Template Library
AI Agent Charter and Runbook
An Excel workbook for one AI agent or AI-supported process: a charter with a readiness checklist, a permissions register, guardrails, a runbook, monthly KPIs against targets, and a change log.
The Charter states the agent's purpose, owner, supervisor, scope in and out, autonomy level, success measures, and promotion criteria, with a 15-item readiness checklist that calculates a readiness percentage. The Permissions sheet lists every system and action the agent may use, with access type, scope, data class, and whether approval is needed, and flags write, send, spend, or delete access without approval.
The Guardrails sheet holds the actions that always need a person, the stop conditions, and the limits per task. The Runbook holds monitoring checks, escalation levels, incident steps, and the kill switch and rollback. Monthly KPIs calculates success, escalation, and override rates and cost per task against the targets on the Settings sheet, and the Change Log records every change with its test and approval. The Dashboard summarizes readiness, permissions, and the latest KPIs.
An example work-order triage agent at an illustrative plant is included so you can see a finished charter before you replace it.
What Is Included in the Workbook
| Sheet | Purpose | What Teams Capture |
|---|---|---|
| Start Here | Orientation and workflow | What the workbook does and a numbered workflow with a link to the sheet for each step |
| Dashboard | Agent summary | Readiness, permissions and permission checks, latest success and override rates, incidents, permissions by access type, and a chart of monthly rates |
| Charter | The agent's job description | Purpose, owner, supervisor, scope in and out, autonomy level, systems and data, success measures, promotion criteria, approval, and a readiness checklist |
| Permissions | Access register (Excel Table) | System, action, access type, scope, data class, approval needed, reason, approver, review date, and a calculated check |
| Guardrails | Approval gates, stop conditions, and limits | What always needs a person, when the agent must stop, and limits on steps, time, tool calls, spend, and rate |
| Runbook | Monitoring, escalation, and incident response | Monitoring checks with owners and frequency, escalation levels, incident steps, and how to stop and roll back |
| Monthly KPIs | Performance against targets (Excel Table) | Tasks, successes, escalations, overrides, incidents, and cost, with calculated rates and a target check |
| Change Log | Record of changes (Excel Table) | Date, change, type, reason, tested before use, approver, and rollback plan |
| Settings | Targets, as-of date, and lists | KPI targets, the as-of date, and the lists for autonomy level, access type, data class, change type, and frequency |
Key Features Inside the Template
Described Like a Job
A charter with purpose, scope, owner, and autonomy level, and a readiness checklist before go-live.
Permissions You Can Audit
Every access listed, with the least needed, and flags for risky access without approval.
A Runbook for When It Goes Wrong
Escalation paths, incident steps, and a tested kill switch and rollback.
KPIs and a Change Log
Success, escalation, and override rates against targets, and a record of every change.
Best Use Cases
- Preparing an AI agent for approval by a governance group
- Documenting an agent's permissions and limits for audit
- Running a shadow test and recording the promotion criteria
- Monitoring an agent in production
- Teaching agent governance
How to Use the Template Effectively
- Write the charter, and complete the readiness checklist honestly.
- List every permission, and give the least access that works. Mark the actions that need approval.
- Set the guardrails: approval gates, stop conditions, and limits.
- Write the runbook, and test the kill switch and rollback before go-live.
- Enter the KPIs each month and compare them with the targets.
- Log every change to instructions, tools, models, permissions, data, and limits, and review the charter at least once a year.
Who Should Use This Template
- Process owners responsible for an AI agent
- Operations and IT leads
- AI governance and audit teams
- Students and trainees learning agent governance
Common Mistakes to Avoid
- Giving the agent broad permissions because it is easier
- Writing the charter after go-live
- A kill switch that has never been tested
- Promoting the agent on feel instead of the written criteria
- Changing instructions or models without a log
Important Notes
AI Agent Charter and Runbook Frequently Asked Questions
How is readiness calculated?
On the Charter sheet, the readiness percentage is the number of Yes answers divided by the number of applicable items, which excludes those answered N/A.
What does the permission check flag?
It flags any permission with write, send, spend, or delete access that is not marked as needing approval, and any use of Restricted data without approval. These are the accesses most likely to turn a mistake into harm.
Can I use it for more than one agent?
The workbook is designed for one agent or process. Save a copy for each agent, so that each has its own charter, permissions, KPIs, and change log.