contorium-workspace-memorylisted

# Contorium workspace memory ## When to use - Starting a new chat or switching models mid-task - User asks to continue where they left off - Exporting compact, AI-ready project context - Detecting session pollution after a task change ## Memory layers | Layer | Source | Lifetime | |-------|--------|----------| | Workspace | `.contora/state.json` (focus, files, Git, notes) | Long | | Session | in-memory events + optional JSONL per `sessionId` | Short | | Cognitive | `.contora/last-intent.json` (BYOK intent + lifecycle) | Medium | ## Instructions 1. Prefer **Current focus** (optional) for user intent; never block export if empty. 2. For clipboard handoff, use **Copy AI-ready context** — not raw event dumps. 3. If focus and active files diverge sharply, suggest **Start fresh AI context session**. 4. When AI intent is stale (low confidence), rely on heuristic operational intent from recent edits. 5. Respect ignore rules (`.contoraignore`, `contora.extraIgnoreSubstrings`) when reasoning about ranked files. ## MCP tools (Codex / Claude Code / Cursor Agent) When the Contorium MCP server is connected (`npm run build:mcp`, see `docs/MCP.md`): - `get_workspace_context` — read extension snapshot from `.contora/state.json` - `store_memory` / `search_memory` / `get_memory` — agent-persisted notes under `.contora/mcp/memories.json` Use MCP for agent-driven recall; use extension commands for sidebar UI and clipboard export. - `contora.exportAIContext` — copy AI-ready context - `c