Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.xpertai.cn/llms.txt

Use this file to discover all available pages before exploring further.

Execution audit is the foundation for productionizing the UOSE system. Every Agent tool call and key user decision should leave a traceable record for review, compliance, and issue diagnosis.

​
Audit Objects

Audit records usually contain:
  • requestId or taskId.
  • traceId.
  • tenantId and organizationId.
  • resourceId.
  • actorType and actorId.
  • toolName or actionTypeCode.
  • mappingVersion.
  • snapshotId.
  • status.
  • policyDecision.
  • decisionSummary.
  • evidenceRef.
  • input.
  • output.
  • duration and timestamp.
These fields let audits answer: who did what, when, based on which resource and semantic version, why the system allowed or rejected it, and what the result was.

​
Audit Entry Points

Users can access audits through:
  • The execution audit page, filtered by status, resource, or time.
  • getAuditTrace, used to review one task flow by taskId.
  • auditRef in approval requests, linking to related execution records.
  • Workbench artifacts, used to review execution effects and visualizations.
The audit page focuses on runtime facts, not configuration snapshots.

​
Evidence References

The evidenceRef in audit records stores structured evidence summaries, such as:
  • Number of matched entities.
  • graphVersion.
  • allowed actions.
  • denied actions.
  • policyId.
  • approvalRequestId.
  • snapshotId.
  • rowCount or result summary.
Evidence references should not store sensitive credentials or put large full results into audits. Large results should be stored in dedicated artifacts or caches.

​
Audit and Governance Loop

Audit results can improve governance:
  • If Agents often hit non-unique entities, add aliases or narrow resource scope.
  • If certain actions are frequently denied, check whether policies are too strict or documentation is misleading.
  • If queries time out, adjust adapter timeout, cache, or sync granularity.
  • If write action approval information is insufficient, require expected effect or more detailed parameters.
Audit is not post-fact archiving. It is a feedback system for continuous UOSE improvement.

​
Best Practices

  • Pass a stable taskId for each Agent task.
  • Reuse the same taskId across multi-step flows to make tracing easier.
  • Record audits for low-risk queries too, so users can review answer sources.
  • Keep approval requests and policy hit reasons for high-risk actions.
  • Do not store tokens, passwords, cookies, or full sensitive business data in audits.