🧠 v2.6.0 - Compaction-Resilient Memory & Structured Task Tracking
Game-changing release: Claude never forgets critical project knowledge, even through context compaction.
🚀 Major Features
Three-Tier Memory System
Persistent notepad system that survives compaction:
- Priority Context - Critical discoveries (API URLs, key files) always loaded on session start
- Working Memory - Session notes with timestamps, auto-pruned after 7 days
- MANUAL - User permanent notes, never pruned
Usage:
# Agents can persist discoveries automatically
<remember>Project uses pnpm not npm</remember>
<remember priority>API client at src/api/client.ts</remember>
# Or save notes manually
/note Database schema uses PostgreSQL with Prisma ORMRalph Loop PRD Support
Structured task tracking inspired by Ralph:
- Product Requirements Document (PRD) format with user stories
- Progress tracking with learnings and patterns
- Completion guarantee - loop continues until ALL stories pass
New Commands:
/ralph-init <task>- Initialize PRD for structured task tracking/ralph-loop <task>- Self-referential loop until completion/ultrawork-ralph <task>- Maximum intensity + completion guarantee/ultraqa <goal>- Autonomous QA cycling workflow/note <content>- Save notes to notepad.md
📦 What's Changed
Added
- Notepad Memory System - Compaction-resilient context persistence
- Remember Tag Auto-Capture - PostToolUse hook for
<remember>tags - PRD Support - Structured task tracking with user stories
- Progress Memory System - Append-only progress log with learnings
- New Commands - /ralph-init, /ultrawork-ralph, /ultraqa, /cancel-ultraqa, /note
- Comprehensive Test Suite - 358 tests (up from 231)
Changed
- Ralph Loop Enhanced - Auto-initializes PRD, PRD-based completion
- Persistent Mode Integration - PRD completion checking
- Installer Enhanced - PostToolUse hook installation
Fixed
- Stale position bug in addPattern
- Type safety in createPrd
- Recursion guard in addPattern
- Todo-continuation infinite loop - Max-attempts counter (5)
- UltraQA/Ralph-Loop conflict - Mutual exclusion
- Agent name prefixing - Standardized Task() calls
- VERSION constant mismatch - Fixed installer VERSION to 2.6.0
📊 Statistics
- 19 Specialized Agents (+ tiered variants)
- 19 Lifecycle Hooks (added notepad, post-tool-use)
- 21 Builtin Skills (added 13 new skills)
- 358 Tests (127 new tests)
See CHANGELOG.md for full details.
🔗 Links
Full Changelog: v2.5.0...v2.6.0