wiki-dreamlisted

# wiki-dream While you sleep, the wiki re-evaluates which old pages have become relevant again. The dreamer reads new ingest entries from `log.md` since the last dream watermark, identifies entities/tags/citations they bring up, and scores every frozen/archived page against that signal. Pages above the threshold become **candidates** — surfaced for your review, never auto-promoted by default. > Auto-dreaming is the only thing in kata that runs without you. It is > intentionally conservative: dry-run by default, capped at 10 candidates > per run, and silently rejecting bad suggestions costs nothing — they > just don't appear next time. ## Implementation The algorithm is in `plugin/scripts/wiki_dream.py`. **The script is the source of truth; the prose below explains its behavior.** The skill calls the script and renders the result. ```bash # Default — read watermark from log.md, write candidates to dreaming/{date}.md python {plugin_root}/scripts/wiki_dream.py --wiki {wiki_path} # Override window — dream over a specific period python {plugin_root}/scripts/wiki_dream.py --wiki {wiki_path} \ --since 2026-04-01 # Explain why one specific page did or didn't qualify python {plugin_root}/scripts/wiki_dream.py --wiki {wiki_path} \ --explain mosaic # Apply — write tier_override:active to selected candidates python {plugin_root}/scripts/wiki_dream.py --wiki {wiki_path} \ --apply --pages 1,2,3 ``` The candidate file is written to `dreaming/{YYYY-MM-DD}.md` (one per ru