cdad check

cdad check is where the workflow starts. It asks one question about your repository: if an agent had to make a safe change here today, using only what the system makes available, how far would it get? The command answers that by combining two kinds of evidence. First a static scan of the repo itself, then a short questionnaire that walks through up to five capabilities and asks the same four legibility questions for each one. The result is a single score from zero to ten, plus a band that says roughly where the repo sits.

The score isn't the point. The report is. Every run writescdad-report.md, and that file is what cdad roadmapreads next. Skip the check and there's nothing for the roadmap step to prioritize against, so this is genuinely the command everything downstream depends on.

A handful of flags let you shape the run. --capabilities <n>changes how many capabilities get walked through the questionnaire, useful if five feels like too many or too few for a first pass.--skip-scan and --skip-questions let you run just one half of the assessment when you already trust the other half. --output <path> moves the report somewhere other than the repo root, and --format json swaps the Markdown report for a machine-readable one if you're feeding the result into other tooling instead of reading it yourself.

Run it again later and you get a second data point, not just a second score. Comparing two reports over a few months is the clearest way to see whether the extraction work is actually paying off.

cdad check
cdad check --capabilities 8 --format json