# AI Agent Role Canvas

Use one canvas per bounded agent role. Replace every prompt below with a concrete, testable statement.

## 1. Job purpose

- Business outcome:
- Workflow boundary:
- Human owner:
- Explicit non-goal:

## 2. Trigger and inputs

- Trigger:
- Required inputs:
- Authoritative sources:
- Freshness requirement:

## 3. Responsibilities and outputs

- Permitted steps:
- Required output:
- Completion condition:
- Evidence to retain:

## 4. Tools and permissions

| System or tool | Data scope | Read | Prepare | Write | Send | Approve | Other limits |
|---|---|---|---|---|---|---|---|
| | | | | | | | |

## 5. Constraints and prohibited actions

- Required validations:
- Limits and thresholds:
- Retry or time limits:
- Prohibited actions:
- Retention rules:

## 6. Escalation

- Escalation triggers:
- Destination and urgency:
- Evidence package:
- Manual fallback:

## 7. Measures and evaluation

- Outcome measure:
- Quality measure:
- Control measure:
- Operations measure:
- Representative test cases:

## 8. Ownership

- Business owner:
- Technical owner:
- Risk/control owner:
- Operations/incident owner:
- Review cadence:

Source: https://shuhaib.tech/blog/ai-agent-job-description