What problem does it solve? Benchmark runs in IC design flows produce FAILs that nobody systematically investigates, and fixes that never leave a local cache die with the session. This Skill closes the loop: it runs the canonical benchmark entry point, classifies every failure, fixes the layer that owns it, proves the fix, and captures it as a PR. ## Core Features & Use Cases - Failure classification: Sorts every FAIL into flow, gate, or tool categories, each with a distinct owner and repository, preventing fixes in the wrong layer. - Cache-based proof: Patches the installed plugin cache and validates fixes against real artefacts with bidirectional falsification and a control input before any repo PR. - Closed convergence loop: Re-runs after upgrade and session restart until the subject's target is met, whether that is flow_compliance_check exit 0, LEF/GDS deliverables, or the highest honest pass@1. - Use Case: After a /vibe-ic-all run on a benchmark IC reports failures, use this Skill to classify each FAIL, fix the owning layer in the vibe-ic plugin or vibeic-eda fork, prove the fix in the cache, and re-run until convergence. ## Quick Start Run the loop2converge skill to converge the latest benchmark run by classifying every FAIL and fixing the layer that owns it.
