fleet eval <list|run> [<agent-name>] [flags]Runs and lists 6-dimension agent evaluations: task output, reliability, quality, efficiency, collaboration, and cost. fleet eval run <agent> computes and stores a fresh score for one agent; fleet eval list shows stored evaluations.
eval requires a subcommand — there is no bare positional form. The brain daemon uses the same scoring model continuously; eval is the on-demand entry point. A low eval score does not quarantine an agent; that is a separate brain risk-model behavior.
| --agent <name> | With list, filter stored evaluations to one agent |
| --limit <N> | With list, maximum results to return (default 20) |
| --json | With list, emit the evaluations as JSON (empty result is `[]`) |
Run an evaluation pass on one agent
fleet eval run frontend-devList stored evaluations for an agent
fleet eval list --agent tech-lead --limit 10List all stored evaluations as JSON
fleet eval list --jsonSix dimensions: task output, reliability, quality, efficiency, collaboration, and cost. Risk is a separate logistic-regression model — eval does not produce a risk score.
No. eval only reports scores. Automated quarantine is a brain-daemon behavior driven by the separate risk model, not by eval.
One binary. Five minutes. See every agent, coordinate every handoff, and keep a full audit trail of what your fleet did.