The Short Version
| Situation | Glyph approach |
|---|---|
| Two agents working on different tasks | Use separate work contexts from the same realm. |
| One agent actively editing a task | Claim the work context with exclusive mode and heartbeat while working. |
| A human reviewing an agent’s task | Inspect the same work context, or use a shared-read claim where policy allows it. |
| One task depends on another | Add a dependency edge and publish in dependency order. |
| Two tasks touch the same files | Run conflict checks before publication. |
| Work is done | Publish, then prune the materialized projection. |
| Another device needs the public source | Use the GitHub mirror or a Git export. |
| Another device needs full Glyph history | Copy or sync .glyph/ only while no process is writing, or wait for remote Glyph store sync in a future version. |
Multiple Users Or Agents In One Project
Each task should usually get its own work context:public realm. They do not need separate Git branches, and they do not fight over one global working tree.
The important habit is to name the task, not the history shape. A branch name often carries several meanings at once: feature, owner, readiness, privacy, and review state. Glyph separates those meanings into work contexts, claims, realms, checkpoints, and publications.
Claims And Heartbeats
A claim says who is actively responsible for a work context right now.Same Workspace, Different Actors
Two actors can inspect the same work context, but only one actor should mutate it at a time unless the project has a policy for shared editing. Recommended pattern:- The active editor holds an
exclusiveclaim. - Reviewers inspect with
glyph diff,glyph read, or a projected copy. - The editor checkpoints meaningful states.
- The editor releases the claim before another actor takes over.
Related Work And Dependencies
Sometimes two contexts are separate but ordered. For example,api-layer should publish after data-layer.
Conflict Checks
Before publication, check whether another active context overlaps:Sharing Work Across Devices
There are three different sharing needs. They should not be mixed up.Share The Public Project
Use GitHub or a Git export. This is the stable compatibility path today.Continue Public Work On Another Device
Today, the simple path is:- Sync the current public realm from the first device:
- On the second device, clone the GitHub mirror and bootstrap a local Glyph store from that public projection:
Move Full Glyph State To Another Device
The full local graph lives in.glyph/: store database, content blobs, workspaces, claims, publications, and exports. In prototype 0, there is no hosted Glyph store sync yet.
If you need to move the full local graph today, copy the project directory including .glyph/ while no Glyph command or editor is writing to it:
.glyph/store.db on a cloud-synced folder and let multiple devices write to it at the same time. SQLite is reliable for local concurrency, but a Dropbox-style shared database is not a distributed source-control protocol.
What Hosted Or Remote Glyph Should Add
Future Glyph should make cross-device collaboration first-class:- remote source graph sync
- per-realm access control
- signed publications
- durable claims with session identity
- stale claim takeover
- conflict records that travel between devices
- remote work context handoff
- policy-controlled private realms
.glyph/ as the local canonical store for one active device at a time.
Recommended Policy
- Use one work context per task.
- Use claims for any task with more than one possible actor.
- Heartbeat during long-running agent work.
- Check conflicts before publishing.
- Add dependencies when publication order matters.
- Publish to a realm before syncing GitHub.
- Prune completed projections after publication.
- Use GitHub for public sharing, not for private Glyph graph state.
- Copy
.glyph/between devices only as an intentional offline handoff.