Documentation
Written-control coverage, explained
CrewScore helps identify public guardrail controls that are not written in AI instructions. It does not test runtime behavior or issue safety grades.
Start locally
Paste instructions into the browser checker or use crewscore scan . from the CLI. Prompt text stays local to the tool you choose.
Protect the controls you mean
Run crewscore init . to create a prompt-free baseline and pull-request workflow. Require public controls such as human_gate.approval_required instead of treating a coverage percentage as a safety threshold.
Read the limitations
The open rules, reproducible validation snapshot, construct-validity limits, and scoring governance are part of the product contract.
Validation and limitations · Scoring governance · Public roadmap