PrymeIntelligence

Agent governance·Template

Agent role charter

An AI agent role charter template: what the charter defines, when to write one, how to fill each section, and a worked sample for a claims intake agent.

An agent role charter is a one-page document that defines a single AI agent's job: its purpose, the work it may do, the work it must refuse, its sources, the depth of each capability, its owner and its review date. It is the agent's job description and the boundary every action is tested against.

What does the charter do?

It turns "the claims agent" into something you can govern. Without a charter, an agent's scope drifts as people add tasks to its brief one request at a time. With one, every addition is a change to a signed document, and someone accountable has to agree.

The charter is also the written form of role scope, the first of the five bounds every action meets. The concept is set out in role scope: what it means for AI agents. The charter states it; the platform enforces it.

When should you write one?

Before the agent is built, not after. Writing the charter first forces the team to agree what the agent is for and, more usefully, what it is not for. Rewrite it when the agent's job changes; amend it when a capability is added or removed.

A charter should cover one agent. If you find yourself writing "and also" more than twice, you may need two agents. Setting scope is explained in how to set an AI agent's role scope.

How do you fill in each section?

  • Purpose is one sentence a customer would understand.
  • In scope lists tasks as verb and object. Be specific: "log a new motor claim" not "handle claims".
  • Out of scope lists what the agent must refuse or route. This list does more work than the in-scope list, because it is what stops the agent going beyond its role.
  • Capabilities and depth gives each capability its depth. A charter states the intended depth; actual depth follows from evidence.
  • Sources names the systems the agent may read, and which are authoritative.
  • Systems it may act in lists the connections it uses to act, separate from what it reads.
  • Accountable owner is a named person, not a team.

For a fuller list of what to include, see what an AI agent's role description should include.

The agent role charter template

Section What to write
Agent name and ID Unique name and inventory reference
Purpose One sentence
Business unit Team the agent works for
Accountable owner Named person
In scope Tasks, verb and object
Out of scope Tasks it must refuse or route, with route
Capabilities and intended depth Each capability with depth 1 to 4
Sources it may read Systems, with which is authoritative
Systems it may act in Connections and the actions in each
Customer limits Amount, volume, time or population limits
Escalation Who receives held items and refusals
Interacts with Other agents and people, and how
Records Where its verdicts and actions are filed
Review Frequency and trigger events
Approved by and date Named approver

A worked sample

A hypothetical home insurer's claims intake agent.

Section Entry
Purpose Take first notice of home insurance claims and prepare the claim file for a handler
Accountable owner Head of home claims
In scope Collect claim details from the customer; check the policy is in force; check the peril is listed on the policy; open the claim record; request photos; draft the acknowledgement
Out of scope Deciding cover (route to handler); discussing settlement amounts (refuse, route); handling suspected fraud (route to fraud team without telling the customer); claims under commercial policies (route to commercial desk)
Capabilities Read policy: 1. Open claim record: 3. Draft acknowledgement: 2. Request photos via portal: 3
Sources Policy administration system (authoritative for cover); claims system; customer's statements (declared)
Systems it may act in Claims system (create record only); customer portal (send photo request)
Customer limits Home policies only; claims notified within the policy period
Escalation Home claims queue, priority flag for escape of water and fire
Review Quarterly, and on any change to the claims process

Notice that "check the peril is listed" is in scope but "deciding cover" is not. The agent can tell a handler that storm damage is a listed peril; it cannot tell the customer they are covered.

The role charter maps neatly onto access rights, in the same way a job role maps to role-based access control: each in-scope action becomes a permission, and nothing else is granted.

Where Pryme fits

In Pryme, Agent Studio assembles a candidate brief from the role you describe, and the charter's in-scope and out-of-scope lists become the role scope bound. Any request outside it is refused at the moment of action, and the verdict names role scope as the deciding bound. More in agent governance is a bound, not a model.

Questions people ask

Who should sign the charter?

The accountable owner, plus the risk or compliance lead for that business line. If the agent touches customer money or decisions, add the senior manager accountable for that area.

How long should a charter be?

One page. If it runs to several, the agent probably has too many jobs, or the charter is carrying detail that belongs in the agent's brief or test plan.

Does the charter replace the agent's brief?

No. The charter is the governance document; the brief is the working instruction set the agent runs on. The brief must stay inside the charter.

Decide what your agents may do. Prove it. Enforce it.

Pryme Intelligence certifies each agent on your own cases, checks every action against five bounds at the moment it happens, and files every verdict.