跳轉至

V12 Changelog

[12.1.2] - 2026-01-14 - Honesty & Hardening 🛡️

🛡️ Quality & Integrity

  • Documentation Honesty (100%): Deep audit of all manuals to ensure 1:1 match with implementation.
  • Codebase Hygiene: Removal of experimental artifacts and ghost features.
  • Verification: Validated boring verify --level FULL and uv/smithery config paths.

🚀 Feature Acceleration

  • Allow-List Skill Installation: boring_skills_install now supports trusted URLs (GitHub, SkillsMP).
  • Universal Skills System (BUSS): Cross-platform loader for .boring/, .gemini/, .claude/ skills. New MCP tools: boring_skill_discover, boring_skill_activate, boring_skill_create, boring_skill_download. Auto-sync enabled.
  • Workspace Hygiene: Added boring clean command to safely remove ephemeral files (.boring/, .boring_memory/) and restore a pristine project state.

[12.0.0] - 2026-01-14 - The True One Dragon Update 🐉🧠

🚀 Major Features

  • One Dragon Flow (V12.0.0): A comprehensive, state-machine driven autonomous development loop. Includes Architect, Builder, Healer, Polish, and Evolver stages.
  • Cognitive Reflex: Integrated semantic search (RAG) directly into the Brain for high-speed fuzzy error correction.
  • Global Swarm Sync: Real-time knowledge synchronization via Git across multiple project instances.

🔧 Fixes

  • CLI Compatibility: Fixed "Requested entity was not found" error for Keyless OAuth users (Gemini CLI).
  • Import Precision: Fixed boring_checkpoint imports in HealerNode.
  • Node Stability: Fixed NameError for boring_speckit_tasks in engine.py by ensuring proper fallback initialization for all optional SpecKit tools.

🛡️ Quality & Stability

  • Tool Count Alignment: Synchronized all documentation and router metadata to reflect verified tool counts (67+ standard, 43 lite).
  • Timeout Protection: Implemented a 1-hour global timeout for AgentLoop within the FlowEngine to prevent perpetual hangs.
  • Integrity Audit: Purged "ghost feature" references to VectorMemory and AutoLearner from all documentation and guides.
  • Standardized Results: 100% of tools now return the BoringResult TypedDict for better LLM reasoning.