Skip to content

RECIPES.md

Latest commit

 

History

History
597 lines (414 loc) · 24.7 KB

RECIPES.md

File metadata and controls

597 lines (414 loc) · 24.7 KB

Things to Try

A recipe book. Pick one. Each entry takes something you probably already have, walks you through what to do with it, and ends with a tangible artifact — a page in your vault, a CSV you can open, an HTML map, a memo, a publishable bio.

Designed so you can scan this list during a coffee break, pick one, and have something to show by lunch.

Convention: every recipe lists Input → Tool → Skill → Output. If a step has nothing in it (e.g., the recipe is skill-only), it'll say "—".

How to use this list

Pick a recipe based on what you already have on disk, not on what skill sounds cool. The fastest path to "ohhh, this is actually useful" is to put your own data through one full cycle.

The recipes are grouped by input type, then a final section by goal if you want to work backwards from "I need to ship a bio."

If you get stuck mid-recipe: run /vault-lint to confirm the system is healthy, then re-read the step you got stuck on. Most failures are skipped steps.

Photos

You probably have the densest archive here — phone, camera, screenshots, Drive, iCloud. Photos carry timestamps, locations, faces, and event context. They turn the vault from "things I wrote" into "things I lived."

Recipe 1 — Map every place you've been

Goal: A CSV of every location you've taken a photo, ready for a map view.
Input A folder of photos with EXIF data (iPhone dump, camera SD, Google Photos export)
Tool photo-processor/scripts/extract_locations.py
Skill —
Time 15 min

Steps:

  1. Drop your photo folder into tools/photo-processor/new/.
  2. Run: python3 scripts/extract_locations.py
  3. Open the output: <organized-dir>/locations.csv. Each row is one photo with lat/lon/date.

What you get: locations.csv — drag this into kepler.gl or umap.openstreetmap.fr for a free, browser-based interactive map of your life in dots. No accounts, no upload limits for normal-sized exports.

Stretch (cohort bonus): the location-analysis/ pipeline turns this CSV into 15+ vault pages — decade maps, travel extremes, biome breakdowns, "where was I on 2014-06-12," art pilgrimage routes. Reach out if you want a copy.

Recipe 2 — Sort the camera roll into events

Goal: Stop scrolling through 40,000 photos. See your life as event clusters.
Input A messy photo folder (10K+ photos works best)
Tool photo-processor/scripts/event_cluster.py
Skill /diarize (optional, after)
Time 30–60 min for the run; 15 min review

Steps:

  1. Run: python3 scripts/event_cluster.py /path/to/photos
  2. The tool clusters photos by date + location proximity. Output is folders like 2024-06-15 — Brooklyn — 47 photos.
  3. Open one cluster you remember. Confirm the system got the event right.
  4. (Optional) Run /diarize <event name> to build a canon page about that event using everything in your vault that mentions it.

What you get: Your photo archive becomes a chronology of events instead of a chronological dump. Plus a diarized page that ties the photos to whoever was there, what was happening, and what came next.

Recipe 3 — Extract every screenshot you've taken

Goal: A separate folder of just screenshots — usually 30%+ of a phone's photos. These are mostly recipes, plane tickets, books, screenshots of articles, and ideas you forgot you saved.
Input Phone photo dump
Tool photo-processor/scripts/extract_screenshots.py + screenshot_report.py
Skill /ingest (optional, on the report)
Time 20 min

Steps:

  1. python3 scripts/extract_screenshots.py /path/to/photos
  2. python3 scripts/screenshot_report.py — generates a Markdown report of what's in there (categorized: travel, food, reading, ideas, work, etc.)
  3. Drop the report into 01_INBOX/To Process/ and run /ingest.

What you get: A canon page of "things I've cared enough to screenshot." Often surprising — surfaces themes you didn't know you were obsessed with. Real example from my vault: I had 280 screenshots of book covers I never bought.

Recipe 4 — The face/place graph

Goal: Photos linked to canon pages of people and places. So a person page automatically shows up with photos of them.
Input Photo dump + your vault with some 04_CANON/Shared/People/ pages
Tool photo-processor/scripts/detect_faces.py + entity_extractor.py
Skill /enrich
Time 1 hr (90% the script run; 10 min curation)

Steps:

  1. python3 scripts/detect_faces.py /path/to/photos — clusters faces.
  2. Manually label the top 10 face clusters with a person's name (one-time annotation cost).
  3. python3 scripts/entity_extractor.py — joins face clusters with canon people pages.
  4. Run /enrich [Person Name] on someone you have lots of photos of — adds a "Photos" section to their canon page with paths.

What you get: A person's canon page now lists every photo of them in your archive, dated. Powerful for memoir / bio work, eulogies (sorry), or just remembering when you last hung out.

Email, Calendar, Contacts (Google Takeout)

This is the highest-ROI ingest. Takeout is huge but it's the receipt of who's been in your life and when.

Recipe 5 — Build your real contact list

Goal: A single, deduped, prioritized contact list. Not Gmail's auto-collected mess — your real people.
Input Google Takeout (Mail + Contacts + Calendar)
Tool takeout-processor (full pipeline)
Skill —
Time 2–3 hours (most is the script running)

Steps:

  1. Request Google Takeout for Mail, Contacts, Calendar. Wait for the email (~1 hour).
  2. Drop the resulting .zips into tools/takeout-processor/00_raw_takeouts/.
  3. Run the pipeline: 
    python3 05_scripts/mbox_extract_bodies.py
python3 05_scripts/build_master_csv.py
python3 05_scripts/match_contacts.py
python3 05_scripts/build_relationship_graph.py
  4. Open 03_master_csvs/relationship_graph.csv — sorted by message volume + recency.

What you get: A CSV of every person you've actually corresponded with, ranked by how-much-they-matter (proxied by message count, calendar density, recency). You'll be surprised who's at the top. Ingest the CSV into your vault and you can /diarize <name> on anyone.

Real example: from my Takeout, I rebuilt a 626-person collector list with $410K of reconstructed sales — half of whom I'd forgotten about. That CSV is now the spine of my collector outreach work.

Recipe 6 — Ask the email archive a question

Goal: Treat 20 years of email as a searchable knowledge base.
Input Takeout-processed mbox
Tool takeout-processor/05_scripts/mbox_extract_bodies.py
Skill /ingest then /diarize
Time 3 hours initial setup; instant after

Steps:

  1. Run mbox_extract_bodies.py to convert the mbox into per-message Markdown files.
  2. Cherry-pick the 50 most-relevant emails on a topic (e.g., "every email about my 2014 show in Berlin").
  3. Drop them in 01_INBOX/To Process/. Run /ingest.
  4. Run /diarize "2014 Berlin show".

What you get: A canon page about that show with every email cited, dated, and the people involved cross-linked to their canon pages. The email archive becomes infrastructure instead of a graveyard.

Press, articles, URLs

If you've ever been in print, this is the recipe that makes you stop losing track.

Recipe 7 — Your press, archived forever

Goal: A folder of every article ever written about you, locally cached so it survives when the original goes 404.
Input A list of press URLs (from your bio file, Wikipedia, Google searches)
Tool press-ingest
Skill /ingest
Time 30 min for 50 articles

Steps:

  1. Make a text file: press-ingest/urls.txt with one URL per line.
  2. Run: python3 process_articles.py. It fetches, cleans, archives.
  3. Run: python3 ingest_to_vault.py. Files land in 02_SOURCES/Articles/.
  4. Run /diarize "Press" to build a press timeline canon page.

What you get: Press articles as plain Markdown. Bylines, dates, outlets, content. Searchable in your vault forever, even after the outlet pulls the article.

Real example: I've ingested 200+ articles about my own work going back to 2014. Half the original URLs are dead. The vault still has them. When I need a quote from a 2017 NYT piece, I grep instead of Google.

Recipe 8 — Generate a press page for your website

Goal: A clean, dated, sortable press list ready to paste into your site.
Input Your press archive from Recipe 7
Tool —
Skill /diarize "Press" then manual format pass
Time 45 min

Steps:

  1. Have Recipe 7 done first.
  2. Run /diarize "Press" — builds a comprehensive press canon page.
  3. Open the page. Ask Claude: "Reformat this as a press list for my website — outlet, headline, date, link."
  4. Save to 06_OUTPUTS/Website Copy/press.md (or .html).

What you get: A ready-to-deploy press section. Update it by re-running on a fresh ingest.

Meetings (Granola)

The fastest-decay archive of your life — meetings disappear into your memory the moment they end. Granola is the cheapest insurance policy you can buy.

Recipe 9 — Your meeting history, synthesized

Goal: Your last six months of meetings as a single rolling thematic archive.
Input Granola export (JSON or per-meeting Markdown)
Tool granola-ingest
Skill /granola
Time 30 min

Steps:

  1. Export from Granola (per-meeting Markdown is easiest).
  2. Drop into tools/granola-ingest/_inputs/.
  3. Run the ingest tool.
  4. In Claude Code, run /granola — synthesizes new meetings into the rolling archive at 03_SOURCE_NOTES/<year>/<year> - Note - Granola Meeting Archive.

What you get: A single source note that summarizes who you met with, what you discussed, what got decided, what's open. Re-run weekly — new meetings merge into the existing synthesis.

Real example: I processed 271 meetings from Nov 2024 → Feb 2026 in one pass. The synthesis surfaced things I'd lost track of: when a particular project actually wound down (Oct 23 2025, not Jan 2026 like I'd been telling people), who I'd promised follow-up to and ghosted, recurring themes across unrelated meetings.

Recipe 10 — Who am I meeting with, and is it the right people?

Goal: A drift report that compares stated priorities against actual calendar.
Input Granola archive (Recipe 9) + filled-in 00_HOME/Studio.md and project pages
Tool —
Skill /drift
Time 5 min

Steps:

  1. Have Recipe 9 done first; project pages should list current priorities.
  2. Run /drift.

What you get: A brutal but honest report: stated priorities vs. how you actually spent your last 90 days. Surfaces "phantom execution" — projects you talk about but don't actually work on. Most people who run this once change their next week.

Real example from my vault on 2026-04-29: drift surfaced 7 projects in "Active" status with zero execution evidence in 90 days. Reorganized my dashboard the same hour.

AI conversations (ChatGPT)

You've already had thousands of conversations with an AI. Most of them you'll never reread. This recipe makes them indexable.

Recipe 11 — What have I been thinking about?

Goal: Topic index of every ChatGPT conversation you've ever had.
Input ChatGPT export (Settings → Data Controls → Export)
Tool chatgpt-ingest
Skill /emerge (optional)
Time 20 min

Steps:

  1. Export from ChatGPT (you'll get a conversations.json).
  2. Drop into tools/chatgpt-ingest/ChatGPT Download/.
  3. Run: python3 parse_conversations.py.
  4. Output: parsed_domains/ with conversations grouped by topic.
  5. (Optional) Run /emerge on the topic index to surface patterns.

What you get: Your AI conversations as a thematic archive — "questions I keep returning to," "things I tried to figure out and never resolved." Combined with /emerge, it's a mirror of what's been on your mind.

Social (Facebook archive)

Old Facebook is most people's longest continuous private journal. The export is a nightmare. The tool fixes that.

Recipe 12 — Reconstruct a relationship from messages

Goal: Every message thread with one person, chronological, in plain text.
Input Facebook archive (Settings → Download Your Information)
Tool facebook-ingest
Skill /diarize <person>
Time 45 min

Steps:

  1. Request your Facebook archive (HTML or JSON; JSON is easier to parse).
  2. Drop into tools/facebook-ingest/your_fb_archive/.
  3. Run: python3 fb_to_obsidian.py.
  4. Output: a folder per person, chronological message thread inside.
  5. Pick a person who matters. Drop their thread into 01_INBOX/To Process/. Run /ingest. Then /diarize <Their Name>.

What you get: A canon page for that person that includes 10 years of message context. Useful for: writing a wedding toast, a eulogy, a long-overdue catch-up email, or just remembering how you became friends.

Already-loaded vault (skill-driven)

These assume you've done at least 5 ingests already. The vault has enough to push back at you.

Recipe 13 — Comprehensive page about a subject

Goal: One canonical page about a person, place, project, or theme — with every source cited.
Input Your vault, with 3+ sources mentioning the subject
Tool —
Skill /diarize <subject>
Time 5–15 min

Steps:

  1. Pick a subject that comes up in multiple sources.
  2. Run /diarize <subject>.
  3. Open the resulting page. Read it. Edit any wrong inferences.

What you get: A definitive page you'll reference for years. Has frontmatter, source citations, themes, chronology. The next time someone asks you about this person/project/place, you start here.

Recipe 14 — Pressure-test a real decision

Goal: 12 expert personas read your vault and stress-test something you're about to commit to.
Input A real decision you're 50/50 on
Tool —
Skill /council <topic>
Time 5–10 min

Steps:

  1. Pick a real decision: a hire, a launch, a partnership, a price, a "should I publish this."
  2. Run /council <decision>.
  3. Read all 12 perspectives. Notice which 2 made you uncomfortable. Engage those.
  4. Save the output to 06_OUTPUTS/Memos/ for the receipts.

What you get: A pressure test that's grounded in your archive (not internet generalities). Personas reference specific past experiences from your vault when relevant. Six months later when you doubt the decision, you have the receipts.

Real example: I've used /council on three real decisions in the last 90 days. One of them (Path A vs B) is the council output I show in cohort tours.

Recipe 15 — Surface what your vault implies

Goal: Patterns the vault contains that you've never articulated.
Input A vault with 30+ pages and at least 50 sources
Tool —
Skill /emerge
Time 10 min

Steps:

  1. Run /emerge.
  2. Read the patterns it surfaces.
  3. For each pattern that resonates, decide: create a canon page about this? If yes, run /diarize <pattern name>.

What you get: New canon pages about themes that were always there but unnamed. This is the single most "magical" skill in the system — it tells you things about yourself you didn't know.

Real example: /emerge surfaced "Identity as Medium" and "Institutional Fiction" as themes running through 12 years of my work. I'd never named those. Both became canon pages and now anchor the artist statement on my website.

Recipe 16 — Connect two things that don't seem related

Goal: Find the bridge between two topics in your archive.
Input A loaded vault
Tool —
Skill /connect <A> <B>
Time 5 min

Steps:

  1. Pick two topics: a recent project + an old one. Or a person + a place. Or two themes that feel unrelated.
  2. Run /connect <A> <B>.

What you get: Often a useful answer ("here are 3 ways these are related"), occasionally an essay you didn't know you had to write. The "huh" moments compound — file them in 09_IDEAS/.

Recipe 17 — Generate ideas grounded in your data

Goal: Brainstorm that's actually about you, not generic GPT output.
Input A loaded vault
Tool —
Skill /ideas
Time 10 min

Steps:

  1. Run /ideas. (Or with a focus: /ideas "things I could ship in 30 days".)
  2. Score the results on 09_IDEAS/Idea Scoreboard.md.
  3. Move the top 1–2 to 05_PROJECTS/Incubating/.

What you get: Ideas you actually might do, because they reference resources, contacts, and history that already exist in your vault. Not "you should start a podcast" generic stuff.

Recipe 18 — External research, vault-aware

Goal: Deep-dive on a topic outside your vault, but linked back to your canon.
Input A topic you want to learn about
Tool —
Skill /autoresearch <topic>
Time 20–60 min depending on depth

Steps:

  1. Run /autoresearch <topic>.
  2. Output goes to 08_RESEARCH/_Active/<slug>/ with frontmatter linking back to relevant canon pages.
  3. Review. If durable, mark it status: stable.

What you get: Structured research on the topic, with linked_canon: frontmatter so the next time you /diarize something connected, the research gets pulled in.

Real example: I ran /autoresearch on 17 debt instruments to map small-business loan options. The output sits in research; the decision gets surfaced into my capital strategy framework on the next pass.

Showcase outputs (publishable artifacts)

Recipes whose end product is something you can ship publicly.

Recipe 19 — Bio refresh

Goal: Three bios — short, medium, long — for one audience type.
Input Vault canon pages about your career
Tool —
Skill /diarize (your name) → manual prompt
Time 30 min

Steps:

  1. Run /diarize <Your Name> if you don't have a self-canon page yet.
  2. Ask Claude: "Using the canon page about me and recent press, write a 100/300/500-word bio for [audience]. The audience is a [conference / gallery / investor / podcast host]."
  3. Save to 06_OUTPUTS/Bios/.

What you get: Three bios you didn't have to write from scratch. Update them by re-running on the latest vault. I have nine bios — three audiences (artist, business, conference speaker) × three lengths.

Recipe 20 — A real talk, drafted from your archive

Goal: A 15-minute talk grounded in your actual life — not made-up examples.
Input A loaded vault, especially with strong canon pages on a few key topics
Tool —
Skill /diarize + manual brief to Claude
Time 2–4 hours

Steps:

  1. Pick the throughline: what's one thing I want this audience to walk away believing?
  2. Run /diarize on each topic the talk will reference. Build the canon spine first.
  3. Brief Claude: "Draft a 15-minute talk on . Beats I want to hit: [list]. Use only material from canon and source notes — no invented examples. Voice: [yours]."
  4. Iterate. Cut hard. Run /challenge on the strongest claim if it feels shaky.
  5. Save to 06_OUTPUTS/Talks/.

What you get: A talk that holds together because it's sourced. When someone asks a question that wasn't in the talk, the canon page is right there.

Real example: I drafted my TED talk in two days using this pattern. Three variant drafts ("Unwalkable Moat" / "Identity Is a Material" / "In the Gap") synthesized into a hybrid. Total: ~6 hours including the research pass.

Recipe 21 — Website that lives in your vault

Goal: Your public website's pages as Markdown / HTML files in 06_OUTPUTS/Website Copy/. Edit the vault, push, deploy.
Input An existing site (or willingness to build one)
Tool —
Skill manual
Time First setup: a weekend. Updates: 10 min.

Steps:

  1. Create 06_OUTPUTS/Website Copy/ if it doesn't exist.
  2. Move (or copy) your site's HTML/Markdown there.
  3. Add a deploy script (Vercel, Netlify, or rsync to your host).
  4. Edit pages in Obsidian. Push. Deploy.

What you get: Your website is now in version control + edited where the rest of your work lives. When you /diarize a new artwork, you can copy the synthesis into the artist page in two clicks.

Real example: my full artist website lives at 06_OUTPUTS/Website Copy/ — eight HTML pages, an llms.txt for AI discoverability, a sitemap, the whole thing. Updates to the artist page are now a vault edit, not a CMS chore.

Recipe 22 — Newsletter draft

Goal: A weekly or monthly newsletter draft sourced from your last X days of activity.
Input A vault used regularly (with /recap entries in the log)
Tool —
Skill /recap (must have been running this) + manual brief
Time 30 min

Steps:

  1. Pull the last 30 days of /recap entries from 00_HOME/Log.md.
  2. Brief Claude: "From the last 30 days of my log, draft a 600-word newsletter for my subscribers. Voice: [yours]. Lead with the most interesting thing. Keep it concrete."
  3. Save to 06_OUTPUTS/Substack/ or wherever your newsletter lives.

What you get: A draft sourced from things you actually did this month, not generic content marketing. Edit, send, repeat.

Recipe 23 — Memo to a stakeholder

Goal: A clean memo to an investor, partner, or client based on what's actually true today.
Input Vault with project pages + recent log entries
Tool —
Skill /council (optional, for stress-test) + manual draft
Time 1 hr

Steps:

  1. Pull the project page + last 90 days of log entries about the project.
  2. Brief Claude: "Draft a 500-word memo to summarizing where stands. Lead with what's working, then risks, then asks. Use only material in canon — no aspiration."
  3. Optional: run /council "Should I send this memo?" for a stress-test.
  4. Save to 06_OUTPUTS/Memos/.

What you get: A memo that's grounded in what actually happened, not what you hoped happened. Keeps you honest with stakeholders.

By goal (cross-reference)

If you want to … Try recipe
See your life on a map 1
Stop losing meetings 9
Rebuild a contact list that matters 5
Find what you're obsessed with 3, 11
Make a publishable bio 19
Pressure-test a decision 14
Find connections you've missed 16
Generate ideas about your own work 17
Surface patterns you haven't named 15
Draft a talk grounded in your archive 20
Move your website into your vault 21
See if your calendar matches your priorities 10
Reconstruct a relationship 12
Make your press archive permanent 7, 8
Send a newsletter you don't have to write 22
Draft a memo that's actually true 23

What's not in this list (yet)

A few recipes I've built or have on the roadmap but haven't packaged for the kit yet. Available to cohort members on request:

  • location-analysis — turns the photo CSV into 17 vault pages: decade maps, travel extremes, "where was I on date X," art pilgrimage routes, biome breakdowns. Real working tool.
  • Mentor pass on a draft — voice-match review against a mentor's body of work. Useful for essay polish.
  • Substack reply skill — drafts contextual replies to comments using your vault.
  • Harvest skill — pulls liked tweets / saved articles into the inbox automatically.
  • Patent / IP mining — scans your archive for patent-able ideas you've casually mentioned.
  • Drift report v2 — adds a "phantom execution" detector.

These are the kind of things that will ship in future paid packs or as community contributions. If you build one, send a PR.

Rules of the road

  1. Start with one source. Five sources is enough for the system to do something useful. You don't need 200 to get value.
  2. Run /recap at the end of every session. Otherwise you'll forget what you did and have to figure it out from the file tree.
  3. /vault-lint weekly. It catches stub pages, broken links, duplicate names, stale frontmatter. The system is self-healing if you let it lint.
  4. Don't re-do work the system has already done. Before writing a bio from scratch, search 06_OUTPUTS/Bios/. Before writing a memo, check 06_OUTPUTS/Memos/. Past you wrote the right thing, probably.
  5. The vault is the workshop, not the gallery. Outputs are public. The vault is private. Don't share the vault folder; share the outputs that came out of it.