agentcli

name	agentcli
description	Use this skill to inspect or change ChatLuna agent admin state — skills, sub-agents, tools, MCP servers, MCP tools, or permission rules. The skill edits a working copy of the agent config inside the sandbox; the user must run `chatluna.agent.sync` to write changes back to the host instance.

This skill ships a CLI that runs inside the sandbox via the bash tool. It reads and writes a working copy of the agent config materialized next to the script (config.json in this skill folder). Local edits do not take effect until the user runs chatluna.agent.sync on the host instance.

How to invoke

Run the script through the bash tool:

node "<skill-dir>/bin/agentcli.cjs" <command> [args...]

<skill-dir> is the location reported in the skill prompt (or by agentcli.cjs --skill-dir). Always quote paths.

Read commands

node bin/agentcli.cjs show overview
node bin/agentcli.cjs show skills
node bin/agentcli.cjs show skill <name|id>
node bin/agentcli.cjs show subagents
node bin/agentcli.cjs show subagent <selector>
node bin/agentcli.cjs show tools
node bin/agentcli.cjs show tool <name>
node bin/agentcli.cjs show mcp servers
node bin/agentcli.cjs show mcp server <name>
node bin/agentcli.cjs show mcp tools
node bin/agentcli.cjs show mcp tool <name>

Write commands

# skills
node bin/agentcli.cjs enable skill <name|id...>
node bin/agentcli.cjs disable skill <name|id...>
node bin/agentcli.cjs remove skill <name|id...>

# sub-agents
node bin/agentcli.cjs enable subagent <selector...>
node bin/agentcli.cjs disable subagent <selector...>
node bin/agentcli.cjs remove subagent <selector...>
node bin/agentcli.cjs set subagent <selector...> <tools|skills|mcp|computer> <all|allow|deny|inherit> [items...]

# tools
node bin/agentcli.cjs enable tool <name...> [--main]
node bin/agentcli.cjs disable tool <name...> [--main]
node bin/agentcli.cjs set tool <name...> <enabled|main|chatluna|character|group|private|authority|subagents> <value> [items...]

# mcp
node bin/agentcli.cjs set mcp tool <name...> enabled <bool>
node bin/agentcli.cjs save mcp server <name> json '<json>'
node bin/agentcli.cjs remove mcp server <name...>

Each write rewrites config.json in place and keeps a config.json.bak.

Workflow

1. Run node bin/agentcli.cjs show overview to understand the current state.
2. For each change, run the matching write command. Verify with another show ... after the edit.
3. When the edits are ready, ask the user to run chatluna.agent.sync. That validates the working copy and atomically replaces the live config.
4. Authoring new skill or sub-agent files still requires loading skill-creator or sub-agent-creator first. Place those files under the sandbox skill / sub-agent directories shown by show overview. The same chatluna.agent.sync command flushes them back to the host.
5. Never invent paths. Read the directories from show overview.

Notes

• This script does not load the live ChatLuna service. show output is derived from the working copy of config.json plus a scan of the sandbox skill / sub-agent directories. Treat it as authoritative for config; for runtime state, the host chatluna.agent console remains the source of truth.
• Removing a skill or sub-agent here only removes its config entry. To erase the actual files, delete them from the corresponding sandbox directory before running chatluna.agent.sync.
• If the bash tool is unavailable, say so instead of guessing.