Scores and verdicts
Every evaluated event carries a composite score between 0 and 1 and a verdict from the agent’s layers. The console reuses a few fixed bands everywhere.
An event also counts as a violation when any layer reported
passed: false, whatever the verdict. See Scoring and actions for how the library decides.
Each violation gets a severity:
Score colors follow the same lines: green from 0.80, amber from 0.55 to 0.80, red below 0.55. Charts draw a dashed line at 0.70, the threshold a certificate dimension must reach to pass.
A session’s composite is the mean of its scored events, updated while the session runs and recomputed when it ends. An agent’s mean is the mean of its sessions’ composites. Agent status comes from that mean:
Fleet
Fleet is the console’s home screen. The range picker covers 24 hours, 7 days or 30 days, or a custom start date up to 365 days back that always ends now. Export downloads the “Posture by agent” table as CSV.Overview tab
A banner at the top shows the worst unresolved incident if it is critical or high. Otherwise it names the worst agent whose composite is below 0.55, and if no agent is that low, it shows the worst incident whatever its severity.
Below the numbers:
- Fleet composite posture charts the trend.
- Fleet health shows agents by status.
- Posture by agent lists every agent, worst first, with its framework, last 8 sessions, composite, change, session count and status.
- Live sessions shows the five newest sessions and updates as sessions end.
- Needs attention states how many agents are below the 0.70 threshold.
Dimensions tab
Posture shape shows the mean score of each layer in the range. Integrity heatmap shows each agent’s pass rate per layer.Agents
Agents lists every agent the workspace has seen. Profiles are built from the session start each agent sends; nothing is configured in the console. The list has views for all agents, agents below the 0.70 floor and critical-tier agents, a status filter and a risk filter, and you can save your own views in the browser.
The risk distribution above the list puts agents with no declared risk tier under “unassigned”.
Find credential takes a 64-character hex fingerprint from
agentegrity.fingerprint(secret) and lists every agent that declared that credential, with its session count and when it was last seen. The console never holds the secret or the fingerprint key, so it can only match what agents declared. See Credential provenance for how fingerprints are computed.
Merge duplicates (admins) merges two agents that are the same agent under different ids. See Merge duplicate agents.