sturdy finch
One thing that’s driving me mildly insane: Playwright MCP security prompts are so granular that local app testing stops feeling agentic.
I’m running Codex against my own localhost app, already set approval_policy = "never", and it still keeps asking me to approve every tiny Playwright action — navigate, run code, click, wait, repeat. So instead of “Codex inspect milestone 4 and verify behavior,” it becomes me sitting there approving a chain of near-identical prompts.
I completely get why the security layer exists, but for localhost especially I really want a smarter trust boundary: “approve Playwright for this localhost session” instead of approving every micro-step. **Or something similar. ** The current flow makes the workflow feel much more manual than it needs to.