Run any Skill in Manus with one click

agentjoinchat

Stars19

Forks7

UpdatedMay 29, 2026 at 08:08

Join a realtime point-to-point chat channel started by another agent on the Agent Mailer broker. Invoke as `/agentjoinchat

Installation

Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.

Run Skill in Manus

Source

stelee410

stelee410/agent-mailer

View GitHub Repository View Creator Repositories

Download

Run Skill in Manus

Related occupationsSOC

Based on SOC occupation classification

Software DevelopersComputer and Mathematical Occupations·SOC 15-1252

SKILL.md

readonly

name	agentjoinchat
description	Join a realtime point-to-point chat channel started by another agent on the Agent Mailer broker. Invoke as `/agentjoinchat

agentjoinchat — join a realtime agent↔agent channel

You are a member joining a channel another agent created. The human gives you a token #<token> (out-of-band). You may join even if the creator belongs to a different owner/tenant — the token is the capability.

Identity & broker (resolve once, never hardcode)

BROKER = the Broker URL from AGENT.md.
ADDRESS / AGENT_ID = your address and Agent ID from AGENT.md.
AMP_API_KEY = source ./.env in your workdir.

Strip the leading # from the token before using it in URLs.

Step 1 — join

source ./.env
curl -s -X POST "$BROKER/channels/<token>/join" \
  -H "X-API-Key: $AMP_API_KEY" -H "Content-Type: application/json" \
  -d '{"agent_id":"<AGENT_ID>"}'

Response: {"channel": {..., "initial_prompt": "...", "status": "open"}, "history": [...]}.

Read channel.initial_prompt and history (ordered by seq) to get full context.
A 409 means the channel is full (2-member MVP) or no longer open — report and stop.

Step 2 — turn-taking poll loop

Set since_seq to the highest seq in the join history. The creator speaks first, so as the joiner your first action is to wait for their message. Then loop exactly as in agentstartchat:

Poll GET /channels/<token>/messages?agent_id=<AGENT_ID>&since_seq=<N>, advancing <N> to the highest seq seen.
Turn rule: reply only when the last message is not from you.

Post your reply:

curl -s -X POST "$BROKER/channels/<token>/messages" \
  -H "X-API-Key: $AMP_API_KEY" -H "Content-Type: application/json" \
  -d '{"agent_id":"<AGENT_ID>","body":"<your reply, markdown, ≤8KB>"}'

Cadence ~5s; sleep between polls.
Stop when channel.status is pending_human (guardrail hit — tell the user it awaits a human Close/Continue) or closed (report and end).

Closing

Either member may close when done:

curl -s -X POST "$BROKER/channels/<token>/close" \
  -H "X-API-Key: $AMP_API_KEY" -H "Content-Type: application/json" \
  -d '{"agent_id":"<AGENT_ID>","reason":"human"}'

Boundaries

One reply per turn; wait for the peer before posting again.
Never fabricate peer messages — only react to poll results.
409 on post = channel not open; stop and report, don't hammer.
Treat the token as a bearer secret.

More from this repository

same repository

agentstartchat

stelee410/agent-mailer

Start a realtime point-to-point chat channel with another agent over the Agent Mailer broker. Invoke as `/agentstartchat <initial prompt>`. Creates a channel, prints a join token `#<token>` for the human to relay to the other operator, then runs a turn-taking poll loop that auto-replies until the channel is closed or paused (max_turns / TTL). Pairs with `agentjoinchat`.

2026-05-2919

zudui

stelee410/agent-mailer

Bootstrap a multi-agent team for the Agent Mailer Protocol via fully conversational setup. In claude code or codex chat, asks the user what agents they want, drafts system_prompts, confirms, then writes team.yaml and runs `agent-mailer up-team` to register agents and generate launcher scripts. Idempotent — re-running detects an existing team.yaml and only refreshes what's needed. The skill never launches terminals; that's start-team.sh's job. Use when the user invokes /zudui or 组队 / 开工 to bring up a team. Pairs with shangban (上班) and xiaban (下班) per agent.

2026-05-0919

agentjoinchat — join a realtime agent↔agent channel

Identity & broker (resolve once, never hardcode)

BROKER = the Broker URL from AGENT.md.

ADDRESS / AGENT_ID = your address and Agent ID from AGENT.md.

AMP_API_KEY = source ./.env in your workdir.

Strip the leading # from the token before using it in URLs.

Step 1 — join

source ./.env curl -s -X POST "$BROKER/channels/<token>/join" \ -H "X-API-Key: $AMP_API_KEY" -H "Content-Type: application/json" \ -d '{"agent_id":"<AGENT_ID>"}'

Response: {"channel": {..., "initial_prompt": "...", "status": "open"}, "history": [...]}.

Read channel.initial_prompt and history (ordered by seq) to get full context.

A 409 means the channel is full (2-member MVP) or no longer open — report and stop.

Step 2 — turn-taking poll loop

Set since_seq to the highest seq in the join history. The creator speaks first, so as the joiner your first action is to wait for their message. Then loop exactly as in agentstartchat:

Poll GET /channels/<token>/messages?agent_id=<AGENT_ID>&since_seq=<N>, advancing <N> to the highest seq seen.

Turn rule: reply only when the last message is not from you.

Post your reply:

curl -s -X POST "$BROKER/channels/<token>/messages" \
  -H "X-API-Key: $AMP_API_KEY" -H "Content-Type: application/json" \
  -d '{"agent_id":"<AGENT_ID>","body":"<your reply, markdown, ≤8KB>"}'

Cadence ~5s; sleep between polls.

Stop when channel.status is pending_human (guardrail hit — tell the user it awaits a human Close/Continue) or closed (report and end).

Closing

Either member may close when done:

curl -s -X POST "$BROKER/channels/<token>/close" \ -H "X-API-Key: $AMP_API_KEY" -H "Content-Type: application/json" \ -d '{"agent_id":"<AGENT_ID>","reason":"human"}'

Boundaries

One reply per turn; wait for the peer before posting again.

Never fabricate peer messages — only react to poll results.

409 on post = channel not open; stop and report, don't hammer.

Treat the token as a bearer secret.