autoresearch-system

# Autoresearch System (Architecture-Level Karpathy Loop) ## When to use - A nightly scheduled self-improvement run - Manual trigger: "optimize system", "autoresearch system" - After major architecture changes to validate everything still works ## Test Scenarios Simulate these end-to-end workflows against the current config: ### Scenario 1: Basic inbound message - The user sends a text message via the chat channel - Expected: the default persona responds, tone matches, response via the chat reply tool ### Scenario 2: Intent routing - A message contains signal-related keywords ("just saw this alpha on a protocol") - Expected: routes to `knowledge/research.md`, invokes the signal-scorer skill logic ### Scenario 3: Persona switch + briefing - The user sends "/persona coach" - Expected: switch to coach, load `memory/personas/coach.json`, respond with the coach greeting + context - Note: only the active personas are switchable. Deprecated personas are archived. If the test input is a deleted persona, the expected behavior = inform the user the persona doesn't exist, stay in the default persona. ### Scenario 4: Driving mode activation - The user sends a voice message saying "I'm driving" - Expected: driving mode activates, the response includes a TTS voice note, uses the default voice ### Scenario 5: Knowledge pipeline flow - A substantive output is produced (analysis, draft, plan) from an **interactive session** - Expected: auto-saves to `outputs/raw/{persona}/`, correct fr...