Run one scenario, all selected scenarios, or a configured tool/model matrix.

Real agent execution is disabled unless AX_EVAL_ENABLED=1 is set, because real adapters may spend LLM API credits and execute agent-driven CLI commands. Use --dry-run without that environment variable to validate scenario loading, fixture setup, cache keys, and run planning without invoking an LLM agent.

Rubric paths in `evaluation.judge.rubric` resolve in this order: relative to the scenario YAML, then the workspace (after fixture copy), then fixtures_path.

Artifacts are written under ax-eval-results/ by default, including reports, transcripts, metrics, and the isolated fixture workspace.

Usage: ax-eval run [OPTIONS]

Options:
  -s, --scenario <SCENARIO>
          Path to scenario file or name

      --all
          Run all scenarios in fixtures directory

      --tags <TAGS>
          Filter scenarios by tags

      --tier <TIER>
          Filter scenarios by tier (0=smoke, 1=quick, 2=standard, 3=comprehensive)
          
          [default: 0]

      --tool <TOOL>
          Tool to test (e.g., claude-code, opencode)

      --model <MODEL>
          Model to use with the tool (e.g., claude-sonnet-4-20250514, gpt-4o)

      --profile <PROFILE>
          Profile to use for matrix run (defined in config)

      --dry-run
          Dry run (don't execute LLM calls)

      --no-cache
          Disable caching

  -v, --verbose
          Enable verbose output (or set RUST_LOG for fine-grained control)

      --judge-model <JUDGE_MODEL>
          Judge model for LLM-as-judge evaluation

      --judge-tool <JUDGE_TOOL>
          Tool to use for LLM-as-judge evaluation (defaults to judge config or opencode)

      --no-judge
          Disable LLM-as-judge evaluation

      --timeout-secs <TIMEOUT_SECS>
          Maximum execution time in seconds per command
          
          [default: 300]

  -h, --help
          Print help (see a summary with '-h')

Examples:
  AX_EVAL_ENABLED=1 ax-eval run --scenario fixtures/my_scenario.yaml --tool opencode
  PATH="$PWD/target/debug:$PATH" AX_EVAL_ENABLED=1 ax-eval run --scenario fixtures/my_scenario.yaml --tool opencode
  AX_EVAL_ENABLED=1 ax-eval run --all --tags smoke --tier 1 --tool claude-code
  AX_EVAL_ENABLED=1 ax-eval run --scenario fixtures/my_scenario.yaml --profile quick
  ax-eval run --scenario fixtures/my_scenario.yaml --dry-run

Start with `ax-eval template scenario` for a copyable scenario schema.
