agentprivacy-wiki-dream-cycle
The dream cycle is how the agentprivacy guide stays rebound to the universe
across /mitch. Where agentprivacy-wiki-resync is the per-turn operation
("I changed a repo → rebuild the wiki"), the dream cycle is the discovery layer
that runs before you know what to change: every current repo dreams about
itself, surfaces its gaps, and proposes what to fill — and the First Person
decides.
It is the Chronicler's loop (it produces chronicles) cataloged by the Librarian
(the wikis/ keeper). The dream files are standing ledgers that travel with
each repo, so the next cycle resumes from them.
The four stages (looped)
1. RESEARCH scan the source dirs → diff vs the live FedWiki → gap report
2. GATE the First Person reads the dreams and ticks what to fill ← the only human step
3. BUILD builders write the approved canon into ~/.wiki (journal-versioned, [[linked]], hub backlinks)
4. SNAPSHOT tools/snapshot.mjs → ./site → review on :3200
↺ loop until comfortable → push → guide.agentprivacy.ai
(the live ~/.wiki stays the editable, tunnel-able carry copy)
Two layers of dream
Thematic dreams (
flow/dream/): cross-cutting passes by site/topic — grimoire+research, tomes/cast/city, skills/personas, atlas/graph, library/narrative, and the relevance-curated wider-universe pass that triages the ~80/mitchdirs into current · stale · irrelevant so the fan-out never dreams on cruft.Per-directory dreams (
<repo>/chronicles/DREAM-<date>.md): each current repo dreams about its own state + its link to the guide. This is the daily work-ledger. Template:# Dream — <DirName> — <date> ## What this is ## Current state & open threads (within the dir) ## Guide coverage (projected vs missing) | site | item | status | source | ## Suggestions (prioritized) | suggestion | target | why | H/M/L | ## To get rolling tomorrow (ordered next actions) ## Notes & uncertainties
Discipline (the keeper's rules)
- Relevance first. Never treat a stale/duplicate/version-named dir
(
v4…,nov11…,…copy,…fork) or a system folder as a gap. Triage, then verify each surviving candidate against the live sites before calling it a gap. - No duplicate information flows. Source canon → one guide projection. Fill what's missing; link, don't re-create, what already lives elsewhere (sibling hosts, other sites). Resolve collisions as cleanups, not new dupes.
- The dreams are research input, not content to paste. They drive what canon to build; they stay as the working ledger.
- Human gate is load-bearing. Nothing is built from a dream without the First Person ticking it; nothing is pushed/published without an explicit ask.
- Honest framing carries. e.g. estimation is not attack; vendored work is not the author's — every dream preserves the source's honesty.
Run it
node flow/run.mjs audit # RESEARCH → flow/gap-report.json
# (read flow/GUIDE-DREAM-SUGGESTIONS.md — the GATE — and decide)
node flow/run.mjs build <site> # BUILD the approved gaps into ~/.wiki
node flow/run.mjs snapshot # SNAPSHOT → ./site → review on :3200
node flow/run.mjs cycle <site> # audit → build → snapshot → audit, in one
The fan-out of per-directory dreams is driven from the relevance ledger; refresh
it when the /mitch layout changes, then dream each current repo. See
flow/DREAM-CYCLE.md for the full convention and flow/builders/ for the
per-site builders (grimoire and research are the worked examples).
Lineage
Born 2026-06-29 building the guide.agentprivacy.ai static snapshot, when the
question shifted from "patch this gap" to "keep the guide standing-rebound to the
whole universe." The Chronicler writes the dreams; the Librarian federates them;
the First Person decides.