Fix CI
Input: optionally one target label to fix first.
The //:ci suite report you start from lists every red target with its
log. Work from the bottom of the dependency graph up: a red build in
packages/utils explains red tests in twenty packages, so fix builds, then
typechecks, then tests, then lints.
- For each red target, read its log and find the first real error (not the
cascade). Reproduce it with the package's own script
(
vitest run <file>, tsc --noEmit, biome check .) before editing.
- Fix the cause in the package that owns it. Do not skip tests, widen
types to
any, add // @ts-ignore outside test files, or loosen a
coverage threshold. A dependency version conflict is fixed in the
manifest and lockfile, not by patching node_modules.
- If a failure is a real regression in behavior, fix the behavior and keep
the test. If the test asserts something the code no longer promises
(and a changeset documents that), update the test and say so.
- Formatting and lint failures: run the package's
biome check --write --unsafe and review the diff before keeping it.
- Re-run the failed target, then the package's full
check, then move to
the next red target. Stop editing when //:ci is green.
Final message: each target that was red, the root cause in one sentence,
and the fix. Anything you could not fix, with the reason, so a human can
take it.
1---2name: fix-ci3description: Fix CI4---5# Fix CI67Input: optionally one target label to fix first.89The `//:ci` suite report you start from lists every red target with its10log. Work from the bottom of the dependency graph up: a red build in11`packages/utils` explains red tests in twenty packages, so fix builds, then12typechecks, then tests, then lints.13141. For each red target, read its log and find the first real error (not the15 cascade). Reproduce it with the package's own script16 (`vitest run <file>`, `tsc --noEmit`, `biome check .`) before editing.172. Fix the cause in the package that owns it. Do not skip tests, widen18 types to `any`, add `// @ts-ignore` outside test files, or loosen a19 coverage threshold. A dependency version conflict is fixed in the20 manifest and lockfile, not by patching `node_modules`.213. If a failure is a real regression in behavior, fix the behavior and keep22 the test. If the test asserts something the code no longer promises23 (and a changeset documents that), update the test and say so.244. Formatting and lint failures: run the package's `biome check --write25 --unsafe` and review the diff before keeping it.265. Re-run the failed target, then the package's full `check`, then move to27 the next red target. Stop editing when `//:ci` is green.2829Final message: each target that was red, the root cause in one sentence,30and the fix. Anything you could not fix, with the reason, so a human can31take it.
evmts