Fruxon
Build, run, and debug Fruxon agents from Claude Code.
Fruxon is a platform for designing, testing, deploying, and monitoring AI
agents. This plugin teaches Claude to drive the fruxon command-line tool,
so you can ask for an agent in plain language and have Claude draft it,
test it against real inputs, publish a revision, and work out why a run
went wrong from its execution trace.
Requirements
-
A Fruxon account and workspace
-
The
fruxonCLI, installed and signed in:uv tool install fruxon # or: pipx install fruxon fruxon login
Then install the plugin inside Claude Code:
/plugin install fruxon --marketplace fruxon-ai/fruxon-agent-plugins
The plugin does not install the CLI for you. If it is missing or older than the skills, a check at the start of each session tells Claude, and Claude asks you to install or upgrade it.
Skills
| Skill | Use it to |
|---|---|
fruxon-meet | Get oriented on Fruxon's concepts and the CLI's main commands |
fruxon-build-agent | Author, test, publish, and deploy an agent revision |
fruxon-use-integrations | Wire connected tools (Slack, GitHub, Jira, Drive…) into an agent |
fruxon-create-integration | Create a new integration with API or Python code tools |
fruxon-debug-trace | Diagnose a failed or surprising run from its trace |
fruxon-agent-mode | Read the CLI's JSON output, NDJSON streams, exit codes, and error envelopes |
Claude loads a skill when your request matches it, or you can ask for one by name.
What the plugin runs and sends
The plugin itself contains only instructions (the skills). It runs no hooks, stores nothing and makes no network calls of its own.
- The
fruxonCLI, which Claude runs when you ask it to, talks to the Fruxon API (https://api.fruxon.comby default) as the workspace you signed in to withfruxon login. It sends what the command you asked for needs: agent definitions, test inputs, integration settings. - Credentials stay with the CLI:
fruxon loginkeeps your API key in the operating system's keychain, or in~/.fruxonwhere no keychain is available. Integration secrets, such as a third-party access token, are entered by you on app.fruxon.com, not passed through Claude. - Usage telemetry: once signed in, the CLI reports which command ran and
whether it succeeded to the Fruxon API, never argument values, output, or
your key. Turn it off with
fruxon config set telemetry falseorDO_NOT_TRACK=1. Details are in the privacy policy.
Support
Contact the Fruxon team at fruxon.com/contact. Bugs in the plugin itself can also go to github.com/fruxon-ai/fruxon-agent-plugins/issues.
License
MIT