Import Project — build the vault around what already exists

Goal: end with (1) every code repo graphed and gitignored, (2) every valuable document/recording captured as a Raw source and compiled into Wiki notes, (3) a project note + topic hub that tie it together, (4) a green maintenance gate, (5) a stamped relay baton. Work incrementally and commit as you go.

Step 1 — Inventory

Have the user place (or confirm) the material in the vault root: repo folders, media files, document folders. Then:

python scripts/setup_vault.py --inventory

This writes _relay/IMPORT-INVENTORY.md (repos / media / documents / loose markdown). Review it WITH the user: confirm what to import now, what to skip, and what order matters. Don't ingest everything blindly — value first.

Step 2 — Code repos → graphify (Layer 1)

For each confirmed repo:

1. Keep outside code out of the vault's git: python scripts/setup_vault.py --gitignore-repos
2. Build its graph + browsable notes (see the graphify skill): /graphify <repo> --obsidian --obsidian-dir graphify/<repo> — or CLI: graphify <repo-path> --obsidian --obsidian-dir graphify/<repo>.
3. Install the auto-rebuild hook inside the repo: graphify hook install (re-run after any fresh clone).
4. Snapshot refresh from now on is python scripts/sync_graphs.py (part of save).
5. Convention that ties code to knowledge: compiled Wiki notes carry a ## Code Graph section linking to [[graphify/<repo>/_COMMUNITY_…]] (or node) notes. Use path-qualified links — community names collide across repos.

Step 3 — Project + topic seed

python scripts/setup_vault.py --name "<Project Name>" --topic "<domain>"

Fill the project note's Goal / Status / Next Steps from what the user tells you — this note is the hub the whole import hangs off.

Step 4 — Documents & notes → Raw sources → Wiki

For each confirmed document (per llm-wiki-ingest):

1. Capture cleaned Markdown in Raw/Sources/<date>-<slug>.md with source frontmatter. PDFs/Word docs: extract text faithfully; web pages: use the defuddle skill; images/diagrams: transcribe content to text, image itself to assets/.
2. search-catalog first; enrich existing notes over duplicating.
3. Compile short, focused Wiki notes (right folder + tag, one idea each), sources + source_count accurate, [[wikilinks]] into the topic/project.
4. Loose .md files at the vault root: either move into Raw/Sources/ with frontmatter (if they're source material) or compile their content into Wiki notes and delete the original (if they're informal notes). Nothing valuable stays unfiled at the root.

Step 5 — Recordings → video-ingest

Meetings, demos, screen captures, reference videos from the inventory's media list: run the video-ingest skill per file (background extraction, frames ⇄ transcript correlation, Raw source + compiled notes + screenshots).

Step 6 — Gate, baton, commit

python scripts/wiki_tool.py doctor
python scripts/wiki_tool.py build
python scripts/wiki_tool.py lint
python scripts/wiki_tool.py source-scan --update --accept-covered
python scripts/wiki_tool.py source-lint
python scripts/audit_public.py

Then hand off per the relay skill (STATE.md rewritten + stamped) and commit. Suggest python scripts/setup_vault.py --prune-demo once real content stands on its own.

Rules

• Never commit outside repos into the vault's git (and never commit vault tooling into theirs — use .git/info/exclude inside a repo if needed).
• Big binaries stay out: media stays where it lives (or Raw/Files/, gitignored); only curated screenshots go to committed assets/.
• Import is compilation, not relocation: the win is short linked notes + graphs, not a pile of moved files.
• Secrets: redact on capture; audit_public.py is the backstop, not the plan.