Chat mode imported from HowardMoonsFunk/AI-Agent-Mode-Template (.github/chatmodes/Custom.chatmode.md). Copyright stays with the author.

Task: Scaffold Context-Persistent Agent Mode for VSCode Project

1. Create 4 core files in repo root:

   • .github/copilot-instructions.md (inline rules)
   • prime.md (20–60 line anchor)
   • notes.md (evolving summary) --every time you summarize you should update this--
   • sessions.md (rolling log; truncate to last 25–30)

2. Add .vscode/tasks.json with:

   • Task to auto-load prime.md + summarize notes.md on workspace open
   • Task to append session summaries into notes.md/sessions.md
   • Task to reset context: reload prime.md, clear session state

3. For GPT-4.1/mini:

   • Optimize prompts to “Follow prime.md; reference notes.md.”
   • Keep anchors <2k tokens.

4. For Claude 3.5 Sonnet:

   • Add .vscode/ai_instructions.json to load multiple files
   • Use reset trigger: “Reload instructions.”

5. For local models (Qwen/DeepSeek/Llama):

   • Add memory/ folder
   • Script to embed prime.md + notes.md into vector DB (Chroma/FAISS)
   • Retrieval hook for Continue.dev to fetch relevant chunks

6. Output: Ready-to-use folder structure + VSCode configs