Hard block on typing passwords breaks legitimate dev/test workflows - needs a permission-gated opt-in for own dev environments

The problem

Claude Code categorically refuses to type passwords into login forms — even documented test-account credentials, on the developer's own application, on localhost / their own test server, with the owner explicitly instructing it to log in.

I understand the default. But as a blanket, non-configurable rule it actively hurts the core dev workflow the tool is otherwise great at:

• My app (ASP.NET WebForms, my own codebase) keeps login state in the in-memory session. Every rebuild restarts the dev server → session gone → re-login needed. That's dozens of logins per day during UI work.
• The agent drives the browser (BrowserClaw / Playwright MCP) for visual testing. The test account (ClaudeTemp) and its password are documented in the repo's testing guide; earlier Claude Code versions filled the login form as part of the documented workflow (browser_fill_form with the credentials was literally written in our project docs by/with Claude). A later version silently stopped doing that mid-workflow.
• "Log in once manually, the profile keeps the session" does not work when the session is server-side and dies on every app restart. Claude's suggested workaround was exactly that, and it collapsed on first contact with reality.

We ended up building a passwordless dev-auto-login endpoint into the app (gated to non-production) — a fine solution, but I had to modify my product's auth code to work around my coding agent's policy. Not every project can do that, and it should not be required.

What I'm asking for

An explicit, user-controlled opt-in for credential entry in development contexts, e.g.:

• a permission rule (settings.json) scoping credential entry to named hosts (localhost, own test domains), and/or
• treating credentials that the user has committed into their own repo docs as usable for the documented workflow, after a one-time confirmation.

Keep the default as-is. Just give the machine's owner a way to override it for their own dev/test systems, the same way other guarded actions are permission-gated rather than hard-refused.

Why this matters

This is a daily-friction issue in exactly the loop (edit → build → verify in browser) where agentic coding lives. Feedback from the session, in the user's words: this is seriously annoying in a development environment, to the point of considering a switch to Codex — which handles the same workflow without the hard block.

Environment: Claude Code (desktop, Windows 11), Opus 4.8.