dream

Dream: Auto-Memory Consolidation

Seven operators, each defined in operators/. This file orchestrates the four-phase cycle. Read operator files only when executing Phase 3.

When to Use

• SessionStart hook nudges via hooks/lib/dream-gate.js when 24+ hours have passed since the last dream AND 5+ memory files have been modified since then. Nudge only — never auto-runs.
• Stop hook nudges after heavy sessions (3+ new memory files in current session).
• Manual: /dream runs immediately, ignores gates.

Provenance

Emit events silently via Bash for each operator action.

node "${CLAUDE_PLUGIN_ROOT}/scripts/provenance-emit.js" '{"agent":"dream","skill":"dream","action":"ACTION","target":"FILENAME"}'

Where ACTION is one of: merge, resolve, abstract, compress, prune, link, normalize.

At start: {"action":"session-start"}. At end: {"action":"session-end","merged":N,"resolved":N,"abstracted":N,"compressed":N,"pruned":N,"linked":N,"normalized":N} + run node ${CLAUDE_PLUGIN_ROOT}/scripts/provenance-consolidate.mjs.

Phase 1: Orient

1. Detect the project memory directory:

   • Use $CLAUDE_PROJECT_DIR if available, else use the auto-memory directory for the current project
   • Verify the directory exists and contains MEMORY.md

2. Read all .md files (excluding MEMORY.md, _dream_log.md, _archived/).

3. Parse YAML frontmatter: name, description, type, confidence.

4. Build inventory: total count, count by type, sorted by modification date, line count per file.

5. Read MEMORY.md. Check links resolve to actual files. Flag orphaned pointers.

6. Report:

   Dreaming: [project name]
   Memory files: N (N feedback, N project, N user, N reference)
   Index entries: N (N orphaned)
   

Phase 2: Gather Signal

1. Group by type, sort newest-last within each group. Order: feedback, user, project, reference. Within each group, oldest first: transformer attention favors recent tokens, so placing the memories you want the consolidator to weight most heavily last in each group keeps them in the recency-favored position.