V5 additions

[lorenss-m]

Note

Improves authentication and eval lifecycle, plus docs and version updates.

• Auth is now injected per-request via httpx/aiohttp hooks, preferring the API key set in hud.eval(api_key=...) and falling back to settings; overrides empty/invalid headers
• Removes static Authorization header from connect_hub MCP config; only sets Environment-Name (auth comes from hooks)
• Tweaks EvalContext.__aenter__ to register trace/enter before environment connect and centralizes cleanup by delegating failures to __aexit__
• Adds docs note for local agent execution dependencies in docs/reference/cli/eval.mdx
• Bumps version and tests from 0.5.0 to 0.5.1 (docs, __version__, pyproject)

^{Written by Cursor Bugbot for commit afb3ffa. This will update automatically on new commits. Configure here.}

[cursor]

Hub auth fails when Environment imported directly

The Authorization header was removed from connect_hub() and now relies on httpx/aiohttp hooks in hud.eval.instrument to inject auth at request time. However, this module is only auto-imported when using import hud. Users who follow the documented pattern in hud/environment/__init__.py (from hud.environment import Environment) won't trigger the instrumentation, causing hub connections to fail with authentication errors. The old code included the header directly in mcp_config, ensuring auth worked regardless of import path.