From memory-recall
Run memory consolidation. Use when the user asks to consolidate, organize, or clean up memories, or says /dream. Synthesizes recent session context into durable, well-organized memory files.
npx claudepluginhub t2ance/memory-recall-pluginThis skill uses the workspace's default tool permissions.
Designs and optimizes AI agent action spaces, tool definitions, observation formats, error recovery, and context for higher task completion rates.
Implements structured self-debugging workflow for AI agent failures: capture errors, diagnose patterns like loops or context overflow, apply contained recoveries, and generate introspection reports.
Compares coding agents like Claude Code and Aider on custom YAML-defined codebase tasks using git worktrees, measuring pass rate, cost, time, and consistency.
An automated memory_curator runs every 4h on the Stop hook, performing aggressive MERGE/DELETE consolidation. This manual Dream skill is for immediate consolidation or for consolidation that requires user judgment (e.g., moving entries between project and global memory, tidying CLAUDE.md).
You are performing a dream -- a reflective pass over your memory files and instructions. Synthesize what you've learned recently into durable, well-organized memories so that future sessions can orient quickly.
You manage three sources:
${CLAUDE_PLUGIN_DATA}/global-memory/. Stores context that applies across all projects. Create this directory if it doesn't exist. You can read and write freely.~/.claude/CLAUDE.md. The user's persistent directives. You can read freely, but MUST ask the user via AskUserQuestion before making any changes.Session transcripts are JSONL files in the project directory (large files -- grep narrowly, don't read whole files).
ls both memory directories to see what already existsMEMORY.md indexes~/.claude/CLAUDE.mdLook for new information worth persisting from the current project. Sources in rough priority order:
logs/YYYY/MM/YYYY-MM-DD.md) if presentFor each thing worth remembering, decide where it belongs based on its content, not its type label:
Use the memory file format (frontmatter with name/description/type) from your system prompt's auto-memory section.
Focus on:
Review ~/.claude/CLAUDE.md for entries that are memory-like rather than instruction-like:
If you find entries that should move:
Also check the reverse: if global memory contains entries that are actually hard directives ("never do X", "always do Y"), propose moving them to CLAUDE.md via AskUserQuestion.
Never edit CLAUDE.md without asking first.
Update MEMORY.md in both memory directories. Each should stay under 200 lines and ~25KB. Each entry should be one line under ~150 characters: - [Title](file.md) -- one-line hook.
Return a brief summary of what you consolidated, updated, or pruned in each location. If nothing changed, say so.