comol/ai_rules_1c/1c-ui-regression

1c-ui-regression — reusable UI tests

Turn a relevant acceptance scenario into a saved, executable test with deterministic assertions. The model decides whether this is worth doing for the current task: recurring workflows, a reproduced UI regression or a requested automated suite are good candidates. Do not create a suite for every UI edit. Prefer extending the project's existing tests over introducing another framework.

Scope and execution

Authoring test files and running a browser are separate operations. The skill can be selected implicitly; selection neither changes UI_TESTING nor authorizes deployment, writes to an infobase or installation of a new service.

• Read content/rules/dev-standards-env.md for the target and effective UI policy. manual permits a run on an explicit UI-test request; essential and auto permit applicable checks with a configured publication; off prevents execution. Interactive checks in the thin client go through QA MCP (content/skills/1c-qa-testing/SKILL.md); this skill covers saved web-client suites. Do not enable UI tests as a side effect of selecting this skill.
• Use an authorized dev/test infobase with the current feature loaded. Resolve the main configuration / extension and source-to-target match through content/rules/extension-workspace.md. Missing runtime prerequisites leave execution unverified; useful local authoring may continue.
• Interactive discovery uses content/rules/ui-testing-tools.md and content/rules/web-client-driving.md. A saved suite uses its verified runner and the applicable TOOL_BROWSER / driver policies. Do not use it to bypass an off or unavailable required provider. A reusable Playwright runner does not require replacing a project's interactive driver.
• Read connection values from the project's .dev.env through an existing loader or explicit runtime environment mapping. Never copy URLs from another project or embed credentials / saved authentication state in a committed test.

Build the smallest useful suite

1. Specify the observation. Use the task's acceptance criteria or an existing OpenSpec scenario. Record initial state, user action and expected visible result before coding. Include the negative or boundary case that exposed the regression. No visual effect to assert means choose a business-logic check; do not change the product just to make a test possible.
2. Choose the runner. Inspect the existing package manifest, lockfile, test configuration and helpers. Reuse them. If none exists and a saved UI suite is useful, create a minimal Playwright Test setup in the target project, merging existing configuration instead of overwriting it. Check the installed version and its official documentation at <https://playwright.dev/docs/intro> before relying on APIs. Retain a lockfile and document the setup/run command. Do not auto-install services or browsers outside the authorized task.
3. Establish real controls. Use a current authorized browser observation and, when relevant, content/skills/1c-form-inspect/SKILL.md. Form metadata identifies logical elements; it does not prove their DOM selectors. Prefer observed accessible roles/names, scoped to the intended form/grid. Do not save ephemeral snapshot refs or generated form-instance IDs as reusable locators. If selectors cannot be verified, report that limit instead of emitting a runnable-looking stub or a passing empty test.
4. Write independent cases. Save tests in the project's established directory (otherwise tests/ui/). Wait for the relevant UI state, not a fixed sleep. Follow the verified input behavior from web-client-driving.md, then assert the value retained after leaving the field. Reuse existing login and data helpers. Each case owns its data or starts from a documented stable fixture; test order must not carry business state between cases.
5. Make failures machine-detectable. Use runner assertions and retain its nonzero exit code. Missing controls, setup failures, timeouts and zero executed cases cannot produce a pass. Emit the runner's JSON/JUnit report, or bounded JSONL observations alongside it, with case ID and expected/actual values. Capture permitted failure evidence without secrets. Logs complement assertions; an AI reading a log is not the test oracle.
6. Run when allowed. Execute the focused suite against the verified target. If test actions write/post/delete, establish the authorized test-data and cleanup scope first; cleanup can remove only artifacts owned by this run. Deployment, if needed and authorized, stays with /update1cbase or /deploy-and-test, never a hidden fixture hook. Preserve failure status even if cleanup also fails.

Completion

The deliverable names the test files, covered acceptance criteria, setup/run command, required target and actual result. Distinguish authored, executed: passed, executed: failed and blocked/unverified. A suite that only loads or lists its cases has not verified the feature. State the checked source / deployed state and report path for a live run.

Fix an in-scope defect through normal project gates. Re-run affected cases only after a relevant source, test or environment change; do not weaken an expected result to make the test green. Repeated deploy/fix cycles remain the explicitly requested /test-fix-loop workflow.

For reusable server-side checks use content/skills/1c-business-tests/SKILL.md; neither skill replaces applicable static validators or the existing focused Gate 3a check.