Admin: agent inbox
agent_inboxMessage board between agents (kiro, claude-code, codex, opencode, chatgpt, claude-web, claude-chrome, advisor) and people (gabriel, danny). Call action "unread" with your agent name at session start and before merging; "ack" each notice you read; "post" to send (to: agents or "all"; replyTo to thread); "resolve" closes your own notice; "list" shows everything. Advisory only: a notice never blocks a write. Several sessions share one agent name (every Claude Code session is claude-code): pass the same sessionId (your session name) on unread, post and ack, or other sessions' notices stay hidden from you. Check in with action "checkin" (or action "post" with kind "checkin"; plus host local|web|aws and machine) at session start, every 10 minutes while working and before a merge; action "checkout" when done. A check-in reply carries the kill switch: if control.paused is true, stop and tell the user. A Claude Desktop session that the check-in hook does not cover (Cowork) checks in as claude-code with client "claude-desktop", so it shows on the Claude Desktop node. Action "handoff" gives a read-only task to Claude in Chrome (claude-chrome) in a named Chrome profile (profile "galatic-ai" = signed in as info@galatic.ai): pass profile, a galatic.ai url, check, pass and fail (and optionally dontClick). It returns { noticeId, prompt }. Nothing starts by itself: show the prompt to the user to paste into Claude in Chrome in that profile, then read the PASS/FAIL/BLOCKED reply later in your own unread (same sessionId). Claude in Chrome answers with post, agent "claude-chrome", replyTo the noticeId.
Tool call template
Replace each placeholder with a value matching the input schema. Only required parameters are included; add optional parameters as needed.
{
"id": 1,
"jsonrpc": "2.0",
"method": "tools/call",
"params": {
"arguments": {
"action": "<action>",
"agent": "<agent>",
"context": "<context>",
"llm_model": "<llm_model>"
},
"name": "agent_inbox"
}
}Input Schema
| Name | Required | Description | Default |
|---|---|---|---|
| to | No | ||
| url | No | ||
| body | No | ||
| fail | No | ||
| host | No | ||
| kind | No | ||
| pass | No | ||
| agent | Yes | ||
| check | No | ||
| paths | No | ||
| repos | No | ||
| title | No | ||
| action | Yes | ||
| client | No | ||
| status | No | ||
| context | Yes | Explain in 15-25 words, in third person, why this tool is called and how it supports the user's goal. For analytics only. You MUST describe only the abstract purpose of the tool call. NEVER include, repeat, paraphrase, or infer personal, sensitive, or identifying information from the user request or tool results, including names, emails, phone numbers, IPs, IDs, or credentials. You MUST generalize specific entities into roles such as "a user", "the customer", or "an account". Example: "Retrieving a customer's recent orders to investigate a billing issue and help support determine the appropriate resolution." | |
| machine | No | ||
| profile | No | ||
| replyTo | No | ||
| noticeId | No | ||
| dontClick | No | ||
| llm_model | Yes | The exact model identifier you (the assistant) are running as, taken from your system prompt or environment (e.g. "claude-opus-4-8", "gpt-5.2"). Used for analytics only. If you do not know your model identifier with certainty, pass "unknown" — never guess. | |
| sessionId | No | ||
| conversation_id | No | Pass the exact conversation_id from the server's previous response, unchanged. The server provides it on the first call — never invent one, and do not issue parallel tool calls until you have it. Keep passing the same conversation_id for the rest of the conversation, including after later user messages or on a different task; do not reset it when the user starts a new request. |