continual-learning

# Continual Learning (template repo) Keep durable agent memory in **local JSON**, not in root `AGENTS.md`. ## Trigger Use when the user asks to mine prior chats, maintain agent memory, or run the continual-learning loop (including stop-hook followups). ## Workflow 1. Call the `agents-memory-updater` subagent (repo override at `.cursor/agents/agents-memory-updater.md`). 2. Return the updater result verbatim. ## Target file - Memory: `.cursor/hooks/state/continual-learning-memory.json` - Schema example: `.cursor/hooks/state/continual-learning-memory.example.json` - Optional helpers: `infrastructure.core.agent_memory` ## Guardrails - Keep this skill orchestration-only — do not mine transcripts or edit files in the parent flow. - Do not bypass the subagent. - **Never** add or edit `## Learned User Preferences` or `## Learned Workspace Facts` in root `AGENTS.md` (public-repo contract). - Do not commit `continual-learning-memory.json` or `continual-learning-index.json`.