Drive, but for .md files.

The Shared Context Drive is the source of truth for every skill, agent, hook, and config that shapes your AI. Sync it, see it, edit it — and bring your teammates and their agents into the same canvas.

Every file in your harness — skills, agents, settings, CLAUDE.md — synced and visible in one place.
Collaboration built-in. Shared projects and working files so teammates and agents work in tandem.
Git-backed underneath. CLI for agents, visual editors for humans, version history for everyone.

Connect your harness See what syncs

git-nativeEvery file is versioned

~30sFrom clone to first sync

CLI + GUIAgents script, humans click

~/your-harness·sql-explainer.md main Synced

sql-explainer.md × settings.json

name: sql-explainer

description: Explains a SQL query in plain English with cost notes.

tools: [Read, Bash, Snowflake]

# SQL explainer

Walk through any query the way you’d brief a new analyst — what it returns, where it joins, where it might be slow.

## Behaviour

Read the query top-down; never paraphrase the SELECT.
Flag full-table scans against tables > 10M rows.
Suggest a cost-aware rewrite when the optimizer is unsure

Priya just pushed 3 commits to on-call-triage.md
your local copy will refresh in a moment

One drive. Every tool your harness already runs in.

Claude Code Cursor Codex Gemini CLI MCP-compatible

YOUR HARNESS, IN ONE PLACE

A single canvas for everything that shapes your agent.

Your skills, agents, hooks, MCPs, settings, and CLAUDE.md — pulled from every tool you use and held in one workspace you can actually see.

One drive, every resourceSkills from Claude. Cursor rules. Codex agents. Hooks, MCPs, and settings — all sitting next to each other instead of scattered across dotfiles.
Sync, don’t copy-pasteEdit a skill once and the change reaches every tool that uses it. No more pasting the same prompt into four different rules files.
See what’s actually shaping outputEvery file your agent reads, in the order it reads them. Finally know why your agent did the thing it did.

All resourcesSkillsAgentsConnectorsSettings

SKILL

pr-review-checklist

skills/pr-review-checklist.md

SKILL

sql-explainer

skills/sql-explainer.md

AGENT

on-call-triage

agents/on-call-triage.md

HOOK

pre-commit-lint

hooks/pre-commit-lint.sh

MCP

snowflake-readonly

mcps/snowflake.json

SETTINGS

settings.json

~/.claude/settings.json

MARKDOWN

CLAUDE.md

~/work/CLAUDE.md

SKILL

changelog-writer

skills/changelog-writer.md

AGENT

spec-reviewer

agents/spec-reviewer.md

Acme · Q3 launch project

MRPSJLDK

messaging.md

projects/q3-launch/messaging.md · 2 active

PS Priya

Headline candidatesPS editing

The drive backend is git-based, so we can talk about it the same way we talk about the code. Every change is a commit; every project is a branch.

AI launch-copy-agent

Audience notesAI drafting

Writing for IC engineers and AI-curious PMs. Lean on concrete file paths and CLI examples rather than abstract talk about “context.”

Open questions

Does the drive ship with a default project for personal use? Should team plans include public projects, or only private ones?

FOR TEAMS AND THEIR AGENTS

Shared projects so every agent works in tandem.

Working files, scratch docs, and shared context that your teammates and their agents can all read from and write to — without stepping on each other.

Shared projectsSpin up a project, drop in the relevant skills and CLAUDE.md, invite the team. Every agent on the project starts from the same context.
Working files agents understandSpecs, briefs, ADRs — written for humans, indexed for agents. Update once, every teammate’s agent picks it up on the next run.
Live presence on every fileSee who’s editing, who’s reading, and which agents are actively pulling context — humans and bots, on the same canvas.

POWERFUL FOR AGENTS · EASY FOR HUMANS

A git-based backend with two front doors.

Agents script the drive through a real CLI. Humans use a calm visual editor. Both write to the same git history — so nothing your agent does is invisible, and nothing you do is locked away.

priya@acme · ~/harness

$ sc pull

fetched 142 resources · 3 changed

M skills/sql-explainer.md

M agents/on-call-triage.md

+ hooks/pre-commit-lint.sh

$ sc edit skills/changelog-writer.md --set tools='[Read,Bash,Linear]'

✓ patched frontmatter · 1 line changed

$ sc commit -m "Add Linear MCP to changelog-writer"

[main a4f1c2] Add Linear MCP to changelog-writer

1 file changed, 1 insertion(+), 1 deletion(-)

$ sc log skills/changelog-writer.md --limit 3

a4f1c2 Add Linear MCP to changelog-writer · 2m

9b2d10 Tighten release-note tone · 1d

331e4a Initial draft from Maya’s notes · 5d

skills/changelog-writer.md v7 · committed

Name

changelog-writer

Description

Drafts release notes from a list of merged PRs.

Tools

[Read, Bash, Linear]

Behaviour

Group commits by area. Lead with user-visible changes. Drop pure refactors unless asked.

History

v7 · 2m ago

Git underneath

Branches, commits, diffs, blame. Your harness behaves the way every engineer already expects code to behave.

CLI for agents

sc pull · sc edit · sc commit — composable verbs your agent can call in a loop.

Visual editors

Frontmatter as forms, markdown as preview, MCPs as toggles. Edit a skill without touching a terminal.

Time travel

Roll back any file to any point. See exactly what your harness looked like the day a workflow stopped working.

“I used to keep skills in three places — a Cursor folder, a Claude config, and a Gist I’d forgotten about. Now I have one drive. My agent reads from the same files I edit, and every change I make is a commit I can roll back. I finally trust my own setup.”

Jordan Liang

Staff Engineer · Northbeam

FAQ

Common questions about the Drive

Where does my harness actually live?

On disk, as plain files, in a git repo you own. Shared Context syncs and visualizes it — but the source of truth is your filesystem, and you can clone it anywhere you’d clone a normal repo.

How does it sync across Claude, Cursor, Codex, and Gemini?

Each tool has its own conventions for skills and rules. The drive maps your canonical files into the right place for each tool — so editing one skill in the drive updates Claude’s skill folder, your Cursor rules, and any Codex agent that imports it.

Can my agent really drive this from the CLI?

Yes. `sc` is a small, composable CLI: pull, edit, commit, log, branch. Agents call it the same way they’d call git. Every action is a normal commit, so anything an agent does is auditable and reversible.

What about shared projects — is that just a folder?

It’s a folder backed by a branch. Invite teammates, drop in the skills and CLAUDE.md the project needs, and every agent on the project starts from the same context. Merge back to main when you’re done.

Is this safe for sensitive context?

Drives are private by default. MCPs run with credentials you control. We never train on your files, and team drives can be self-hosted if you’d rather keep the git remote on your own infrastructure.

Drive, but for .md files.

# SQL explainer

## Behaviour

A single canvas for everything that shapes your agent.

pr-review-checklist

sql-explainer

on-call-triage

pre-commit-lint

snowflake-readonly

settings.json

CLAUDE.md

changelog-writer

spec-reviewer

messaging.md

Shared projects so every agent works in tandem.

A git-based backend with two front doors.

Git underneath

CLI for agents

Visual editors

Time travel

Common questions about the Drive

Your harness deserves a real home.

Your harness deserves
a real home.