Write symbolic Foundry tests that verify Solidity properties across all inputs within bounds.
npx skills add https://github.com/andresdefi/cryptoskills --skill halmos-andresdefi
Please help me install this Agent Skill. Skill: halmos Source: https://github.com/andresdefi/cryptoskills/tree/main/skills/halmos Command: npx skills add https://github.com/andresdefi/cryptoskills --skill halmos-andresdefi
check_ prefixed Foundry tests whose parameters become symbolic variables covering every possible value within configurable loop and depth bounds. - Counter-example generation: When a property fails, Halmos outputs concrete hex-encoded witness values that can be reproduced as standard Foundry tests. - Bounded model checking with CI integration: Configure --loop, --depth, and solver timeouts to balance verification confidence against runtime in GitHub Actions pipelines. - Use Case: Verify that an ERC-20 transfer never changes totalSupply, or that a vault's totalAssets always covers totalShares, for all possible deposit and withdrawal amounts rather than a few hundred random samples. ## Quick Start Write a Foundry test function prefixed with check_ that uses vm.assume and assert, then ask the agent to run halmos against it and interpret any counter-examples.High-intent search queries and answers about installing and using this skill.
Write a Foundry test function prefixed with check_ instead of test_, and its parameters become symbolic variables covering all possible values. Use vm.assume to constrain inputs and raw assert() for properties, then run the halmos CLI after forge build.
Halmos is a bounded symbolic executor that checks all inputs within loop bounds using local z3 solving. Echidna is a coverage-guided fuzzer for stateful exploration, while Certora provides unbounded formal verification via its CVL language and paid cloud service.
No, Halmos supports a subset including vm.assume, vm.prank, vm.deal, vm.store, vm.load, vm.warp, and vm.roll. Forking cheatcodes like vm.createSelectFork, vm.ffi, file I/O, and snapshots are not supported during symbolic execution.
Over-constraining with vm.assume can make assumptions contradictory, leaving zero execution paths so the test passes vacuously. Check the path count in the output, and use raw assert() instead of assertEq, which reverts and is treated as a valid path.
Increase the limit with --solver-timeout-assertion, tighten input ranges with vm.assume, and reduce the --loop bound. Breaking complex assertions into smaller sub-properties and avoiding non-linear arithmetic with multiple symbolic variables also helps z3.
Halmos only verifies properties up to the configured --loop iteration bound, so bugs appearing at higher iterations are missed. It also lacks fork mode, has limited cross-contract symbolic execution, and is pre-1.0 software with breaking changes between releases.
urufu-labsApplies testing conventions for OpenZeppelin contracts across Hardhat, Foundry, Halmos, and Certora.
ccashwellVerify Solidity contracts with Certora CVL specs and Halmos symbolic testing.
hairyfFuzz Ethereum smart contracts to falsify invariants and trigger assertion failures.
0xinitVerify smart contract correctness with Certora Prover and CVL specifications.