P01 — One Repo / One Agent
One bounded Run works against one repository and returns one explicit result.
Problem
An agent needs a clear unit of work and a clear handoff. An unbounded process with implicit ownership is difficult to validate or stop.
Context
Use this for a small coding task with one repository and no need for concurrent mutation. Start the interactive driver at the Git control-repository root; the selected repository can live under repos/ without becoming part of the control repository’s history.
Pattern
Create one Run, give it one objective and one mutation scope, then validate its Candidate before integration. ws run creates the isolated worktree for the managed repository; the driver remains in the workspace root.
Diagram
request → Run → isolated checkout → validate → Candidate → integrate
Example
profile: interactive-coder
target: api-service
path: repos/api-service
objective: add request ID middleware
result: commit + test evidence
Trade-offs
This is easy to reason about and cheap to operate. It does not solve cross-repository context, concurrency, runtime collisions, or scheduled maintenance.
Failure modes
The Run grows beyond its stated objective, edits the canonical checkout, or reports success without validation evidence.
Extension points
Swap the Agent Provider, execution provider, or integration policy without changing the Run boundary.
When not to use
Do not add bounded coordination for a single, short, human-supervised command unless the workspace root and its generated guidance are already useful to the team.