midnight-verify:zkir-regression

To detect verification toolchain regressions, run a curated suite of ZKIR claims through the full pipeline. This process executes classification, domain dispatch, compilation, and PLONK proof verification to output a claim-level pass/fail report with diagnostics.

You can run targeted regression testing by invoking the skill with a specific category argument like zk-proof. This executes a focused sweep across the selected domain instead of a full sweep, generating a summarized pass/fail report for those specific claims.

ZKIR claim categories supported for compiler upgrade verification include arithmetic, types, state, privacy, zk-proof, and transcript. Each category is processed through claim classification and domain skill loading to isolate behavioral changes.

Yes, you can use the generated pass/fail reports for CI gating and pre-release checks. The report provides claim-level verdicts and failure interpretations, making it suitable for automating quality control during suspected toolchain issues.

A behavioral change check fails when the verification pipeline detects discrepancies during claim compilation, execution, or PLONK proof verification. The output report includes failure interpretations to help diagnose the exact toolchain regression.

No external dependencies are required to run a full ZKIR regression sweep. The skill operates independently to orchestrate claim classification, contract generation, compilation, execution, and proof verification.