Run Godlint

Version updated for https://github.com/tomerwave/godlint to version v0.5.0.

  • This action is used across all versions by 0 repositories.

Action Type

This is a Composite action.

Go to the GitHub Marketplace to find the latest changes.

Action Summary

Godlint is an executable engineering constitution that automates the enforcement of rules across Rust, TypeScript, JavaScript, and Python repositories. It provides guardrails for AI coding agents to catch architectural drift before generated code reaches review, share boundaries and thresholds across a polyglot repository, enforce deterministic policy locally and in CI, account for accountable exceptions with expiring suppressions, and ensure local compliance without dependencies on external LLMs.

What’s Changed

Added

  • ci/stale-action-refs — makes full commit pins reviewable without network access. It reports a pin without an inline version label at warning, and reports repository-proven contradictions at the configured severity when the same action and SHA carry different labels or the same action and label name different SHAs. A single leading v is normalised before comparing, because v4.6.2 and 4.6.2 name the same release and reporting them as a contradiction would spend the rule’s only asset — that it speaks when a label lies. allow-in removes paths from reporting and comparison. The rule deliberately cannot verify that a label names the pinned commit; zizmor’s online stale-action-refs and ref-version-mismatch audits cover that external check.
  • ci/no-silenced-failure — reports checks that cannot make a workflow fail: literal continue-on-error: true settings and scripts ending || true, ; exit 0, or || exit 0. A same-job read of steps.<id>.outcome or .conclusion proves a deliberate soft step and stays silent. Corpus-common continue-on-error and || true findings are capped at warning; explicit exit-zero endings stay at the configured severity.

Changed

  • ci/no-monolithic-job raises the recommended@1 step limit from 7 to 20, the p90 of 231 jobs across 94 workflows in nine widely used repositories. The former limit came from this repository alone and reported 36% of real jobs. Raising it also made this repository’s own release.yml exemption unnecessary, which is what confirmed the number was wrong rather than the workflow.
  • ci/no-inline-script keeps its limit of 8 after the same measurement — across 981 real scripts it is p85, and 15% exceed it — so the number is now corpus-backed rather than repository-derived.
  • A condition written without braces is read as an expression by every rule that reads conditions. GitHub treats an if: as an expression whether or not it is wrapped in ${{ … }}, so ci/bot-conditions was missing the idiomatic spelling entirely and ci/no-silenced-failure’s escape hatch did not open for it. Both now share one reader.

Fixed

  • style/no-comments no longer reports a returned string literal as a docstring. The check confirmed the string was the first thing in a block but never that the string was the statement, so a Python function whose first statement returned a literal — return "active" — was reported as a comment at error under recommended@1. A real docstring, including a module docstring, still reports. Found by writing deliberately bad Python to probe for rules Godlint is missing.
  • A declared drift in .github/accepted-drift.md survives the merge that lands it, so a pull request that deliberately relaxes a rule no longer has to be re-declared on every subsequent branch.
  • validate-pull-request.py runs its change-scoped checks whether or not --release-line is passed, and fails when it cannot find a release line to compare against. Only CI passed the flag, so a local run reported one fewer check than CI ran and printed that all of them passed — including, on one branch, the check that then failed the pull request.