lantern-rag-dollhouse

name	lantern-rag-dollhouse
description	Build and use the single flat Lantern OS RAG dollhouse from local repos, GitHub repo metadata, literal COMET LEAP PDFs, 30-day art images, charts, convergence manifests, and future intake rules. Use when Codex needs a flat RAG-ready file, literal PDF/image assets in a skill, cross-repo Lantern/COMET LEAP consolidation, or a mass-clone/intake plan.

Lantern RAG Dollhouse

Use this skill from C:\tmp\lantern-os when the user asks for one flat local RAG file, all literal COMET LEAP PDFs/images, or cross-repo dollhouse intake.

Core Files

• Flat RAG file: references/LANTERN-OS-RAG-DOLLHOUSE.flat.md (gitignored, machine-generated — rebuild with npm run regen:rag / make regen-rag; see issue #2313)
• Asset manifest: assets/ASSET-MANIFEST.sha256
• PDFs: assets/pdfs/
• COMET LEAPER operating narrative: assets/pdfs/COMET-LEAPER-OPERATING-NARRATIVE-v1.pdf
• 30-day art images: assets/images/comet-leap-30day/
• Charts: assets/images/charts/

Intake Rule

Separate every source into one of these states:

• local_inspected: exists on disk and git/read-only state was checked.
• local_asset_copied: literal artifact copied into this skill.
• github_metadata_only: repo exists on GitHub but has not been cloned into the local dollhouse yet.
• external_llm_summary: compressed research from another LLM, not raw source.
• external_search_snippet: search result snippet or brief web finding.
• not_yet_cloned: named repo absent from local disk.
• held: needs operator approval, credentials, hardware, or destructive action.

Use skills/bayesian-world-model/SKILL.md to turn these states into priors, evidence classes, posteriors, and promote/hold/reject decisions.

Update Loop

1. Inspect git status --short --branch.
2. Inspect local repos read-only.
3. Copy only selected literal assets into assets/.
4. Rebuild assets/ASSET-MANIFEST.sha256.
5. Update references/LANTERN-OS-RAG-DOLLHOUSE.flat.md.
6. Validate skill and artifact counts.
7. Commit and push.

External LLM / Web Cache

When caching Googled information from another LLM, use:

data/rag-intake/external-llm-web-cache/
scripts/Add-ExternalRagCacheItem.ps1

Store compressed claims, source URLs, dates, rights state, evidence class, and confidence. Do not store raw article dumps or long copyrighted passages.

Purge Lane (post-ingestion)

After Sync-RagAndPdf.ps1 ingests markdown into the flat RAG file, the derived PDFs in reports/PDF/ become re-generatable and can be aged out. Run scripts/Invoke-PostIngestionPurge.ps1 (batch job post-ingestion-purge, interval 245 min, runAfter: sync-rag-pdf) to enforce:

Target	Rule
reports/PDF/*.pdf	Delete if older than 7 days (re-derivable from markdown)
manifests/evidence/*.json/.md	Keep newest 30 per class prefix, drop the rest
data/kalshi/*-2026-*-*.json	Delete dated snapshots older than 7 days (superseded by -latest.json)

Status Cube row (purge lane):

x	y	z	t
reports/PDF/, manifests/evidence/, data/kalshi/	purge lane	re-derivable only; never purge source markdown or -latest.json	receipt written to manifests/evidence/purge-{stamp}.json

Boundaries

Do not pretend metadata-only repos have been cloned. Do not mutate source repos while building the dollhouse. Do not import dirty source state blindly. Do not purge source markdown — only derived PDFs and superseded snapshots.