Moshe Avdiel

Posted on Sep 11 Originally published at github.com

Claude Code Finished the Turn Blind — Mirror Phone live device leaf on the Admin Dashboard Path

I have sat next to the admin-dashboard wall at 21:03 while Claude Code was this close to doing the wrong thing.

Not a model failure. A posture failure.

The host had started with device-live frozen in env / argv / a skill file. Someone on the floor said, out loud, the tile is not a rumor. it is the same tree. Claude Code was still holding the old process. The only “safe” move anyone trusted was:

Kill Claude Code (or its MCP server)
Edit a file
Restart the host
Lose the operator's last ten minutes of diagnosis

I have watched that restart more times than I want to admit. It feels responsible. It is a ceremony. status tile green while the agent host started earlier does not wait for ceremonies.

The Aha: Mirror Phone live device leaf is not a binary you reboot. It is a function that should read live posture from Kiponos.io on every call. The host stays up. The leaf moves.

The problem: Mirror Phone live device leaf lived in the process, not in the turn

Claude Code is good at calling tools. It is not born with a shared, instant, restart-free control plane.

So teams hide Mirror Phone live device leaf in the only places agent frameworks actually ship:

Where the gate hid	What you restart	What you lose
MCP server env / argv	The MCP process	Open tool sessions
Skill file on disk	The agent turn, sometimes the host	Context the model already paid for
Host-local JSON	Whatever still has the file open	Agreement between two agents
Hard-coded `if` on `device-live`	A release	The incident clock

The admin-dashboard wall already knew. Claude Code did not, because it had started earlier.

That is the missing piece: the framework gave you tools. It did not give you a live hub.

What teams believe

Belief	Production
The skill file is the source of truth	Skills instruct. They do not fan out
Put the SDK in the SPA	Connect tokens do not belong in a browser
Feature flags cover this	Flags are another product, another delay
Paste the new device-live into chat	Two agents, two pastes, two lies

The Aha: local get, live write, host stays up

Kiponos holds a nested tree. Java and Python SDKs keep the latest values in memory, patched over WebSocket deltas. The hot path inside a Claude Code tool is a local get — no HTTP RTT per admin dashboard lookup.

Hub leaf for this essay:

examples/agentic-dev-1101-am-mirror-live/device-live = live

Runnable proof: examples/java/agentic-dev-1101-am-mirror-live

Public SDKs: Java, Python, plus React/Angular server peers (createFromEnv). Never put Connect tokens in the SPA.

Config tree (admin dashboard + peers)

examples/
  agentic-dev-1101-am-mirror-live/
    device-live: live          # Mirror Phone live device leaf
apps/
  admin-dashboard/
    live:
      device-live: live

Integration — Java hot path

Kiponos kip = Kiponos.createForCurrentTeam();
Folder gate = kip.getRootFolder()
        .folderOrCreate("examples")
        .folderOrCreate("agentic-dev-1101-am-mirror-live");
if (!gate.hasKey("device-live")) {
    gate.set("device-live", "live");
}
String posture = gate.get("device-live");
// Claude Code tool: refuse the dangerous call when posture moved

Same leaf from a Python tool (Claude Code just calls it):

from kiponos import Kiponos

k = Kiponos.connect(quiet=True)  # env: KIPONOS_ID, KIPONOS_ACCESS, KIPONOS
try:
    posture = k.get("examples/agentic-dev-1101-am-mirror-live/device-live", "live")
    if str(posture) == "live":
        raise PermissionError("Mirror Phone live device leaf gated live — host not restarted")
finally:
    k.disconnect()

The Claude Code process does not recycle. The next tool call already sees the dashboard edit.

Real scenarios

Event	Without Kiponos	With Kiponos
Status tile green while the agent host started earlier	Restart Claude Code; lose the operator's last ten minutes of diagnosis	Set `device-live` live; next Claude Code tool call already obeys
Peer host still on old device-live	Paste the value into the other chat	One hub leaf; both processes `get()` locally
admin-dashboard wall shows the new posture	Claude Code started earlier so it writes anyway	Dashboard and tool share the same memory tree
Incident over, resume	Another Claude Code restart	Set `device-live` back; session continues
Senses firing while the turn finishes blind	Two ceremonies, two lost threads	Same tree, two products, no paste

Performance (this path, not a generic table)

Claude Code tool get() is an in-process map lookup after bootstrap.
One WebSocket per process lifetime — not per admin dashboard line.
A dashboard edit is a delta of device-live, not a config-file reload.
You do not pay model tokens to “please restart Claude Code.”
A second host converges without a third paste onto senses firing while the turn finishes blind.

Compare to alternatives

Approach	Honest fit	Why it still restarts
Env file + Claude Code reboot	Simple at 09:00	The freeze is at 21:03
Skill markdown as policy	Good instructions	Not a live bus
Redis poll inside the tool	Shared, but RTT on the hot path	You invented a hub with worse UX
Feature-flag SaaS	Product experiments	Rarely session-safe for Claude Code
`@RefreshScope` / actuator	JVM apps	Does not restart Claude Code

When not to use Kiponos

Situation	Why
Tool schema itself changed (new argument)	That is a code/Claude Code restart
Secret rotation of Connect tokens	Credentials are not live knobs
One-off local script, no peers	A hub is overkill
Browser-only “SDK in the SPA”	Forbidden — tokens leak or defaults lie

Rehearsal beats slides

In staging: set a painful device-live, prove Claude Code recovers without a host kill, prove clamps reject nonsense, prove last-known-good when the hub is firewalled. That drill ends half the architecture arguments about Mirror Phone live device leaf.

Why Claude Code is the wrong restart target

Claude Code is good at calling tools. It is not a control plane. Killing it to flip device-live teaches the on-call that judgment requires a process ID. The admin-dashboard wall already disagrees.

What the admin dashboard operator actually said

At 21:03 someone said, out loud: the tile is not a rumor. it is the same tree. That sentence is the whole product. If it cannot land in the running Claude Code process in seconds, you do not have posture. You have a wiki.

Getting started (15 minutes)

TeamPro on kiponos.io → Connect → KIPONOS_ID / KIPONOS_ACCESS / profile ['my-app']['v1.0.0']['dev']['base'].
Clone github.com/kiponos-io/kiponos-io.
cd examples/java/agentic-dev-1101-am-mirror-live && cp kiponos.local.env.example kiponos.local.env
./gradlew test run — prints examples/agentic-dev-1101-am-mirror-live/device-live=...
In the dashboard, change device-live. Keep the process up. No rebuild.
Point your Claude Code tool at the same leaf. Do not ship a new server binary to flip Mirror Phone live device leaf.

The moral

If flipping Mirror Phone live device leaf requires restarting Claude Code, you do not have a gate. You have a hope with a process ID.

Agent frameworks already know how to call tools. Kiponos is the live hub they do not ship — so the admin-dashboard wall can change its mind without killing the session.

How to try: examples/java/agentic-dev-1101-am-mirror-live and ./gradlew test.

DEV Community