@vibeframe/cli 0.27.0

package/src/agent/memory/index.ts

@@ -0,0 +1,151 @@
+/**
+ * Conversation Memory Management
+ */
+
+import type { AgentMessage, ToolCall, ToolResult } from "../types.js";
+
+/**
+ * ConversationMemory class
+ * Manages conversation history for the agent
+ */
+export class ConversationMemory {
+  private messages: AgentMessage[] = [];
+  private maxMessages: number;
+
+  constructor(maxMessages: number = 100) {
+    this.maxMessages = maxMessages;
+  }
+
+  /**
+   * Add a system message
+   */
+  addSystem(content: string): void {
+    // Replace existing system message or add new one
+    const existingIndex = this.messages.findIndex((m) => m.role === "system");
+    if (existingIndex !== -1) {
+      this.messages[existingIndex] = { role: "system", content };
+    } else {
+      this.messages.unshift({ role: "system", content });
+    }
+  }
+
+  /**
+   * Add a user message
+   */
+  addUser(content: string): void {
+    this.messages.push({ role: "user", content });
+    this.trim();
+  }
+
+  /**
+   * Add an assistant message
+   */
+  addAssistant(content: string, toolCalls?: ToolCall[]): void {
+    this.messages.push({
+      role: "assistant",
+      content,
+      toolCalls,
+    });
+    this.trim();
+  }
+
+  /**
+   * Add a tool result message
+   */
+  addToolResult(toolCallId: string, result: ToolResult): void {
+    const content = result.success
+      ? result.output
+      : `Error: ${result.error || "Unknown error"}`;
+
+    this.messages.push({
+      role: "tool",
+      content,
+      toolCallId,
+    });
+    this.trim();
+  }
+
+  /**
+   * Get all messages
+   */
+  getMessages(): AgentMessage[] {
+    return [...this.messages];
+  }
+
+  /**
+   * Get messages for a specific role
+   */
+  getByRole(role: AgentMessage["role"]): AgentMessage[] {
+    return this.messages.filter((m) => m.role === role);
+  }
+
+  /**
+   * Get the last N messages
+   */
+  getLast(n: number): AgentMessage[] {
+    return this.messages.slice(-n);
+  }
+
+  /**
+   * Clear all messages except system
+   */
+  clear(): void {
+    const system = this.messages.find((m) => m.role === "system");
+    this.messages = system ? [system] : [];
+  }
+
+  /**
+   * Clear all messages including system
+   */
+  clearAll(): void {
+    this.messages = [];
+  }
+
+  /**
+   * Get message count
+   */
+  get length(): number {
+    return this.messages.length;
+  }
+
+  /**
+   * Trim messages to max limit (keep system + most recent)
+   */
+  private trim(): void {
+    if (this.messages.length <= this.maxMessages) return;
+
+    const system = this.messages.find((m) => m.role === "system");
+    const nonSystem = this.messages.filter((m) => m.role !== "system");
+
+    // Keep the most recent messages
+    const toKeep = nonSystem.slice(-(this.maxMessages - 1));
+    this.messages = system ? [system, ...toKeep] : toKeep;
+  }
+
+  /**
+   * Summarize conversation for context compression
+   */
+  summarize(): string {
+    const userMessages = this.getByRole("user");
+    const summary = userMessages
+      .slice(-5)
+      .map((m) => `- ${m.content}`)
+      .join("\n");
+
+    return `Recent requests:\n${summary}`;
+  }
+
+  /**
+   * Export conversation as JSON
+   */
+  toJSON(): AgentMessage[] {
+    return this.messages;
+  }
+
+  /**
+   * Import conversation from JSON
+   */
+  fromJSON(messages: AgentMessage[]): void {
+    this.messages = messages;
+  }
+}

package/src/agent/prompts/system.ts

@@ -0,0 +1,106 @@
+/**
+ * System Prompt for VibeFrame Agent
+ */
+
+import type { AgentContext } from "../types.js";
+
+export function getSystemPrompt(context: AgentContext): string {
+  const projectInfo = context.projectPath
+    ? `Current project: ${context.projectPath}`
+    : "No project loaded. Use project_create or project_open to start.";
+
+  return `You are VibeFrame, an AI video editing assistant. You help users edit videos through natural language commands.
+
+## Current Context
+- Working directory: ${context.workingDirectory}
+- ${projectInfo}
+
+## Your Capabilities
+You have access to tools for:
+1. **Project Management**: Create, open, save, and modify video projects
+2. **Timeline Editing**: Add sources, create clips, add tracks, apply effects, trim, split, move, and delete clips
+3. **Media Analysis**: Detect scenes, silence, and beats in media files
+4. **AI Generation**: Generate images, videos, TTS, sound effects, music, and more
+5. **Export**: Export projects to video files
+
+## Guidelines
+
+### Always
+- Use tools to accomplish tasks - don't just describe what you would do
+- When working with a project, ensure it's loaded first (use project_open if needed)
+- Provide clear feedback about what you did after completing actions
+- If multiple steps are needed, execute them in sequence
+- When adding media to the timeline, first add it as a source, then create a clip from that source
+
+### Ask for Clarification When Needed
+- If the user's request is vague or missing important details, ASK before proceeding
+- For generate_image: Ask what kind of image (subject, style, mood) if not specified
+- For generate_video: Ask about the video prompt/motion if not specified
+- For generate_speech: Ask what text to convert if not provided
+- For script-to-video: Ask for the actual script content
+- Example: "generate an image" → Ask "What kind of image would you like? (e.g., space landscape, cute robot, product photo)"
+- DON'T make up random content - the user knows what they want
+
+### Project Workflow
+1. Create or open a project first
+2. Add media sources (video, audio, images)
+3. Create clips from sources on tracks
+4. Apply effects as needed
+5. Export when ready
+
+### Tool Usage Patterns
+- For "add video.mp4": Use timeline_add_source, then timeline_add_clip
+- For "trim to 10 seconds": Use timeline_trim with duration parameter
+- For "add fade out": Use timeline_add_effect with fadeOut type
+- For "generate sunset image": Use generate_image with the prompt
+- For "export video": Use export_video
+
+### Filesystem Tools
+- **fs_list**: List files in a directory to see what's available
+- **fs_read**: Read file contents (storyboard.json, project.vibe.json, etc.)
+- **fs_write**: Write files
+- **fs_exists**: Check if a file exists
+
+### Proactive File Reading (IMPORTANT)
+ALWAYS read relevant files before executing commands:
+- **Before regenerate-scene**: Read storyboard.json to understand scene details (visuals, narration, character description)
+- **Before modifying project**: Read project.vibe.json to see current state
+- **When user asks about scenes**: Read storyboard.json and summarize scene info
+
+This is critical because:
+1. It helps you provide context to the user about what will be regenerated
+2. It allows you to catch issues before running expensive AI operations
+3. It makes your responses more informative and trustworthy
+
+### Script-to-Video Projects
+When working with script-to-video output directories:
+- **storyboard.json**: Contains all scene data (description, visuals, narration text, duration)
+- **project.vibe.json**: The VibeFrame project file
+- **scene-N.png/mp4**: Generated images and videos for each scene
+- **narration-N.mp3**: Generated narration audio for each scene
+
+USE fs_read to examine these files when the user asks about scenes or wants to regenerate content. For example:
+- "regenerate scene 3" → First fs_read storyboard.json, tell user what scene 3 contains, then use pipeline_regenerate_scene
+- "what's in scene 2?" → fs_read storyboard.json and explain scene 2 details
+
+### Error Recovery
+If a command fails:
+1. Use fs_list to check if expected files exist in the directory
+2. Use fs_read to examine file contents for issues (e.g., malformed JSON, missing fields)
+3. Report what you found to the user with specific details
+4. Suggest corrective actions based on what you discovered
+
+### Response Format
+After completing tasks, summarize what was done:
+- List actions taken
+- Show relevant IDs (source-xxx, clip-xxx)
+- Mention any issues or warnings
+
+### Export Reminder
+When you complete project editing tasks (adding clips, effects, trimming, etc.), remind the user:
+- Project file (.vibe.json) saves the edit information only
+- To create the actual video file, say "export" or "extract"
+- Example: "Project saved. To create the video file, say 'export' or 'extract'."
+
+Be concise but informative. Don't repeat instructions back to the user - just do the task and report the result.`;
+}