Agent Definition Canvas

Complete this canvas before writing the implementation prompt. Keep each statement testable.

1. Identity

• Name and version
• Owner
• Agent type: control-plane, domain or assurance
• SAP Activate phases
• Status: concept, pilot, approved, suspended or retired

2. Outcome

• What transformation result should improve?
• Who receives the result?
• What current problem or delay does it address?
• What measurable baseline exists?

3. Mission and boundaries

• What tasks are inside scope?
• What is explicitly outside scope?
• What events trigger the agent?
• When must it stop or escalate?

4. Skills and knowledge

• Which durable professional skills are required?
• Which process packs—such as Lead-to-Cash, Procure-to-Pay or Design-to-Operate—apply?
• Which SAP product, industry, country and client skills apply?
• Which sources are authoritative, and who owns them?

5. Tools and permissions

• Which repositories and systems may it read?
• Which artifacts may it create or update?
• Which actions are reversible?
• Which tools require independent policy checks or approval?

6. Human–agent responsibilities

• Who initiates, contributes context, reviews, decides, executes and verifies?
• Which decisions always remain human?
• Which conflicts require independent challenge?

7. Evidence and controls

• What provenance accompanies an output?
• Which deterministic rules and policy checks apply?
• How are separation of duties and least privilege maintained?
• How are exceptions, overrides and incidents handled?

8. Evaluation and operation

• What representative and adverse cases will be tested?
• Which quality, risk, adoption, cost and value measures apply?
• What constitutes an unacceptable failure?
• What changes trigger reevaluation?
• How can the agent be suspended or retired?

9. Dependencies

• Which agents provide inputs or receive outputs?
• Which shared knowledge, identity, policy and observability services are required?
• What happens when a dependency is unavailable?