Capability Model
How commands, skills, tools, and permissions are surfaced to human and agent execution.
Purpose
This document defines the capability layer of the toProject system.
Block Position
Inside the toProject board, this page belongs to the Operating System block.
Core Position
Sandbox answers what is isolated.
Capability answers what is allowed.
These are related but distinct.
Capability Categories
The system should distinguish at least:
- commands
- skills
- tools
- workflows
- hooks
- privileged actions
Capability Questions
Each capability should be answerable through:
- who can invoke it
- in which runtime it is valid
- whether network is required
- whether write permission is required
- whether approval is required
Why This Matters
Without a capability model:
- sandbox policy becomes too vague
- execution mode becomes hard to enforce
- agents appear more autonomous than they actually are
Boundary
This page defines executable capability classes.
It does not define:
- project object semantics
- run state transitions
- artifact storage
Those belong elsewhere in the architecture.