Channels

Channels — messaging status + launch

Diagnose messaging channel plugins and give the user the exact command to load them. This skill does NOT install channels (use /agent:messaging) or authenticate them (per-channel skills like /whatsapp:configure) — it reports state and hands back a launch command.

This is a CORE feature. See docs/channels.md for details.

Dispatch

List flow

1. Call channels_detect({ format: "table" })
2. Print the returned text verbatim — it already has the table, next-steps hints, the launch command, and a ⚠️ Runtime block when a channel server is up but not delivering inbound (see Runtime warnings below)
3. If any channels show ❌ under Installed and the user is new, remind: "Install a channel with /agent:messaging <name>"

Status flow (single channel)

1. Call channels_detect({ format: "json" })
2. Find the entry where name matches (case-insensitive)
3. If not found, respond: "I don't track a channel called <name>. Known channels: whatsapp, telegram, discord, imessage, slack, fakechat."
4. Otherwise print a compact block:

📡 <label>
   kind:           <development|official|integration>
   installed:      <✅ | ❌>  (<detail>)
   authenticated:  <✅ | ❌ | ⏸️>  (<detail>)
   active:         <❓ | ❌>  (<detail>)
   next:           <setupHint>

5. If the entry has a runtime field, surface it — see Runtime warnings.

Runtime warnings

The JSON entry may carry a runtime field read from the channel's live status.json (currently WhatsApp). When runtime.problem is true, the channel server is running but inbound is NOT reaching this session — the classic "I see 'typing…' on my phone but the agent never answers" failure, usually after an in-session update/reload or when a second session (interactive, service, or a still-alive scheduled-task session) holds the single-device lock.

When runtime.problem is true, do NOT report the channel as healthy. Lead with the warning, verbatim from runtime.detail, and then runtime.remediation if present. For example:

⚠️ WhatsApp inbound is NOT active in this session — another instance (PID <runtime.holderPid>) holds the single-device lock, so incoming messages go to that session, not this one.
Fix: <runtime.remediation>

Key runtime.status values: idle_other_instance (lock held by another session — the main one), logged_out (re-link needed), lock_error (filesystem/PID-file problem). Never tell the user to run /whatsapp:configure reset for idle_other_instance — it's a lock-ownership problem, not a link problem; the fix is to get down to one session. On claude-whatsapp ≥ 1.21 the waiting session takes over the lock automatically (within ~15 s of the holder exiting), so closing the extra session is enough; on older channel versions the waiting session stays idle forever and needs a full relaunch after the holder is closed. See claude-whatsapp docs/troubleshooting.md → "You see 'typing…' but get no reply".

Live voice setup

For attaching ClaudeLive to this existing agent, use /agent:live setup. channels_detect can describe launch flags after opt-in, but does not enable the bridge or configure credentials. /agent:live status distinguishes saved settings, running listener and native verification.

Launch flow

1. Call channels_detect({ format: "launch" }) (with the appropriate flags from the user's command)
2. Print the command in a code block
3. Remind the user: "Preserve these channel arguments when preparing /agent:service install; it must receive the existing launcher arguments explicitly."

If the user included --skip-permissions, ADD a warning above the command:

⚠️ The --dangerously-skip-permissions flag pre-approves every tool call. Use only for background/service runs.

Response style

• CLI / WebChat: full table
• WhatsApp / Telegram / Discord: collapse to one line per channel using the icons, e.g. 📡 WhatsApp:✅✅❓ · Telegram:✅❌❌ · iMessage:✅✅❓. The full table is too wide for mobile.

Never

• Do NOT try to install or authenticate channels from this skill. Redirect to /agent:messaging or the channel's own skill.
• Do NOT lie about active state — the tool says ❓ unknown when it can't tell. Pass that through; the user can verify with /mcp or by sending a message.
• Do NOT generate a launch command with --skip-permissions unless the user explicitly asked for it.

References

• docs/channels.md — full reference
• lib/channel-detector.ts — pure detection logic
• skills/messaging/SKILL.md — channel installation
• skills/service/SKILL.md — always-on, consumes the launch command