maida scenario run

Runs pinned headless Claude Code prompts in isolated fixture workspaces and evaluates their captures against checked-in baselines.

maida scenario run [MANIFEST] [--scenario ID] [--format text|json|markdown]

MANIFEST defaults to .maida/scenarios.yaml. The runner validates the whole manifest and local environment before invoking Claude. Each selected scenario gets a fresh temporary workspace containing only its declared Git-tracked fixture files, a unique non-persisted Claude session, an ephemeral loopback OTLP receiver, its own native USD cap, turn cap, and process-group timeout. The exact project settings and strict MCP configuration are passed explicitly; permissions use dontAsk and dangerous permission bypass is never enabled.

Reports contain no raw Claude stdout or stderr. Per-scenario status is pass, assertion_failed, or agent_failed.

Exit codes: 0 all pass; 1 one or more assertion failures; 2 invalid manifest, selection, config, executable, or version preflight; 10 any agent, capture-import, or runtime failure. Agent failure takes precedence over assertion failure when multiple scenarios run.

See Run isolated scenarios for the manifest schema and a complete example.