Testing

npm run test:unit         # mocha over out/test: unit, conformance, commands, CLI, MCP, Plugin API, text, golden, layout, interop
npm run test:integration  # the extension inside a VS Code test instance (vscode-test)
npm run check:webview     # the webview page contract (no framework, no stylesheet, module rules)
npm test                  # test:unit and check:webview
npm run test:cli          # only the CLI and MCP tests

Run npm run compile first: the tests run from out/.

Golden images#

test/golden renders fixture designs and compares them with the expected images. After an intended rendering change, npm run golden:update rewrites the expected images; review them before committing.

Agent scenarios#

test/agent-scenarios holds end-to-end scenarios in which Claude Code designs through ud mcp with the shipped skills: a login screen, variables with modes, button variants, a frame to React, a review, quick actions and the AI panel’s own pipeline. Each scenario sets its prompt, fixture, models, allowed tools and checks (selectors, lint, renders, image similarity, code).

npm run test:agents -- --scenario S1 --runs 1 --claude test/agent-scenarios/mock-claude/claude.mjs   # smoke run, no model
npm run test:agents -- --scenario S1                                                              # real run, needs Claude Code
npm run test:agents:replay                                                                        # replay recorded transcripts

Benchmarks#

npm run bench measures the performance budgets of ARC-001 §12 on generated fixtures (npm run bench:gen).

CI#

.github/workflows/test.yml runs on Linux, macOS and Windows: npm ci, compile, lint, the webview check, the licence check, the schema check, the unit tests and the integration tests, plus the CLI tests on Node 22 and a .vsix size check.