Agent Skills
Teach your AI agent how to use Agent Messenger CLIs effectively.
Agent Messenger includes an agent-messenger entry skill for selecting a platform and acting identity, resolving conversations, verifying delivery, and handling reply follow-ups. It routes to the existing platform skills, so agents do not need every platform's instructions in context.
What Are Agent Skills?
When you give an AI agent access to a CLI tool, it needs to know:
- What commands are available
- How to format arguments
- What output to expect
- Best practices and common patterns
Agent Skills provide this knowledge in a structured format that AI agents can understand.
Installation
For AI agents
When asked to use Agent Messenger, install the entry skill in the current project, unless it is already installed:
npx -y skills add agent-messenger/agent-messenger --skill agent-messenger --yesRead its SKILL.md at the path reported by the installer. To target a specific supported agent, append --agent codex, for example. Do not add --global unless requested.
The entry skill explains how to install and read only the required platform skill, for example:
npx -y skills add agent-messenger/agent-messenger --skill agent-slackbot --yesSkill installation does not install or authenticate the CLI, grant permission to send messages, or start listeners. Install the CLI through the quick start if necessary, and preserve the user's account and approval constraints.
Skills CLI (interactive selection)
To choose platform skills interactively instead:
npx -y skills add agent-messenger/agent-messengerInstallation paths depend on the selected agent and scope. See the installer's output and skills.sh for details.
Claude Code
Add Agent Messenger skills to Claude Code:
# From terminal
claude plugin marketplace add agent-messenger/agent-messenger
claude plugin install agent-messengerOr from within Claude Code:
/plugin marketplace add agent-messenger/agent-messenger
/plugin install agent-messengerHow It Works
Once installed and read, the skills guide your agent to:
- Discover the CLI - Check the installed version's commands and flags
- Select the acting identity - Reuse the intended user or bot account; set up authentication only when authorized
- Resolve the destination - Identify the recipient, conversation, and original thread
- Perform the requested operation - Send, read, reply, or react within the task's scope
- Verify the result - Retain the receipt, reconcile uncertain writes, and verify a reply-receiving mechanism before promising automatic follow-up
Example Interaction
With Agent Messenger skills installed, you can ask your AI agent:
"Send a message to #general saying the build passed"
The agent will:
- Select the intended Slack user account and verify its identity
- Resolve the exact
generalchannel and confirm the installed send syntax - Execute the authorized
agent-slack message send general "The build passed" - Confirm the returned delivery result, without treating it as proof the recipient read it
Skill Contents
Each platform skill includes:
| Section | Description |
|---|---|
| Authentication | How to extract and manage tokens |
| Commands | Full command reference with examples |
| Output Format | Expected JSON structure for parsing |
| Error Handling | Common errors and how to recover |
| Best Practices | Tips for reliable agent workflows |