Changelog

All notable changes to the playwright-qa-agents plugin. Versions follow semantic versioning; before 1.0, a minor version can change agent behaviour.

0.10.0 - 2026-11-12

Changed

• ui-test-healer runs every playwright-cli command in its own named session (-s=heal-<name>), deletes any data it creates and reports what it couldn't, and keeps the repo's worker count.
• ui-test-generator confirms the plan was approved before generating. Both generators now stop and ask when they can't tell.
• Both generators follow Playwright's isolated-test practice: tests pass alone and in any order, a dependent journey is one test with a test.step per stage, and shared data is created per worker. Serial mode stays banned.
• page-mapper states that the test repo's CLAUDE.md wins over its defaults, like the other agents.

Added

• docs/worked-example.md: the cart feature of playwright-e2e-demo, from page map to the healer's final diff, with token use per stage. It links the reviewed cart plan, plans/ui-cart.md, pinned to the e2e commit that added it.
• README: how to check the plugin loaded (Claude Code no longer has /agents), the VS Code reload command, "What review caught", "Limitations" (convention vs enforced guardrails, measured cost, when not to use it) and "Background", which links the published blog post.
• README "Convention or enforced": serial mode and skipped or conditional tests are now lint errors in playwright-e2e-demo.
• README "What review caught": the orphaned test account was deleted afterwards with deleteAccount.
• CI fails when plugin.json and marketplace.json have different versions.

Fixed

• README described the healer as leaving plan-vs-page mismatches failing. It marks them test.fixme() with the expected and actual result.

[0.9.0] - 2026-11-12

Fixes from building the playwright-e2e-demo suite.

• page-mapper: named sessions, shared-site etiquette and cleanup, login and resume steps, repo locator rules, observed vs expected behaviour, template sections for input behaviour, success states and flows.
• api-test-planner: a Schemas section; undocumented negative cases go to open questions.
• ui-test-planner: the requirement wins over the page, with an "Expected findings" table; @smoke checked against project tags; every element checked for a locator; id ranges for parallel planners; no locator code in plans.
• api-test-generator: never serial, UPDATE_SCHEMAS=true for new schemas, the repo's status helper, unconditional cleanup, read helpers before relying on them.
• ui-test-generator: repo locator rules, named sessions, delete what exploration creates, don't raise workers, commons BasePage checks.
• README: terminal install, a new session after installing, claude plugin update, playwright-test-commons v0.7.0.

[0.8.1] - 2026-11-10

• Planners: whether plans are committed is up to the test repo.

[0.8.0] - 2026-11-10

• Agents read the test repo's per-folder CLAUDE.md guides; update-docs keeps CLAUDE.md split into a root file and folder guides.

[0.7.1] - 2026-11-10

• update-docs removes comments that only made sense to their author.

[0.7.0] - 2026-11-10

• Added the update-docs skill: detailed READMEs, JSDoc everywhere, only comments that explain why.

[0.6.1] - 2026-11-10

• Generators document reusable code with JSDoc.

[0.6.0] - 2026-11-10

• Response bodies are typed in the API clients; no environment settings.

[0.5.0] - 2026-11-10

• Bundled Playwright's playwright-cli skill in the plugin.

[0.4.1] - 2026-11-10

• Agents use the test repo's playwright-cli skill; removed the frontmatter check script from CI.

[0.4.0] - 2026-11-10

• Agents write TypeScript: .ts files, typed bodies, and lint that includes the type-check.

[0.3.0] - 2026-11-10

• Numeric test ids, unique across the repo (#0012); the kind of test comes from tags.

[0.2.1] - 2026-11-10

• Run tests with npx playwright test instead of npm test.

[0.2.0] - 2026-11-10

• Dropped briefs; plans open with acceptance criteria.

[0.1.1] - 2026-11-10

• Locators scope to a good container instead of requiring data-* anchors.

[0.1.0] - 2026-11-08

• First release: page-mapper, API and UI planners and generators, the UI healer, and the author-tests, run-spec, check and bug-report skills.