Eve's filesystem-first agent model: an agent is a directory of files

What this is

Eve's central design claim is that an agent is a directory of files. The project layout defines the parts of an agent as files and directories rather than as configuration buried in application code:

• instructions.md -- the system prompt.
• agent.ts -- model and runtime configuration.
• tools/ -- typed functions the model can call.
• skills/ -- procedures loaded contextually.
• channels/ -- Slack, Discord, and HTTP integrations.
• schedules/ -- cron schedules.
• subagents/ -- specialist agents for delegation.
• connections/ -- tools from external MCP servers.
• sandbox/ -- a controlled workspace for files and commands.
• hooks/ -- lifecycle and stream-event handlers.

Why it matters

A file-backed definition is reviewable and version-controllable: the system prompt, the tool surface, the skills, the channels, and the approval-relevant sandbox are all artifacts in a repository, not state assembled at runtime. That makes an agent's full operating context inspectable from its directory, and it makes changes to that context show up as diffs.

Operator consequence

• An operator can read what an agent is, what it may call, and how it is wired by reading its directory, and can gate changes through normal code review.
• The model is a research lens for the thesis that the shaped environment around the model is itself a first-class, ownable layer.

Open question

• How much agent behavior is fully captured by the file layout versus resolved at runtime (model defaults, connection state, sandbox provisioning)? Worth confirming against a working project as the API settles.