Agent Quality Record

Write the release decision down

super agent evaluate runs an agent's BDD specification and tells you what passed. It leaves nothing behind that a colleague or an auditor can read months later.

One flag writes that down, in the open format published as SuperGauge.

super agent evaluate developer --gauge-out record.yaml

What the record holds

Scenarios as the task set

BDD scenarios in the playbook are what the agent was measured against. A scenario marked held-out counts toward the held-out portion, and the manifest is fingerprinted by name and expected output, so editing the wording of a scenario leaves the seal intact.

What the agent was permitted to do

The authority block is read from the compiled playbook. Tools become capabilities, and the runtime block supplies the sandbox and egress posture. A record naming the agent without naming its permissions can be structurally valid and substantively wrong.

Routing invocation

An agent published over A2A is selected by another agent reading its card. Where selection fails, the agent receives no work, and its completion rate describes traffic that never arrived. No other measure in the registry covers discoverability.

A record

record.yaml
supergauge: "0.1"
profile: {id: sg/framework-agent, version: "0.1", tier: T1}

subject:
  agent: developer
  harness_digest: sha256:9c1e...        # the playbook, fingerprinted
  model: {provider: ollama, id: qwen3:8b}
  authority:                            # what the agent could do
    sandbox: docker
    egress: deny-by-default
    capabilities: [search, fetch]

task_set:
  manifest_digest: sha256:7d02...       # the scenarios, fingerprinted
  held_in: 6
  held_out: 2
  sealed: true

measures:
  - {id: task.completion,             value: 0.83, n: 8}
  - {id: interop.routing_invocation,  value: 0.75}

decision:
  verdict: hold
  actor: "you@example.com"
--gauge-out PATH

Write the record. A .json suffix selects JSON, otherwise YAML.

--gauge-tier T0 T1 T2

The risk tier the record claims. Higher tiers demand more before a release.

--gauge-sealed

Assert the held-out scenarios were closed to anything that tunes the agent.

Measures appear only where the run produced them, so an evaluation without repeated attempts carries no reliability measure and reaches a lower conformance level. Record emission never fails an evaluation.

An open format

Any SuperGauge implementation reads the output, so a team running SuperOptiX alongside a coding-agent harness gets one record shape across both.