Claude Test plugin grows new tools, a richer pane and safer explorer instructions

What

Claude Test is a plugin (add-on package) bundled with Claude Code for running tests on your app. It has grown in several ways:

• New tools: claude_test_allow, claude_test_start_app, claude_test_confirm_run, claude_test_sign_in_as, claude_test_fix_offer, claude_test_more_tests, claude_test_remember and claude_test_browser_installed. Several appear in the lists of tools the claude-test agent may call, next to the existing claude_test_app_up and claude_test_show_run, and in at least one other agent's disallowedTools list. Whether they are switched on was not checked.
• Pane: a large block of new code reads the .claude-test/runs and .claude-test/specs folders and the files progress.ndjson, names.json, decisions.json, results.json, pane.json, signin.json and fix.json. It tracks whether the app started and whether the run was accepted, and builds rows for the Claude Test pane. Stop reasons such as "Couldn't start the test agents" and "Nothing ran: every test needs sign-in" become pane text. The pane also gets reportHref, installNoteText and isMadeHere, which check a made-here file with a MAC key and an install-progress.json file.
• Plugin permissions: the plugin now handles tool.call and lists many more calls it may make, including fs.read, fs.list, fs.stat, mcp.call, tool.check, clock.every, agent.list, session.root, ui.panes and ui.invalidate.
• Explorer agent: agents/explorer.md is renamed to agents/author.md. Its instructions now say it has only Read, Grep and Glob, list credential files it must never open, tell it not to grep with a catch-all glob, and tell it to stay inside the project folder.

Why

The plugin can do more, including starting your app and confirming a run, and its explorer is now told not to open files that hold credentials or private data or to read outside your project.