What's new
Notable changes and highlights across bond releases.
bond 1.10.0: hub mode
Until now, bond helped you work with coding agents in standalone git repositories: bond init set up the configuration, bond opened your workstation in Herdr, and bond team distributed parallel branches among agents in their own worktrees.
In real-world projects, however, code is frequently split across multiple repositories: a frontend web app, a backend API service, and perhaps shared libraries. When a feature or bug fix requires touching both the frontend and backend at once (such as creating a new API endpoint and immediately consuming it in the UI), coordinating separate branches and agents by hand quickly becomes tedious.
That is why hub mode is here.
A hub with a React frontend (submodule) and an Express backend (symlink): bond hub init, bond hub check and a pi orchestrator that splits the story into a vertical branch with bond team, one pi agent per repo, and checks it all at the end.
What is a hub?
A hub is an operations git repository that groups together your project repositories (declared in hub.toml). It does not replace your repositories or force them into an artificial monorepo: each repository preserves its own git history, remotes, and independence.
Repositories can be linked into a hub in two ways:
- As a git submodule (
link = "submodule"): ideal when sharing the hub repository across your team through a remote git forge. - As a local link (
link = "symlink"): ideal for working with existing local clones already on your machine, without altering their structure. Machine-specific paths to local clones are saved in.bond/hub.local.toml(which remains local and is never committed to git).
Vertical work with bond team
The real power of hub mode unlocks with bond team. In an individual repository, bond team delegates independent branches to separate agents. In a hub, bond team operates vertically:
- A branch with the same name across all repositories: when you run
bond team fix/login, bond creates the branch in the hub and checks out an isolated worktree with that same branch in each session repository (or in the subset specified via--repos). - One agent per repository: each repository gets its own agent in an isolated worktree (
~/.herdr/worktrees/<hub>/<repo>/<branch>), equipped with that repository's own configuration and tools. - One vertical branch at a time: to maintain focus and prevent interdependency conflicts, bond ensures only one vertical branch is active in the hub at any given moment.
- A unified live table: Herdr opens a single
tabla · <hub>pane tracking all repositories of the session in real time.
When to use a hub?
Use a hub when:
- Your system is split across several interdependent repositories (frontend, backend, microservices, libraries).
- You want an orchestrator agent (such as pi or Claude Code) to plan global milestones and coordinate specialized minion agents across repositories.
- You need to implement or verify features that cross repository boundaries with synchronized branches.
If you are only working on a single self-contained repository, you do not need a hub: bond init and bond continue to work just as they always have.
Hub directory layout
At the root of a hub, session configuration, linked repositories, and operational documentation live together:
my-hub/
├── hub.toml # Session declaration (name, base branch, repos)
├── AGENTS.md # Global agent instructions with generated bond:hub block
├── .bond/
│ ├── bond.toml # Hub bond configuration (agents and skills)
│ └── hub.local.toml # Local symlink paths on this machine (.gitignore)
├── repos/ # Default folder for linked repositories
│ ├── frontend/ # Submodule or local symlink
│ └── backend/ # Submodule or local symlink
├── docs/
│ ├── plans/ # Multi-repo plans and milestone tracking
│ └── decisions/ # Hub decision memory (optional)
└── .gitignorehub.toml: declares the session name, the default base branch (e.g.main), and each repository with its path, base branch, link type (submoduleorsymlink), default agent, and an optional description.repos/: the standard folder where linked repositories live to keep the hub root organized and uncluttered.docs/plans/anddocs/decisions/: the hub centralizes team plans and, if decision memory is enabled via[decisions], agent task requests and completion reports (runs) are recorded here.- The
bond:hubblock inAGENTS.md:bond syncautomatically maintains a marked block containing a summary table of the session repositories, their descriptions, base branches, and operational rules (such as not editingrepos/directly from the hub, but throughbond team). Content outside the marker is left untouched.
The end-to-end workflow
Working in a hub follows six straightforward steps:
# 1. Initialize or prepare the hub
bond hub init
# 2. Verify that the session is ready
bond hub check --fix
# 3. Fast-forward base branches
bond hub pull
# 4. Open the vertical branch and assign tasks
bond team feat/auth
bond team prompt feat/auth "Add JWT authentication: endpoint in backend and form in frontend"
# 5. Collect work from agents
bond team collect feat/auth --repo backend
bond team collect feat/auth --repo frontend
# 6. Clean up worktrees
bond team clean feat/authbond hub init: interactive wizard that initializes the hub, runsbond initif needed (including theplan-teamskill), imports submodules or local symlinks, and sets up bond in member repositories that do not have it yet.bond hub check: verifies every session repository. It checks that repositories exist, link types match, base branches exist, agent configurations are in sync, and worktrees are clean. With--fix, it safely repairs broken links and initializes submodules. In--jsonmode, it provides structuredleveldetails (ok,warn,fail) and issue tallies.bond hub pull: updates the base branch of the hub and each repository usinggit fetchand fast-forward advance only. If a repository has uncommitted changes or divergent history, it is skipped safely without merging or rebasing.bond team <branch>: opens the vertical branch in the hub and creates worktrees and agents in each repository.bond team promptdistributes instructions, automatically annotating the prompt with teammate names and folders so cross-repo work fits together cleanly.bond team collect: retrieves the report and commits from each agent. If the hub has decision memory configured, the task and report are recorded into a centralized run.bond team clean: removes worktrees across session repositories. With--delete-branch, it also deletes integrated branches once work is completed.
What the hub does for you
- Cascaded configuration: when adding repositories with
bond hub initorbond hub add, bond applies the agent configuration to repos that lacked bond, saving you from configuring each one manually. - Automatic agent context: through the generated
bond:hubblock inAGENTS.md, any agent running in the hub immediately understands what repositories exist, what each is responsible for, and where plans are kept. - Guaranteed structure: the
repos/default location, outdated base warnings, and single-vertical-branch rule prevent cross-repo conflicts and common multi-repo mistakes.
For complete CLI options and details, see bond hub in the Reference and the vertical branch guide in The workstation.

