Changelog

v0.1.3 — 2026-09-26

Fullscreen browser execution improvement for desktop UI testing.

Changed

• Prefer Google Chrome in headed, isolated Fullscreen sessions.
• Use maximized mode as a fallback when Fullscreen cannot be established.
• Preserve isolated state per independent case and one session across that case's steps.

Validation

• Chrome 154, headed Fullscreen, viewport tracking, isolated restart behavior, localhost and remote navigation, URL-visible evidence, redaction, and fallback behavior validated with Playwright MCP 0.0.82 on macOS.

• Sanitized authenticated-application dogfooding: PASS for Fullscreen, form interaction, authentication flow, post-login navigation, and URL-visible evidence. Private targets and artifacts are not included.

• Post-release fresh-clone installation and public example.com browser smoke: PASS; the installed Skill was discovered, Chrome opened headed in Fullscreen, and URL-visible evidence was verified.

Notes

• Computer Use is not required for normal Fullscreen browser execution.
• Fullscreen runtime validation covers macOS only. Other operating systems and multi-monitor placement remain unvalidated; headless runs do not use desktop Fullscreen.

v0.1.2 — 2026-09-26

Browser-execution and test-isolation improvements for the existing Skill.

Changed

• Prefer Google Chrome and headed browser execution for UI tests.
• Start each independent case in a fresh isolated session; retain one session across that case's steps.
• Require explicit intent to reuse supplied authentication or browser state, and report that deviation.
• Use best-effort maximization while continuing in a headed window if it is unavailable.

Documentation

• Clarified that Playwright isolated state serves the private-session intent without claiming Chrome's native Incognito UI.
• Clarified that Computer Use is not required for normal browser testing.
• Documented that true OS fullscreen is not guaranteed and maximization is best-effort.

Validation

• Google Chrome 154, headed execution, isolated sequential cases, explicit synthetic storage-state reuse, best-effort maximization, localhost interaction, remote navigation, SPA URL update, and URL-visible evidence validated with Playwright MCP 0.0.82 on macOS.
• Other operating systems and real authenticated profile reuse were not runtime validated.

v0.1.1 — 2026-09-26

This privacy and release correction supersedes the withdrawn v0.1.0; it contains the same initial Skill functionality.

Security / Privacy

• Re-published the initial release from sanitized Git metadata.
• Added permanent Git metadata privacy validation to the release checks.

v0.1.0 — 2026-09-26

• Prompt-driven black-box UI testing for remote sites and localhost applications, using Playwright MCP as a separately configured browser capability.
• Natural-language planning and execution for forms, common controls, file-input uploads, redirects, and SPA navigation, with Expected vs Actual validation and PASS, FAIL, BLOCKED, or INCONCLUSIVE results.
• URL-visible screenshot evidence composed from browser-observed URLs, with conservative URL redaction, retained raw captures, and isolated per-test workspaces. Concurrent runs were validated when each uses separate browser sessions and evidence paths.
• Project and User/Global Skill installation guidance, Plugin packaging, and Plugin runtime validation in Codex CLI on macOS, plus safety policies, report assets, and generic public-safe examples.
• Known limits include unvalidated custom controls and interaction variants, real SSO/MFA handoff and profile reuse, some dynamic UI/recovery behavior, and runtime validation beyond Codex CLI on macOS. See the README compatibility and Known Limitations sections.