The plugin's UserPromptSubmit and SessionStart hooks call `claude-mailbox` with no --url flag, so they previously always hit the hardcoded http://127.0.0.1:47822/mcp default. If port 47822 was held by another local service (e.g. ClaudeDo), the daemon couldn't bind there and every hook was talking to the wrong process. CLI default for --url now resolves to $CLAUDE_MAILBOX_URL when set, falling back to http://127.0.0.1:47822. Doctor gained a Step 2 that probes /health on 47822, identifies foreign occupants, picks a free port, writes both ~/.claude-mailbox/mailbox.json and the CLAUDE_MAILBOX_URL entry in .claude/settings.json env so the hooks follow along automatically. Also adds a fallback hint when Windows schtasks /Create fails with Access is denied (Group Policy restricts non-admin task creation): run install-autostart from an elevated shell, or accept an ephemeral serve for the current session.
5.5 KiB
description, allowed-tools
| description | allowed-tools |
|---|---|
| Diagnose and auto-fix the Claude-Mailbox setup (binary install, port-conflict detection, daemon autostart, smoke test, optional base-prefix). | Bash, Read, Edit, Write |
You are running the Claude-Mailbox doctor. Walk through these checks in order. After each step, print a one-line ✓ / ✗ with the action you took. End with a summary block.
Use Bash only for claude-mailbox subcommands, npm, where/which, and HTTP probes. Use Read/Edit/Write for .claude/settings.json and mailbox.json. Never run sudo automatically — if elevation is needed, stop and ask.
Step 1 — daemon binary on PATH
Run: claude-mailbox --version
-
Exit 0 → ✓ record the version. Continue.
-
Command not found → binary missing. Install path:
Platform Command Windows npm install -g @kuns/claude-mailbox(no admin)macOS / Linux npm install -g @kuns/claude-mailbox(may fail with EACCES — never run sudo automatically; ask the user)Prerequisite:
npm config get @kuns:registrymust point athttps://git.kuns.dev/api/packages/releases/npm/. If not:npm config set @kuns:registry=https://git.kuns.dev/api/packages/releases/npm/After install, re-run
claude-mailbox --version. If it still fails, stop and report.
Step 2 — port-conflict check (before autostart!)
Default port is 47822. Probe whether anything is already on it:
curl -sf http://127.0.0.1:47822/health
- Returns a JSON body with
"status":"ok"and aversionfield that matchesclaude-mailbox --version→ it's already our daemon, ✓ skip to Step 4. - Returns 200 with
"status":"ok"but a differentversion→ it's an older claude-mailbox; treat as running, ✓. - Returns non-200, non-JSON, or any other foreign response → port conflict. Some other process owns 47822.
- Connection refused → port is free, ✓ continue to Step 3.
If port conflict detected:
- Tell the user which process holds the port (Windows:
Get-NetTCPConnection -LocalPort 47822 | Select-Object OwningProcess, thenGet-Process -Id <pid>; macOS/Linux:lsof -i :47822). - Pick a free port. Default suggestion: 47900. Verify it's free:
curl -sf http://127.0.0.1:47900/healthshould fail with connection refused. - Read
~/.claude-mailbox/mailbox.json(create empty{}if missing) and merge{"port": <chosen>}. Write back. - Also write the override into
.claude/settings.jsonenv so the plugin's hooks find the right URL:Merge into existing env, preserving other keys."env": { "CLAUDE_MAILBOX_URL": "http://127.0.0.1:<chosen>" } - Mark
restart_needed = true.
Step 3 — daemon autostart and running state
Run: claude-mailbox status
Running→ ✓ continue.Stopped→claude-mailbox start, re-check.NotInstalled→claude-mailbox install-autostart, thenclaude-mailbox start, re-check.
If install-autostart fails with "Access is denied" on Windows: Group Policy may block non-admin schtasks /Create. Two fallbacks:
- Tell the user to run
claude-mailbox install-autostartfrom an elevated PowerShell themselves (one-time). - For this session, run
claude-mailbox serveas a background process so the rest of the doctor's checks can pass — the daemon won't survive logoff, but that's fine for verification.
If status doesn't reach Running after the fallback, stop and report.
Step 4 — health probe
Hit http://127.0.0.1:<port>/health (use the configured port, not necessarily 47822). Expect a JSON body with "status":"ok" AND a version matching claude-mailbox --version. If unreachable or version mismatch, stop and report.
Step 5 — mailbox identity (base prefix)
No prompt by default. Each Claude Code session gets a unique mailbox name auto-derived from its session_id (e.g., claude-a8b3c1d2).
Read .claude/settings.json and look for env.CLAUDE_MAILBOX_NAME.
- If set → ✓ "Mailbox prefix is
<X>." (real name will be<X>-<short_session_id>). - If unset → ✓ "Mailbox name will be auto-derived (
claude-<short_session_id>)."
Ask once: "Want to flavor your mailbox names with a memorable prefix (e.g., backend, frontend)? (yes / no / <name>)"
On yes/explicit name: merge env.CLAUDE_MAILBOX_NAME = <name> into .claude/settings.json, preserving other keys. Mark restart_needed = true.
Step 6 — smoke test
Use two ephemeral names — we don't need the real session name here:
claude-mailbox send --from doctor-probe-a --to doctor-probe-b --body "ping from doctor"
claude-mailbox check --name doctor-probe-b
(If the port was changed in Step 2, pass --url http://127.0.0.1:<port> to both.)
The check output must be a JSON array with one message: from: doctor-probe-a, body matches. ✓ on success, ✗ otherwise.
Step 7 — summary
Claude-Mailbox doctor
binary: <version>
daemon: Running (port: <port>, what you did if anything)
health: ok
port conflict: none | resolved (moved from 47822 to <port>)
base prefix: <name from settings, or "auto-derived (anonymous)">
smoke test: passed | failed
restart hint: yes if restart_needed, otherwise no
End with one of:
- All ✓ and no restart needed → "Setup is healthy. Your mailbox name this session will be revealed by the SessionStart hook on next session start."
- All ✓ and restart needed → "Restart Claude Code (or open a new session) so the SessionStart hook picks up the new env values."
- Anything ✗ → "Setup incomplete: ."