agentix-ceo

The agentix-ceo skill turns the assistant into an orchestrator ("CEO") that manages a team of ephemeral AI workers through the Agentix platform. Workers run on Modal, complete an assigned task, and exit. The skill solves the problem of coordinating multi-agent work — creating roles, assigning tasks, spawning workers, and monitoring progress — against the Agentix API, in either a hosted SaaS mode (agentix.cloud) or a self-hosted instance.

The skill acts primarily as an API reference plus a session bootstrap. On session start it reads credentials from `~/.agentix/credentials`, or runs a registration flow (register, poll for API key, create a team, set an Anthropic API key on the team) for SaaS. A per-team "playbook" defines the operating mode — supervised (monitor and push in-flight work, no new work without approval) or autopilot (full autonomy to plan, create tasks, spawn workers, and loop). It documents endpoints for teams, roles, tasks, and workers, and supports optional git integration via a GitHub token and repo URL. The skill repeatedly instructs that API keys, tokens, and secrets must never be displayed in chat and must be read from and written to the credential file only.

It targets developers and power users who want to run autonomous or supervised multi-agent workflows on the Agentix platform. Two security-relevant characteristics stand out: credentials (Agentix API key, Anthropic key, GitHub token) are stored in a plaintext file under the home directory rather than an OS keychain, and the self-hosted path explicitly runs with no authentication on the local network. Combined with an autopilot mode that spawns workers and loops autonomously, these warrant a modest score reduction, though the skill uses official platform endpoints, encrypts keys server-side, and takes care never to leak secrets to chat output.