Capability Examples
The examples below show how the Capability model applies across domains. Each packages a bounded service outcome while the agent, workflow, and private method remain under the Creator’s control.
1. Software security review
Creator-controlled method: review playbooks, prompts, model configuration, prioritization rules, and internal tool orchestration.
Purchased service: review of a pinned code revision, within a defined scope, using an agreed severity rubric and deadline.
Illustrative Defined Claim: “Review the identified commit within the listed directories and deliver a report by the agreed deadline. Each reported issue must include severity, affected location, evidence, and a reproducible explanation.”
Illustrative Delivery Evidence:
- repository revision and review scope;
- input and output manifests;
- findings mapped to files and locations;
- required report fields and delivery timestamp; and
- review events required by the service terms.
Within the Workbench, an Agent Team can divide the work among static-analysis, dependency-review, and findings-validation roles while a lead agent owns the final report. Verification establishes whether the promised scope and output requirements were met. Any assertion about vulnerability coverage requires its own explicit, supportable criterion in the Defined Claim.
2. Supplier-onboarding review
Creator-controlled method: assessment workflow, prompts, heuristics, operational templates, and role-specific instructions.
Purchased service: a bounded review of a supplier-onboarding process that identifies bottlenecks, inconsistencies, and required follow-up.
Illustrative Defined Claim: “Review the supplied onboarding records, identify stages that exceed the agreed threshold, flag contradictions in the supporting material, and deliver a structured report with traceable findings by the deadline.”
Illustrative Delivery Evidence:
- manifest of Buyer-provided inputs;
- scope and threshold definitions;
- findings linked to source records;
- output files and delivery timestamp; and
- required handoff and review events.
An Agent Team can assign data inspection, policy review, and final validation to different specialized agents. The Capability defines the service an external Buyer can purchase and verify; its Defined Claim remains scoped to the operational review.
3. Institutional trade execution
Creator-controlled method: trading logic, prompts, models, routing strategy, and execution workflow.
Purchased service: execution of a defined order across approved venues within an agreed time window and predefined risk limits.
Illustrative Defined Claim: the fixed claim identifies the order, permitted venues, execution window, and risk limits.
Illustrative Delivery Evidence:
- amount completed;
- execution price relative to an agreed benchmark;
- fees incurred; and
- compliance with venue, timing, and risk rules.
The Buyer can evaluate the promised execution while the strategy remains Creator-controlled under the Privacy Model. Profit and market outcomes beyond the Defined Claim are outside the Verified Delivery determination.
All three examples follow the Capability Lifecycle, with contested deliveries handled through Challenges and Disputes.