    validatedOptions: QueryOptions;
+};
+/**
+ * Send a question in an interactive session
+ */
+export declare const sendInteractiveQuery: (filteredScript: ExtendedScript, question: string, session: InteractiveQuerySession, options: QueryOptions) => Promise<string>;
+/**
+ * Clear conversation history
+ */
+export declare const clearHistory: (session: InteractiveQuerySession) => void;
+/**
+ * Get conversation history
+ */
+export declare const getHistory: (session: InteractiveQuerySession) => ConversationMessage[];

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

@@ -0,0 +1,44 @@
+import { queryOptionsSchema } from "../../../../types/query.js";
+import { executeLLM, filterScript } from "../../llm.js";
+import { buildInteractiveUserPrompt, getInteractiveSystemPrompt } from "./prompts.js";
+/**
+ * Create an interactive query session
+ */
+export const createInteractiveSession = (script, options = {}) => {
+    const validatedOptions = queryOptionsSchema.parse(options);
+    const filteredScript = filterScript(script, validatedOptions);
+    const scriptTitle = script.title || "Untitled";
+    const session = {
+        scriptTitle,
+        beatCount: filteredScript.beats.length,
+        history: [],
+    };
+    return { session, filteredScript, validatedOptions };
+};
+/**
+ * Send a question in an interactive session
+ */
+export const sendInteractiveQuery = async (filteredScript, question, session, options) => {
+    if (filteredScript.beats.length === 0) {
+        return "No content available to answer the question.";
+    }
+    const systemPrompt = getInteractiveSystemPrompt(options);
+    const userPrompt = buildInteractiveUserPrompt(filteredScript, question, session.history);
+    const answer = await executeLLM(systemPrompt, userPrompt, options, options.verbose ? `Interactive query: ${question}` : undefined);
+    // Add to history
+    session.history.push({ role: "user", content: question });
+    session.history.push({ role: "assistant", content: answer });
+    return answer;
+};
+/**
+ * Clear conversation history
+ */
+export const clearHistory = (session) => {
+    session.history = [];
+};
+/**
+ * Get conversation history
+ */
+export const getHistory = (session) => {
+    return [...session.history];
+};

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

@@ -0,0 +1,26 @@
+import type { QueryOptions, ConversationMessage } 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;
+/**
+ * Default system prompt for interactive query
+ */
+export declare const DEFAULT_INTERACTIVE_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\n- You may reference previous conversation when answering follow-up questions";
+/**
+ * Get system prompt for interactive mode
+ */
+export declare const getInteractiveSystemPrompt: (options: QueryOptions) => string;
+/**
+ * Build user prompt with conversation history for interactive mode
+ */
+export declare const buildInteractiveUserPrompt: (script: ExtendedScript, question: string, history: ConversationMessage[]) => string;