A FIELD GUIDE FOR AGENTIC SYSTEMS

Agents need context.

Agent Workspace helps agents understand and work across independent repositories, selecting shared context while keeping each change bounded and reviewable.

Start here →Browse patterns ↗
THE CORE PATH
01Independent reposseparate histories
↓
02Shared contextdeclared relationships
↓
03Bounded Runnarrow write scope
↓
04Durable resultCandidate + lineage
00 / MANIFESTO

The architecture is open.
The reference implementation is only one implementation.

These docs describe principles, trade-offs, and boundaries for self-managed agentic engineering. The goal is a shared language—not another CLI manual.

01 / ORIENTATION
01

Start with the why

Give agents enough cross-repository context without widening their write boundary.

→02

Learn the vocabulary

System, Layer, Resource, Task, Run, Candidate, and lineage.

→03

Use the patterns

Composable boundaries with explicit owners, limits, and extension points.

→
“
Context scope can be larger than mutation scope.

Agents may read across a system while every Run keeps its write boundary explicit and isolated.

See the pattern →
DOCUMENTATION STATUS58 pages in the first slice
DELIVERY MODELStatic HTML · PWA · GitHub Pages
SOURCE OF TRUTHMarkdown in the repository