Define --full as the composition of --all, --max-steps 100, and --auto-respond. Also fix a few reward parsing bugs.

[pimpale]

Note

CLI and config updates

• Replace full with all in config (hud/cli/eval.py) and add --all flag; --full now sets --all, --auto-respond, and --max-steps 100.
• auto_respond is now an explicit boolean flag (no implicit default via --full); max_steps defaults to 10.
• Task selection uses all for multi-task runs; display updated to show all and explicit max_steps.

Reward handling changes

• Reward is computed in EvalContext.__aexit__ from evaluate tools; runners stop manually setting ctx.reward (hud/datasets/runner.py).
• Telemetry exit uses self.reward directly (hud/eval/context.py).

Tool result structure

• Environment returns MCPToolResult with structuredContent for both local and remote tool calls (hud/environment/environment.py).
• Improve reward parse logging to print result.structuredContent (hud/agents/base.py).

^{Written by Cursor Bugbot for commit 2b60eb4. This will update automatically on new commits. Configure here.}