Improve the method with evidence.

Open source / MIT / Catalyst Forge

The most useful contribution connects a specific job, observed behavior, proposed change, and a check that can establish the result.

UXcalibur on GitHub is the method source. The public npm package, uxcalibur, installs the skill and references for supported coding-agent hosts. UXcalibur is licensed under MIT.

Report a useful issue

Use the issue tracker. Include the method/package version, coding-agent host, chosen outcome, focus areas and exclusions, reproduction, expected and observed behavior, and relevant evidence. State what remains unverified.

For an installer problem, include the command, operating system, Node version, selected agent and destination, and the error. Remove credentials, private source, customer data, and unrelated personal paths before sharing evidence.

For a method problem, a minimal synthetic example is particularly useful. Explain whether the issue concerns an unsupported inference, missing behavior contract, ignored boundary, circular packet dependency, or misleading verification claim.

Work from the versioned source

The repository's skills/uxcalibur/ directory is the method source. Personal installed copies are separate installations. Changes to the method should retain the skill's relative reference links and agent metadata. Keep the focused pass as the default and detailed specs explicit.

Use the repository's setup and verification instructions before opening a pull request. The synthetic proof demonstrates the existing checks; the fixture is an app being reviewed, not a product analyzer. A release install does not require the fixture's development harness.

Give a reviewer something concrete

Describe the problem and resulting behavior. Connect changes to evidence, record meaningful checks and limits, and preserve accepted decisions, user data, useful architecture, and unrelated work. Method changes should include a bounded before/after example where it helps establish the improvement.

Do not turn a prediction into a measured efficacy claim. Keep synthetic examples labeled, observed and proposed behavior separate, and unimplemented recommendations visibly unimplemented. A report with no actionable cuts is valid.