Claude Agent Orchestration
A documented Claude Code development harness with distinct planning, building, evaluation and documentation roles.
- Agent Orchestration
Under the hood
Technical architecture
The public repository currently contains a development-rules document. It specifies a workflow for Claude Code; it is not a deployed orchestration service or a separately implemented agent runtime.
Orchestration
Claude Code instructions
Defines phase boundaries, task clarification and structured handoffs.
Specification
BRIEF.md · SPEC.md
Describes the task context and the proposed implementation contract in files.
Sprint loop
Builder and evaluator roles
Specifies separate implementation and evaluation phases, with failure context returned for correction.
Handoff
Documentation and cleanup phase
Defines a final documentation stage after the sprint checks pass.
Failure and fallback path
The document specifies blocked states for missing inputs and revision loops after failed evaluations. Following those instructions depends on the host agent; this portfolio does not claim runtime enforcement.
One request, end to end
- Clarify the task and prepare a brief and specification.
- Run the documented builder/evaluator loop against a sprint contract.
- Document and hand off the result after the required checks.
Why it is built this way
- Use file-based contracts for handoffs.
- Separate implementation from evaluation responsibilities.
- Expose the actual repository artifact rather than imply an executable product exists.