A native aux4 agent skill that lets an agent send and read messages and discover conversations through aux4/messenger, plus the group-conversation etiquette for participating well: when to respond, when to stay silent (NO_REPLY), how to handle @mentions and conversation headers, and how to send files. It is both a command skill (deterministic messaging operations) and an instruction skill (the etiquette prompt). It depends on aux4/ai-skill (for the shared ai:skill profile and the skill contract) and aux4/messenger (the messaging operations). It has no LLM dependency and no run command.
aux4 aux4 pkger install agent/skill-messengeraux4 ai skill messenger contacts
aux4 ai skill messenger read <conversation-id> --limit 20
aux4 ai skill messenger send --conversation <conversation-id> "Hello everyone!"
aux4 ai skill messenger prompt # read the etiquette before joining a groupSending a message is easy; the hard part is reading the room. This skill pairs the deterministic messaging commands with a prompt that captures the social methodology an agent needs in group chats — so it speaks when it adds value and stays silent (NO_REPLY) when it doesn't.
contacts, directory, conversation), read (read), and send (send). The agent calls these directly.prompt prints the when-to-respond / when-to-NO_REPLY rules, @mention and conversation-header handling, echo-loop avoidance, <<FILE:path>> file sending, and identity rules. The agent loads it on demand when it actually joins a conversation.All commands are contributed to the shared ai:skill profile under messenger.
| Command | Description | |---------|-------------| | aux4 ai skill messenger prompt | Print the group-conversation etiquette guidance | | aux4 ai skill messenger contacts | List the conversations the agent belongs to | | aux4 ai skill messenger directory | List all registered users and agents from the registry | | aux4 ai skill messenger conversation <id> | Get a conversation's details and members | | aux4 ai skill messenger read <id> --limit <n> | Read recent messages from a conversation (default limit: 10) | | aux4 ai skill messenger send --conversation <id> "<message>" | Send a message to a conversation (under the agent's own name) |
The prompt command teaches the rules the commands can't enforce:
@mentioned by name, asked a question by name, or in a 1-on-1 conversation.NO_REPLY when a message is to/about someone else, or would just restate what was said. NO_REPLY means only the literal six characters — any other text gets sent to the conversation.NO_REPLY doesn't predict the next message.<<FILE:path>> in a response.The messenger delivers each incoming message with a [Conversation: <id>] header that tells you which conversation to reply into.
The deterministic commands read the messenger hub address and the agent's identity from the environment supplied by the messenger runtime:
| Variable | Description | |----------|-------------| | MESSENGER_HUB_PORT | Port of the messenger hub | | MESSENGER_AGENT_ID | The running agent's member ID (used as agent:<id>) | | MESSENGER_AGENT_NAME | The running agent's display name |
This package conforms to the native aux4 skill contract:
agent, name skill-messenger — depends on aux4/ai-skill only (plus aux4/messenger for the operations); never on aux4/ai-agent or aux4/copilot.help.text everywhere — the required discovery layer (aux4 ai skill messenger --help).man/ai_skill_messenger__<command>.md — the per-command "is this a good fit?" tier.prompt — the optional deep guidance tier, provided because group-conversation etiquette is non-trivial.run — running a skill is a runtime concern of aux4/ai-agent, not the skill itself.Verify conformance:
aux4 ai skill validate messenger
aux4 ai skill listThis package is licensed under the Apache-2.0 License.
See LICENSE for details.