AR Policy Debugger

Overview

This is the iterative refinement skill, the equivalent of AWS's documented "Kiro CLI" policy-repair workflow. It loads the policy definition, quality report, and failing-test findings; explains why a test fails using the two-step model; and applies targeted annotations so you don't hand-edit formal logic.

⚠️ Golden rule: check the translation first. The SMT validation step is mathematically sound. When a result is wrong, look at the finding's premises/claims/confidence:

• Right variables, right values, wrong result → the issue is in your rules.
• Wrong/missing variable assignment → the issue is in your variable descriptions (cheaper, safer to fix).

Reference: references/debugging-decisions.md (the full decision table + annotation recipes), ../../shared/references/findings-reference.md, smtlib-rules.md, ar-api-context.md.