Grade iterate
Phase 3 of building a Claude Managed Agent — the bounded grade→iterate loop. Define a CMA outcome (a required markdown rubric graded by an isolated grader), read each verdict, decide the next move (sharpen / re-run / promote to schedule), and once a version passes, run held-back eval cases in parallel.
What it does
Phase 3 of building a Claude Managed Agent — the bounded grade→iterate loop. Define a CMA outcome (a required markdown rubric graded by an isolated grader), read each verdict, decide the next move (sharpen / re-run / promote to schedule), and once a version passes, run held-back eval cases in parallel.
This is a reusable SKILL.md workflow from Alireza Rezvani. It gives an AI agent task-specific instructions to follow when you ask for this kind of work. The source page contains the full workflow, requirements, and any supporting files.
How to try it
- Read the original skill.
Check that the workflow fits your task, and review any tools, accounts, or permissions it needs.
- Follow the publisher’s setup guide.
Installation depends on your agent. Use the source repository’s instructions and include any required supporting files.
- Give your agent a concrete task.
Describe your goal and provide the relevant inputs. Try a small, reversible example and review the result.
The explanation is based on the publisher’s skill metadata. The source file was retrieved on 25 Sep 2026; we have not executed or independently validated the skill. Agent compatibility and any paid service requirements should be checked upstream.