Install Second-Brain OS

The three tiers

Pick how deep to go.

All three tiers clone the same public repo — github.com/the-sid-dani/second-brain-harness. The difference is how much of the harness's optional surface area you turn on the first time you launch.

Order matters

For Lite and Full, ./scripts/install.sh runs before claude. Don't skip that order — Claude Code reads .mcp.json, .claude/settings.json, and env vars at session-start. Running the installer afterwards means restarting Claude. See the "Why install.sh runs first" callout below.

Tier 1

Minimal

Fastest start

Just the core. Chief-of-staff workflows (/briefing, /contact, /find, /new-project) and the design skills (decks, dashboards, landing pages). No advanced research tools. You can upgrade later, anytime — the installer is idempotent.

git clone https://github.com/the-sid-dani/second-brain-harness ~/Desktop/second-brain-harness
cd ~/Desktop/second-brain-harness
claude

Then inside Claude Code: /bootstrap

Best if: you want to try it without installing anything beyond Claude Code itself. Lowest commitment, fastest to first useful output.

Tier 2 · Recommended

Lite CCv4

Recommended

Everything in Minimal, plus the continuous-claude harness — a set of skills and automations that make long, complex work easier. Adds:

/research — deep-dive research on a topic, summarized for you
/autonomous — Claude works through multi-step tasks on its own with checkpoints
/premortem — pre-implementation failure analysis (what could go wrong, before you start)
/review — code or doc review pass
/create-handoff — generates a session handoff doc when you stop
Plus 5 background automations (statusline, pre-compaction snapshots, file-read optimization, auto-handoff, post-edit lint checks)

Skips the FastEdit integration (the heavy code-editor optimization — not needed unless you're a heavy coder).

git clone https://github.com/the-sid-dani/second-brain-harness ~/Desktop/second-brain-harness
cd ~/Desktop/second-brain-harness
./scripts/install.sh --no-fastedit-model
claude

Then inside Claude Code: /bootstrap

Best if: you want the full experience without the heaviest dependency. The recommended starting point for most teammates.

Tier 3

Full

Adds FastEdit

Everything in Lite, plus the FastEdit integration — a faster, smarter code editor that uses a small AI model (1.7B parameters) to make precise surgical edits on large files. Most useful for engineers and PMs who edit a lot of code. The installer downloads the model on first run, so allow a few extra minutes.

git clone https://github.com/the-sid-dani/second-brain-harness ~/Desktop/second-brain-harness
cd ~/Desktop/second-brain-harness
./scripts/install.sh
claude

Then inside Claude Code: /bootstrap and /mcp

Best if: you write code daily or work in large repos. The extra model download pays for itself within an afternoon of editing.

Important · Read this

Why `install.sh` runs before `claude`.

Claude Code reads three things at session-start, and only at session-start:

.mcp.json — the list of MCP servers to spawn for this session
.claude/settings.json — hook wiring, permissions, env vars
Environment variables — API keys (Anthropic, Exa, NIA) from your .env

The CCv4 installer is what puts those three things in place. Specifically, ./scripts/install.sh:

Registers FastEdit MCP in .mcp.json (Tier 3 only)
Wires the CCv4 hooks through .claude/settings.json — the hook files are already in .claude/hooks/, but they reference bloks, tldr, and fastedit binaries the installer puts on your PATH
Puts your API keys in .env for Ouros / Exa / NIA harness tools

All of that has to be on disk before claude starts. Reading the same config after Claude is already running means exiting and relaunching the session.

The previous wording was wrong

The earlier version of this page said "run install.sh after the Minimal flow above." That made teammates launch Claude first, hit a half-broken session, run the installer, then have to exit and restart. The fix is the linear flow you see in each tier — clone, install, launch. One pass per tier, no restart.

For Minimal, there's no install.sh step because that tier doesn't depend on FastEdit MCP, the CCv4 binaries, or harness API keys. The persona + design skills work straight from the cloned repo.

Compare

What each tier adds.

Quick side-by-side. Pick the one that matches what you'll actually use this week. You can rerun a higher tier anytime — the installer is idempotent.

Feature	Minimal	Lite	Full
Chief-of-staff skills (`/briefing`, `/find`, `/new-project`…)	●	●	●
Design skill family (26 skills — landing, deck, dashboard, poster…)	●	●	●
PARA workspace + persona files	●	●	●
`/bootstrap` interactive setup	●	●	●
CCv4 skills (`/research`, `/autonomous`, `/premortem`…)	—	●	●
CCv4 hooks (status, pre-compact, tldr-read, auto-handoff)	—	●	●
Ouros / Exa / NIA harness tools	—	●	●
FastEdit MCP (AST-aware code edits)	—	—	●
First-launch time	~30 sec	~2 min	~5 min

Which one should I pick?

Minimal — you mostly want the chief-of-staff workflow (briefings, contacts, projects, design). You don't write code daily and don't need long-running research agents.
Lite — recommended default. You want the chief-of-staff stack and the CCv4 research/autonomous pipelines. You write code occasionally; the standard Claude Code Edit tool is fine.
Full — you write code daily in larger files and want surgical AST-aware edits via FastEdit. Adds a ~3 min one-time model download.

All upgrades are idempotent — re-run ./scripts/install.sh later to move from Lite to Full, or first-time-run it to move from Minimal to Lite. Nothing gets clobbered; the installer skips what's already there.

Once installed

Use it day-to-day

Day-to-day workflow →