mirror of
https://github.com/ruvnet/RuVector.git
synced 2026-05-24 05:43:58 +00:00
0af0ffe8ca
Implements a complete daily learning and optimization loop for RuVector: ## GOAP Planner - A* search-based Goal-Oriented Action Planning - 8 default actions: scan_codebase, analyze_patterns, compute_novelty, log_discovery, submit_to_pi, consolidate_sona, refresh_connection, reason_with_gemini - LearningWorldState with 20+ state variables - Goal conditions: Equal, GreaterOrEqual, LessThan ## Discovery System - CodebaseScanner for pattern discovery across multiple languages - PatternAnalyzer with signature matching for common patterns - DiscoveryLog with full method provenance and tool attribution - QualityAssessment with novelty, usefulness, clarity, correctness, generalizability scores ## Integration - PiRuvIoClient for π.ruv.io cloud brain submission with retry logic - GeminiGoapReasoner for GOAP planning using Gemini 2.5 Flash - SecretManager for Google Cloud secrets (via gcloud CLI) - Submission queue with exponential backoff ## Scheduler - DailyLearningScheduler with scheduled and opportunistic triggers - Trigger detection: idle period, git commits, trajectory buffer - Configurable quality thresholds and scan parameters ## Consolidation - SonaConsolidator with EWC++ integration - Pattern pruning and memory pressure management - Cosine similarity for pattern search ## ADR & Design - ADR-115: Daily Learning and Optimization Loop architecture - Comprehensive GOAP design document with milestones Co-Authored-By: claude-flow <ruv@ruv.net> |
||
|---|---|---|
| .. | ||
| daily-learning-loop-goap-design.md | ||