Dream

Overview

Dream handles the bookkeeping of the Datarim knowledge base: cross-references, deduplication, organization, staleness checks, and structural consistency. It runs periodically or on demand via /dr-dream, keeping the knowledge base healthy as it grows. Inspired by the LLM Wiki pattern where the tedious part of a knowledge base is not the thinking — it's the maintenance.

Three Operations

• Ingest — process new items. Finds misplaced files, missing task IDs in filenames, broken links, archives without matching reflections, and proposes corrections.
• Lint — periodic health check. Detects contradictions, stale content, orphan files, duplicate content (>70% overlap), oversized files (>500 lines), inconsistent naming, and broken cross-references.
• Consolidate — deep maintenance. Merges duplicates, extracts recurring patterns from reflections, builds or updates the index, syncs progress tracking, and adds bidirectional links between related documents.

Contradiction Handling

When two documents make conflicting claims, Dream does not silently overwrite. It adds a [!contradiction] callout to both documents, logs the conflict, and leaves resolution to the human. Both claims may be valid in different contexts.

When It's Used

The librarian agent runs Dream via /dr-dream (full maintenance), /dr-dream lint (quick check), or /dr-dream index (rebuild index). It is auto-suggested after archiving if more than 5 documents are unindexed, or after every 10 completed tasks.

What Dream Does NOT Do

• Modify source code — only datarim/ and documentation/archive/ content
• Delete documents without approval — all removals are proposed, never automatic
• Resolve contradictions — it flags them for human judgment
• Create new knowledge — it organizes existing knowledge only