Ace-Context-to-Memory

Converts source documents to markdown and chunks them for agent memory. Pipeline: convert (documents → markdown + images) → chunk (markdown → smaller files) → embed (vector index for semantic search).

When to Activate

• User asks to "add content to memory" or "refresh memory"
• Wants to convert a folder of documents (PPTX, PDF, DOCX, XLSX) for agent context
• Mentions "convert and chunk", "ingest for agent", or "memory pipeline"
• Has added new files and wants them processed
• "Use memory" / semantic retrieval: User says "use memory", "search memory", "what does memory say about X", "from our content", "from ABD materials", "what do we have on [topic]" — run search_memory "<query>" and inject results into your response
• Adding content from any folder: When adding a folder's content to memory, run link_workspace_source.py --path <folder> first (on request) if the link does not yet exist

CRITICAL: Respect User Scope

• One file: When user says "one file", "this file", "just X.pdf", or names a specific file → use --file <path>. Process ONLY that file.
• Folder: When user says "folder", "all", "everything in X", or explicitly requests a folder → use --memory <path>.
• Do NOT process entire folders when the user asked for a single file.