Agent#

In scion, an Agent is an isolated process running an LLM + harness loop against a task. It is the fundamental unit of execution.

Properties#

Each agent has:

  • Identity: Unique name, container ID, and (in hosted mode) a UUID and URL-safe slug
  • Home directory: Mounted at /home/<user> — contains harness config, credentials, agent-info.json
  • Workspace: Mounted at /workspace — a dedicated git worktree or cloned repo
  • Template: The template blueprint that seeded its configuration
  • Harness: The LLM-specific adapter (Claude, Gemini, OpenCode, Codex)

State Model#

See agent-state-model for the full three-dimensional state tracking system.

Isolation#

Agents are strictly isolated from each other:

  • Dedicated filesystem and home directory
  • Shadow mounts (tmpfs) prevent access to .scion data or other agents
  • Environment variables explicitly projected
  • Credentials mounted read-only or injected per-agent

Sub-agents#

Scion agents are not limited by a harness’s built-in sub-agent feature. Through templates (home folder content, env vars, custom mounts), they have full agent capabilities even when acting as sub-agents from the orchestrator’s perspective.

See Also#

Backward Agent State Model Context Management Forward