An open-source skill for your coding agent
Help people accomplish more.
Make the result better, too. UXcalibur helps your coding agent find a few justified interface changes from evidence in your app—within the focus areas and exclusions you choose.
Codex · Claude Code · Cursor Versioned method · MIT
01
Set the boundary.
Name what a person needs to accomplish or improve. Choose the flow, focus areas, exclusions, and observable criteria.
02
Inspect real behavior.
Follow the relevant interface and source. Separate what was observed from what is inferred. Preserve what already works.
03
Make a useful cut.
Get a ranked recommendation with its behavior contract and acceptance check. Implement when authorized, then verify.
A short list. A clear reason for every cut.
Each recommendation connects an observed obstacle or opportunity to the chosen outcome, the proposed behavior, supporting implementation, preservation rules, confidence, and an observable check.
For broader, explicitly requested work, the same method produces a detailed upgrade specification with complete flows, implementation packets, dependencies, and a living tracker. A focused pass stays focused; there is no required finding count.
The same four continuity cases fail at the expected assertion on the preserved baseline and pass after implementation. This demonstrates actionability on a synthetic flow; it does not measure customer outcomes or general method accuracy. View the full proof.
Bring a task worth improving.
The installed skill and reference files are the product. Your agent supplies its model, tools, permissions, and processing environment.