bks-lab/open-bridge

**Stop re-explaining your work to your coding agent every morning.**
A git repo of markdown and YAML that Claude Code, Codex or Copilot CLI reads at the start of every session. Your repos, your clients, your open tasks and yesterday's work log are already in context, so the first answer picks up where you stopped. No database, no hosted service: plain files in a private repo you own. The BKS-Lab team runs its own work on it every day (status).
That is the shipped demo workspace answering. The clip below is the same morning, 15 seconds without sound (click it for the 1:25 film on YouTube); the full live session adds the morning session start, an incident taken from log triage to a tested fix, and first-run onboarding.
Set it up
Your data needs a private home before you write any of it: your own private repo becomes origin, BKS open-bridge stays a read-only upstream. The easiest way is to let your agent do it.
Paste this into Claude Code, Codex or Copilot CLI
Every step the agent takes is in SETUP.md: check your tools, show you the plan, re-home the remotes to your private repo, arm the push guard, prove it, then ask you to restart it inside the new folder. Why setup ends with a restart: docs/install.md.
Not ready to set anything up? Look at a running one first
Two commands, nothing to configure. The repo ships a runnable demo (examples/agency/): a fictional two-client agency with a filled board, two days of logged work, and a P1 incident in flight. You need an agent CLI, for example npm install -g @anthropic-ai/claude-code.
Ask good morning, then where was I on the payment retry? and why is the cart task in review?. Everything it answers is read from plain markdown in that folder: open work/log.md next to it and the trick disappears. Do not git push from this clone; it points at the public repo.
One person wearing several hats (a consultancy partner who is also a freelancer, with a household and a home server)? examples/portfolio/ is that shape: three personas, a managed client service, life admin, and digests that stay silent until something is red.
Who it is for
- **For** people who run several clients, repos or roles with a coding agent, from a solo operator wearing several hats to a small consultancy.
- **For** anyone tired of re-explaining who they are, which client they are on, and what happened yesterday, every session, in every tool.
- **If you work in one repo**, a CLAUDE.md is probably enough. open-bridge pays off once you juggle several repos, clients or roles.
- **Not for** people who want a hosted app. There is nothing to host; the value is files your agent reads.
What you get
Four pieces do the daily work:
| Piece | What it is |
|---|---|
| **Registry** | ecosystem.yaml: your repos, clients and workspaces in one file, read as a table of contents with one entry fetched on demand (context-index) |
| **Task management** | work/: one folder per task with a STATUS.md, a generated board, and an append-only daily log (work-system) |
| **Skills** | one skills/ tree of plain-language verbs (/briefing, /debrief, /archive, …) that every supported tool loads (commands) |
| **Standing orders** | always-on rules in protocols/standing-orders/, loaded every session and scoped per sub-agent |
And capabilities you switch on when you need them:
| Capability | One line |
|---|---|
| Org overlays | subscribe to your organisation's shared config by git URL, without a fork |
| Workloads | one file per scheduled job or daemon on your machines, checked against the live service manager |
| Inbox | everything that needs you waits in one place until the live source says it is done, and the briefing shows it first |
| Workplace | plan the day as workspaces with one agent tab per task, and steer the tabs from one place through a driver for your terminal tool |
| Cockpit (alpha) | a local web page that shows the briefing, inbox, tasks, agent tabs and machines and acts on them with a click; an early version, off until you switch it on, and still being extended |