Tools for verifying AI-generated code. The goal is to know that AI-generated code is correct without trusting an AI to say so. docs/VISION.md states the vision.
crosscheck/is the Crosscheck Claude Code plugin. It verifies code with Dafny and Lean, runs differential random testing against Lean models, and carries the specification and assurance skills.cgv/is the contract graph verifier. It extracts contracts from Python code (Django models and plain-Python data classes) and checks them across component boundaries. Its checker has machine-checked soundness proofs in Lean.
The plugin is published through the nicholls marketplace:
claude plugin marketplace add nicholls-inc/claude-code-marketplace
claude plugin install crosscheck@nichollsSee crosscheck/README.md for prerequisites. See cgv/README.md to build and run CGV.
docs/VISION.mdis the vision the whole suite shares.docs/assurance/ROADMAP.mdlists the steps towards the vision, anddocs/TASKS.mdis the ordered queue of tasks.docs/assurance/,docs/gates/,docs/decisions/,intent/,.assurance/,evals/,REVIEW.md,JOURNAL.md, andplan.mdhold the development framework and its records.formal-verification/holds the specs and fixtures that the Lean and Dafny pipelines are tested against.logic-distribution/holds the research behindcrosscheck/docs/research/logic-distribution-analysis.md.