AgentOps package
AgentOps and AI Agent Evaluations appear in the platform capability list.
Solutions · AI agent assurance
Evaluate AI agents before you trust them in production.
AgentOps and AI agent evaluations are a first-class capability. Bound what agents can do, run checks on agent-related code and configs you provide, and preserve evidence — without pretending a chat transcript is a security proof.
What this check covers
Governed execution
Approvals, budgets, dry runs, and stop controls apply when remediation is in scope.
Code-level checks
Release Check still applies to agent tooling code: secrets, auth, unsafe patterns in the repo you scan.
Example signal
[SIGNAL] Agent tool can call unrestricted shell without approval gate [PLAN] Bound tool permissions + require human approval for write actions [VERIFY] Re-test original condition after policy change [PROVE] Evaluator identity + outcome recorded
In scope
- Product positioning for agent evaluation and controlled execution
- Deterministic checks on agent-related source you submit
- Partner and enterprise design for evaluation frameworks
Limitations
- Not an automatic red-team of every third-party model endpoint
- Does not claim formal model certification
- Deep agent simulation may be package/roadmap scoped — ask what is live for your plan
Local result
Recorded result
Verified evidence
FAQ
Do you store agent prompts and chain-of-thought?
Do not assume retention. Product paths emphasize metadata and evidence over dumping raw user content. Confirm the workflow you use.
How is this different from AI code security?
AI code security focuses on generated application code. Agent assurance focuses on agent tools, permissions, and evaluation before production trust.