dragoon is experimental research-grade tooling for design hygiene in ai-coded projects.
a heuristic scanner. it reads your source files with regex-level parsing, infers structural patterns, and outputs measurements and findings. those findings are useful guidance, not legal compliance, not gospel.
not a security tool. not a legal compliance audit. not an accessibility audit (it surfaces a few accessibility signals; it does not replace a real wcag audit). not a substitute for human design review.
- always validate findings against your actual design system.
- treat scores as relative, not absolute. compare your project to itself over time.
- the inference is heuristic. expect false positives. the rules are tuned to be useful at scale, not perfect on any single file.
- edit
dragoon.jsonrules by hand when the auto-derived defaults don't match your intent.
dragoon makes no network calls. it reads files locally and writes one json file locally. no telemetry. no analytics. no cloud.
none. see LICENSE for the full mit terms. the software is provided as is.
dragoon is built openly. issues, prs, and proposals welcome. the goal is design quality tooling that respects the realities of shipping software with ai. it's not perfect. it improves with use.