LintLang 0.6.0 lets teams turn on instruction-quality CI while reviewing an existing backlog. Record findings you have reviewed, commit that baseline, and gate remaining findings:
python -m pip install --upgrade lintlang==0.6.0
lintlang scan AGENTS.md --write-baseline .lintlang-baseline.json
# Review the findings before committing the baseline.
lintlang scan AGENTS.md --baseline .lintlang-baseline.json --fail-on reviewThe GitHub Action supports the same baseline through hermes-labs-ai/lintlang@v0.6.0 and its baseline input, including SARIF output. Complete baseline guide.
Also included:
- H2 handles explicit local verification bounds and negated retry prohibitions.
- The Claude Code plugin installs from the repository marketplace and resolves the scanner from a neutral working directory.
- Claude Code and OpenCode adapters now request LintLang 0.6.0.
- The merged MegaLinter plugin and browser-playground entry point are included.
Baselines are opt-in. Default scan behavior is unchanged; input errors still fail. A baseline acknowledges reviewed findings and does not establish runtime safety.