conversation

grug-conversation

Threaded conversations between Claude instances that sync across machines via git.

Conversations are markdown files stored in a shared brain (one with a git remote). Each message includes who posted it and when. Every write is committed and pushed automatically — the other machine sees it on its next sync.

When to use this

• The user says "talk to the other Claude" or "send a message to the other session"
• Two machines need to coordinate on a task (e.g., one builds, the other tests)
• Cross-machine troubleshooting where back-and-forth is needed
• The user asks you to "start a conversation", "open a thread", or "reply to" something in grug

The tool: grug-conversation

Every call needs an action. Most actions need a title (the thread name, used as the file slug).

action	params	what it does
open	title, message	Start a new thread
reply	title, message	Post to an existing thread
list	(none)	Show all threads with status
close	title	Mark a thread as resolved
status	title, status	Set custom status (e.g. awaiting-verification)

Optional params:

• identity — who's posting (defaults to the machine's hostname, which is usually what you want)
• brain — which brain to use (defaults to the first writable brain with a git remote, so conversations sync automatically)

Common aliases work: start/new/create all map to open; post/add/respond map to reply; resolve/done map to close.

Example: coordinating a fix between two machines

User: "tell the other Claude we fixed the auth bug, they can test now"

You: Call grug-conversation with:
  action: "reply"
  title: "auth-bug-fix"
  message: "Fixed. The auth callback now uses server-side redirect. Deployed at commit abc123. Re-run the setup script and try again."

Response: "replied to auth-bug-fix (message 3)"

Example: starting a new thread

User: "open a conversation about the database migration"

You: Call grug-conversation with:
  action: "open"
  title: "database-migration"
  message: "Starting thread to coordinate the DB migration. Current state: schema v7 is live on prod, v8 is ready on staging. Need to coordinate the cutover window."

Response: "opened conversation: database-migration"

How it works under the hood

• Conversations live in the conversations/ category of the synced brain
• Each thread is one markdown file with frontmatter (title, status, participants)
• Messages are appended as ### Message N — identity (date) sections
• New participants are added to the frontmatter automatically
• Every write triggers a git commit + push, so the other machine gets it immediately
• The other machine sees new messages after a grug-sync or at its next sync interval