Problem
Using agmsg across multiple desktop agents is currently hard to operate because the user cannot see the communication state as a whole.
The scripts can show a registered team roster and per-agent history/inbox, but they do not provide a single operator-facing view for:
- who is registered in the team
- who is actually alive / stale / unknown now
- which delivery mode each agent/session is using (
off, turn, monitor)
- which monitor/watch processes are attached to which identity
- which Codex identity is bound to which Desktop thread/session
- recent traffic by sender/receiver/topic/thread/intent
- warnings such as stale pidfiles, monitor not recovered after reboot, ambiguous loaded threads, or possible cross-injection risk
This becomes especially painful after a PC reboot. Some Claude-side sessions may recover automatically, others may not. Codex monitor is more fragile because delivery targets an ephemeral TUI thread, so recovery requires safe identity-to-thread rebinding. Without visibility, the user cannot tell whether the network is healthy or silently degraded.
Desired UX
Provide an operator dashboard or status command that answers, at a glance:
- Who exists in this team?
- Who is currently reachable?
- Who is receiving by
turn vs monitor vs off?
- For Codex Desktop, which identity is bound to which thread/session?
- What messages recently moved between whom?
- What needs attention?
A first version can be CLI/TUI/HTML; a polished GUI can come later.
Initial scope for the fork
Keep this local to the fork while experimenting. Do not use upstream issues as internal scratch.
Possible phased approach:
- Phase 1:
agmsg status / agmsg doctor style summary using existing official script data
- Phase 2: per-identity presence and delivery status, including stale/unknown states
- Phase 3: Codex monitor route view: identity -> project -> thread/session -> bridge/watch process
- Phase 4: lightweight local GUI showing roster, presence, routes, recent traffic, and warnings
Notes
This is related to, but separate from, Codex monitor correctness. Monitor safety still needs a strict no-cross-injection invariant. This issue is about making the network observable enough to operate safely.
Problem
Using agmsg across multiple desktop agents is currently hard to operate because the user cannot see the communication state as a whole.
The scripts can show a registered team roster and per-agent history/inbox, but they do not provide a single operator-facing view for:
off,turn,monitor)This becomes especially painful after a PC reboot. Some Claude-side sessions may recover automatically, others may not. Codex monitor is more fragile because delivery targets an ephemeral TUI thread, so recovery requires safe identity-to-thread rebinding. Without visibility, the user cannot tell whether the network is healthy or silently degraded.
Desired UX
Provide an operator dashboard or status command that answers, at a glance:
turnvsmonitorvsoff?A first version can be CLI/TUI/HTML; a polished GUI can come later.
Initial scope for the fork
Keep this local to the fork while experimenting. Do not use upstream issues as internal scratch.
Possible phased approach:
agmsg status/agmsg doctorstyle summary using existing official script dataNotes
This is related to, but separate from, Codex monitor correctness. Monitor safety still needs a strict no-cross-injection invariant. This issue is about making the network observable enough to operate safely.