[codex] add Playwright healer skill

[shubh24]

What changed

• Adds a minimal playwright-healer skill for repairing existing TypeScript and Python Playwright workflows.
• Composes with the repository's existing browser-trace skill instead of adding a coding-agent runtime, Browserbase Agent, MCP server, or helper script.
• Requires the host coding agent to run and repair the real Playwright entry point, retain context across successive failures, and verify both the normal command and a fresh traced Browserbase execution.
• Adds three eval prompts covering multi-stage repair, local-versus-remote behavior, and unsafe replay handling.
• Registers the skill in the README and Claude plugin marketplace; installing the marketplace plugin includes browser-trace automatically.

Why

Customers already use Claude Code, Codex, or another coding host that can inspect files, edit code, run commands, and iterate. The skill should provide the recovery contract and Browserbase observability guidance without embedding a second agent or orchestration layer.

Customer impact

Customers can install one small skill package and point it at an existing Playwright repository. Their code remains ordinary Playwright, trace artifacts come from the same remote session the script drives, and completion requires the actual business assertions to pass.

Validation

• node scripts/validate-skills.mjs --skill playwright-healer — passed with zero warnings.
• node scripts/validate-skills.mjs — all 17 skills passed with zero warnings.
• Marketplace and eval JSON parse successfully.
• The workflow was exercised against both a two-stage locator/redirect fixture and a multi-page checkout fixture with successive planted failures.