ralph-referencelisted

# Ralph Reference (lazy-loaded) Reference material moved out of the always-loaded `CLAUDE.md` to save resident context. The operational rules (Aristotle, config location, security hooks, agent teams, quality gates, repository isolation, model routing, memory system) remain in `CLAUDE.md`. ## Batch Task Execution (v2.88.0) New skills for autonomous multi-task execution: | Skill | Purpose | Usage | |-------|---------|-------| | `/task-batch` | Execute task lists autonomously | `/task-batch docs/prd/file.prq.md` | | `/create-task-batch` | Interactive PRD creator | `/create-task-batch "Feature name"` | **Key Features**: - Handles MULTIPLE tasks (not single task) - MANDATORY completion criteria per task - VERIFIED_DONE validation guarantee - Fresh context per task execution - Auto-commit after each completed task See: `docs/batch-execution/BATCH_SKILLS_v2.88.0.md` ## Skills/Commands Unification (v2.87.0) **Unified Skills Model**: All commands now use the SKILL.md format per Claude Code best practices. | Location | Purpose | Format | |----------|---------|--------| | `.claude/skills/<name>/SKILL.md` | Source of truth (repo) | Skill directory | | `~/.claude/skills/<name>` | Symlink to repo | Symlink | **Key Changes**: - Removed duplicate command files from `~/.claude/commands/` - All Ralph skills symlinked from global to repo - Single source of truth: changes in repo reflect globally - Version aligned to v2.87.0 across all skills See: `docs/architecture/UNIFIED_ARCHITECTU