← c15c15 / wf2 - Agent-Native Pass (agent-as-actor) / Activity·agent-native-v1-v1 · 2026-06-03 · draft
Qpoint
QP
Activityagent-as-actor · grounded in real telemetry

Agent Governance · Activity

What agents did — a feed of actions and decisions.

What we observe

Each entry is an observed action (a tool or model call). We see the call and the data in it; we infer the 'decision' from the sequence — we do not see the agent's reasoning between actions.

12 actions3 flagged1 blocked
WhenAgent actionStatus

invoked model generativelanguage.googleapis.com

2m ago · req_9f2a

ALLOW

Acting agent

support-copilot

sourceExe: /app/support_agent

runtime: Node 20

owner: Support

governed

The action

endpointId: generativelanguage.googleapis.com

class: unapproved ⚠ unapproved

method · status: POST · 200

authorization: ALLOW

Data exposed in this action

EMAIL_ADDRESSUS_SSN

Risk labels

auth-token

Why this matters

Sensitive data (EMAIL_ADDRESS, US_SSN) was sent to an unapproved tool or model. This exposure was allowed under the default ALLOW stance.

Create guardrail from this action →

Fields: requestId · sourceExe · endpointId · class · method · status · action (ALLOW|DENY|BLOCK) · qscan sensitiveDataType · riskLabels