Developer toolLive

Agent Governance Checklist Scorer

Score your agent governance against the checklists auditors now use.

Answers · 16 of 16

CSA agent identity governance

01
Every agent has an owner on record.CSA · weight 3
02
Every agent has a named human sponsor.CSA · weight 2
03
Static API keys for agents are eliminated.CSA · weight 3
04
Agent credentials expire, with a lifetime that cannot be extended.CSA · weight 3
05
Delegation is constrained: a callee never holds more than its caller.CSA · weight 3
06
Grants require a declared intent, a time limit and a scope limit.CSA · weight 2
07
Four kinds of step always wait for a person: widening scope, touching another agent's identity, reaching a higher-tier system, keeping data beyond the job.CSA · weight 3
08
You can reconstruct the full delegation chain for any action.CSA · weight 3
09
A change in the agent registry reaches every token issuer.CSA · weight 2
10
Multi-tenant agents are isolated at the identity layer.CSA · weight 2
11
Decommissioning removes credentials, grants, sessions and cached context.CSA · weight 2
12
Agent behaviour is monitored against its declared intent.CSA · weight 2

OWASP agentic top 10

13
Inputs are inspected for goal hijacking before an agent reads them.OWASP ASI01 · weight 3
14
Agents authenticate each other on every call.OWASP ASI07 · weight 3
15
Tool access follows least privilege.OWASP ASI02 · weight 2
16
A rogue agent can be stopped with one action.OWASP ASI10 · weight 3

Partly counts as half. n/a is left out of the score. Unanswered counts as no.

3 free runs, no sign-up

Answer sixteen questions about how your agents are run: twelve from the Cloud Security Alliance agent identity governance checklist, four from the OWASP top 10 for agentic applications. Each answer is yes, partly, no or not applicable.

The score is the weighted share of yes, with partly counting as half and n/a left out. It places you in one of the three maturity bands the checklist uses: Foundation, Intermediate or Advanced. The gaps come back heaviest first, each with one line on why it matters, and a table maps every control to what Ares does, or says plainly where a control is outside Ares.

These are the questions a security review now asks before an agent goes to production. Better to answer them here first. Everything runs in your browser. We store nothing from your input except an anonymised event: which tool ran and when.

What it checks

Ownershipan owner and a named human sponsor for every agent
Credentialsno static keys; lifetimes that expire and cannot be extended
Delegationconstrained hop to hop; grants with intent, time and scope limits
Human approvalthe four kinds of step that always wait for a person
Auditthe full delegation chain, reconstructed for any action
Lifecycleregistry changes reach issuers; decommissioning leaves nothing behind
OWASP agenticgoal hijack, inter-agent authentication, tool misuse, rogue agents
Mappingeach control against what Ares does

Sample input

{
  "owner": "yes",
  "sponsor": "no",
  "no-static-keys": "partly",
  "ttl": "no",
  "constrained-delegation": "no",
  "intent-grants": "partly",
  "human-triggers": "partly",
  "chain-audit": "no",
  "registry-propagation": "yes",
  "tenant-isolation": "na",
  "decommission": "partly",
  "behaviour-monitoring": "no",
  "goal-hijack": "partly",
  "inter-agent-authn": "yes",
  "tool-least-privilege": "partly",
  "kill-switch": "no"
}