Five skills do 90% of the work. Verify the install. Push through the first week. The interesting work starts now — using the system well, evolving the harness, building the habits.
Second-Brain OS lives in ~/Desktop/second-brain-harness/. Whichever interface you use — Terminal, Claude desktop app, or VS Code — select that folder before you start a new session. Otherwise Claude Code loads the wrong project context and skills, hooks, and memory all silently fail to fire.
Open ~/Desktop/second-brain-harness/ first. Start the conversation second. Everything else just works.
Open Terminal, cd into the folder, then run claude. The session is anchored to that folder for its lifetime.
cd ~/Desktop/second-brain-harness && claudeclaude launched inside the open folder.Click New session. Before typing anything in the message box, look at the bottom of the window for the folder selector chip (it shows Local next to your current folder). Click it, then pick second-brain-harness from the recent list — or use Open folder… if it's not there. Then type your message. The session now has the full Second-Brain OS context loaded.
If you don't have it yet, grab VS Code (free, cross-platform). Install the Claude Code extension from the Extensions marketplace. Then:
second-brain-harnessCmd/Ctrl+Shift+P → "Claude Code: New Session")The intent-detector hook still fires but slash commands won't resolve correctly (/new-project, /briefing, /find all read CLAUDE.md + skills from the active folder). If commands feel broken, the first thing to check is which folder Claude is running in.
The most common workflow. /new-project handles both Beru-style meta-projects (planning, content, research) and code repos. Here's what running it actually looks like.
~/Desktop/second-brain-harness/ (see above if you skipped that step), type /new-project directly, or just say it in plain English: "let's start a project on Q3 capacity planning", "kick off a new MCP server for Notion sync", "scaffold a project for the migration". The intent-detector hook surfaces the suggestion if you forget the slash.
workspace/1-Projects/. Code repo — MCP server, agent, library, CLI, bot, anything that gets its own git history. Lives in workspace/2-Coding/<scope>/<name>/ with optional gh repo create.
q3-capacity-planning), the project type (planning / strategy / content / research / meetings / ongoing), and the one-line summary. Scaffolds the folder from 3-Resources/templates/. Two files written: CLAUDE.md (project rules + frontmatter) and memory.md (the append-only decision log).
work / personal / forks), stack (TypeScript / Python / Rust / Bash / etc.), and whether to create a private GitHub repo. If yes, runs gh repo create <handle>/<name> --private and clones it. Appends a row to 3-Resources/code-projects.md (the one allowed index — code repos are gitignored from the workspace, so the index is the only durable record).
workspace/1-Projects/2026-05-q3-capacity-planning/
├── CLAUDE.md ← project-specific rules + status frontmatter
└── memory.md ← append-only — every decision logged with dateCLAUDE.md has YAML frontmatter — status: active, project type, owner, date — that /prune-projects and /archive-project read later. The memory.md starts empty; you append a one-liner every time something material happens.
The two-file rule. Every project gets exactly two meta-files. Not three. Not five. More than that and people stop opening the folder, the project goes cold, and the system fails. Append-only memory.md means decisions never get rewritten — the trail of why things are the way they are stays intact.
The skill you'll run more than any other. /briefing composes calendar, email, Slack, and project state into one focused brief — every morning, with coffee.
second-brain-harness folder selected in your session (Terminal, desktop app, or VS Code — all work), type /briefing in the morning, or just say "brief me", "what's on my plate today", "what should I work on this morning". The skill triggers broadly on day-orientation language.
gws), email (via gws), Slack (via the MCP), Jira (via Atlassian MCP), GitHub (via gh). If something isn't wired up, the briefing gracefully degrades — fork users with zero MCPs still get a useful brief from local state alone.
USER.md). Priority Slack channels (the ones flagged in USER.md). Active projects with their staleness scores. Open commitments from your contacts directory.
contacts/<slug>.md). Calendar conflicts. What you promised in previous threads. Anything urgent that crosses two signal sources at once (calendar conflict + an email about it = priority surface).
docs/briefings/morning-briefing-YYYY-MM-DD.md. Surfaces in chat: top items, conflicts, what to focus on. You can open the HTML file in a browser for the fuller view, or just read the chat summary.
# Morning Briefing — 2026-05-11
## Today at a glance
- 3 meetings, 1 conflict (10am 1:1 vs 10am cross-functional sync)
- 12 unread emails — 2 HIGH priority (manager, direct collaborator)
- 1 stale project flagged (2026-04 launch prep, 21 days untouched)
## Calendar — what matters
- 9:00am — Standup with AI Task Force (Omar, Jaya, you)
Last interaction with Omar: 2026-05-09 — you committed to
finishing the eval framework spec by Friday.
- 10:00am — CONFLICT. Cross-functional sync vs your 1:1 with
Jaya. Decision: tell the sync you'll send notes async,
keep the 1:1.
## Top emails
- Sarah (Sales) — Q3 forecast review, needs your input by EOD
- Andy (Marketing) — campaign brief draft, no urgent actionIt connects signals instead of summarizing each in isolation. A calendar event with a name in it triggers a contact lookup, surfaces the last interaction, and reminds you what you committed to. Your priority signals from USER.md (direct collaborators = HIGH, calendar conflicts = URGENT) drive the ranking. It's not a feed — it's a chief-of-staff handing you a one-pager.
Five skills do 90% of the work. Learn these first. Everything else is decoration.
/new-project1-Projects/YYYY-MM-slug/ with CLAUDE.md + memory.md + status frontmatter. Also handles code repos (branches into 2-Coding/<scope>/<name>/ with optional gh repo create). Step 0 of this skill runs /find first — connect before create./save-resource0-Inbox/ into 3-Resources/ with the right convention per type. Research goes to a date-prefixed topic folder. Reference goes flat. Meeting notes go to meetings/YYYY-MM-DD-<topic>.md./briefing3-Resources/briefings/morning-briefing-YYYY-MM-DD.html. Run it with coffee./prune-projects/archive-project for each one you approve./archive-project1-Projects/ to 4-Archive/ and flips its status frontmatter from active to done. Optional one-paragraph retro lands in memory.md before the move so decision context survives.Run /briefing in the morning. Capture anything weird into 0-Inbox/ throughout the day. Run /prune-projects on Friday. That's the floor. Everything else compounds on top.
Two commands. Each should report a clean logged-in state. Run them in any terminal.
gh auth statusShould print Logged in to github.com with your username and the scopes gist read:org repo workflow.
gws auth statusShould print authenticated with your @samba.tv account plus the enabled API surface (Calendar, Drive, Gmail, Docs, Sheets, etc.).
Re-run the matching auth command from the install output (the 5 one-time commands the installer printed at the end). Each opens a browser to sign in. If gws auth login says "Access blocked", ping your team lead — the GCP consent screen needs to be set to Internal.
If something went sideways, the log tells you exactly what.
tail ~/samba-onboarding.logSecond-Brain OS is live. The interesting work starts now — using it well, evolving the harness, building the habits.
/briefing every morning for five days straight. Even if it's rough at first.0-Inbox/ — meeting notes, random ideas, files you weren't sure where to put./new-project. Anything. The first one's the hardest.SOUL.md, USER.md, IDENTITY.md — edit anything that doesn't sound like you./prune-projects on Friday. Even if you have nothing to prune. Build the rhythm./find <some-topic> before scaffolding anything new. Connect before create.beru-workspace/1-Projects/2026-05-second-brain-design/system-design.md in your clone. Why the harness is shaped the way it is.The harness is opinionated, but it's yours. Edit the persona files. Rewrite the rules that don't fit. Add skills for the leaves of your work tree. The system gets better as you tune it — that's the whole point.