Gas expectations

Some Forge integration tests assert on exact gas usage and resource counts, for example:

• assert_gas - exact GasVector consumed by a test,
• assert_syscall / assert_builtin - number of syscalls / builtins used,
• the l2_gas: ~NNNNNN number inside the expected available_gas error strings.

These expected values are hardcoded in the test sources and change whenever the underlying gas model changes (typically after a Scarb or blockifier bump). Updating them by hand is tedious and error-prone.

The affected files are:

• crates/forge/tests/integration/gas.rs
• crates/forge/tests/integration/resources.rs
• crates/forge/tests/integration/available_gas.rs

The update-gas-expectations skill

We ship update-gas-expectations skill, that automates the refresh. It lives at .claude/skills/update-gas-expectations/SKILL.md.

The assertion helpers print both expected: and actual: on failure, so the actual value is always available in the panic message. The skill drives the relevant tests, reads the reported actual: value for each failing assertion, and rewrites the matching expectation in the source. It then re-runs until every assertion is green.

Usage

From an interactive Claude Code session in the repo root, invoke it explicitly:

/update-gas-expectations

or simply ask Claude to "update the gas expectations" / "fix the gas tests" - the skill's description makes it trigger automatically for those requests.

When to run it

Run it after any change that affects gas computation - most commonly after bumping Scarb or blockifier - and commit the regenerated expectations together with the bump.

Notes

• Only numeric expectation literals in the three files listed above are edited. Assertion logic, test bodies (the Cairo source inside test_case!) and production code are never touched.
• The skill runs the tests with exact assertions (without the non_exact_gas_assertions feature), so the recorded values are precise rather than margin-based.
• Tests that deliberately assert on wrong values (the assert_gas diagnostics tests, such as assert_gas_failure_shows_gas_diff_and_test_case_name) are excluded - the skill never "fixes" them.
• Any failure that is not an expected-vs-actual mismatch (compile error, contract deploy failure, missing scarb, a genuine logic panic) causes the skill to stop and report instead of editing numbers.