Free resource · Word + Excel kit
Claude Agents Kit: Create, test and manage your agents
Project, skill, CoworkClaude Tag, sub-agent, or Agent SDK: there are six ways to create an agent with Claude. This kit helps business teams and AI experts in large organizations choose the right one, design the agent, test it, and manage it, with 20 agent ideas per function.
The Agents Claude kit, in brief
- What it is
- A kit to create agents with Claude in a large organization: choose the right building block, design the agent with a sheet in eight sections, test it on twenty cases, supervise it and manage it for 30 days.
- For who
- Business teams that want to automate a recurring task, AI experts who support agent creators, and CIOs who need to set the framework.
- Format
- 12-page Word document in 8 chapters and Excel workbook in 7 tools (choice of agent type, design sheet which generates instructions, catalogue of 20 ideas, set of 20 tests, risk grid, agent register, pilot and measurement of time saved).
- What it contains
- What is called a Claude agent; the six ways to create one; how to choose; the design sheet and instruction template; twenty agent ideas per function; the testing method; the eight framing rules; the 30-day pilot.
- What is it based on?
- Anthropic's help center and documentation accessed on October 25, 2026, Claude Tag and Managed Agents' announcements, and our large organization agent creation workshops.
- Price
- Free, in exchange for a professional email address. No unsolicited emails.
- Update
- Edition October 2026, information verified on October 25, 2026. Several features are in beta: check them before deploying.
- What it doesn't replace
- Neither your organization's security policy nor the specific analysis required for high-risk uses as defined by the AI Act.
- Go Further
- Claude Skills kit, Cowork kit, Deploying Claude in the Workplace Kit, Agent Party.
The word "agent" encompasses very different realities. To quickly decide, a simple definition suffices: an agent is an assistant to whom you entrust a goal rather than a question. It plans its steps, uses tools (files, search, connectors to your applications) and delivers a result, sometimes without anyone asking it to.
With Claude, most useful enterprise agents can be built without writing a single line of code. The difficulty isn't technical: it lies in choosing the right component, ensuring precise design, and establishing a framework before deploying the agent. This kit addresses these three points.
1. The autonomy scale of an agent Claude
| Level | What Claude does | Example |
|---|---|---|
| 1. Assistant | Responds within a defined framework (instructions, reference documents) | An assistant who answers questions about travel policy |
| 2. Equipped Agent | Uses connectors and skills to read and produce | An agent preparing a client appointment using the CRM |
| 3. Delegate Agent | Complete a multi-step project and deliver a file | An agent who consolidates ten tables and writes the summary |
| 4. Recurring Agent | It triggers automatically, according to a schedule. | Competitive intelligence delivered every Monday |
| 5. Integrated Agent | Runs within an application, without a chat interface | Automated processing of incoming requests |
The higher you go on the scale, the higher the value, and the more robust the framework needs to be. The golden rule: start at the lowest level that works, then move up a level when the result is reliable.
2. Six ways to create an agent with Claude
| Brick | What it is | For who |
|---|---|---|
| Project with instructions | A space with permanent guidelines and reference documents, shareable with a team | All |
| Skill | A skill that is documented once Claude applies it in the chat. Cowork and Claude Code | Reference persons, experts |
| Cowork and planned tasks | Delegated work in several stages, whether one-off or recurring; since October 6, 2026, the new Pro and Max sessions run in the cloud, even when the computer is turned off. | Professions, managers |
| Claude Tag in Slack | A Claude shared by channel, which maintains the team context; in beta since June 23, 2026 on Team and Enterprise | Teams on Slack |
| Sub-agents Claude Code | Specialized agents described in a file in the .claude/agents/ folder, with limited tools | IT, data, business developers |
| Agent SDK and Managed Agents | Agents integrated into your applications in Python or TypeScript, or hosted by Anthropic (public beta since April 2026) | Technical teams |
3. Choosing the right type of agent in five questions
- Answering questions about a stable corpus? A project with instructions is sufficient.
- Always apply the same method or the same model? Add a skill that can be used everywhere.
- Produce a complete deliverable, or start over every week? Cowork, with a scheduled task if it is recurring.
- Serving an entire team in their daily interactions? Claude Tag if the team lives in Slack, otherwise a shared project.
- Integrate the agent into an application or process large volumes? Agent SDK or Managed Agents, with the IT department.
If your organization also uses Microsoft 365 Copilot, the Copilot agents (Agent Builder, Copilot Studio) remain closely aligned with Microsoft 365 data; the Copilot Studio kit treats this slope.
4. The design sheet in eight sections
| Status | Example: appointment scheduling agent |
|---|---|
| 1. Assignment | A one-page document for the salesperson before each client meeting. |
| 2. Trigger | The salesman, the day before the appointment |
| 3. Entrances | Customer name, CRM history, public news |
| 4. Tools | Read-only CRM connector, web search, "customer record" skill |
| 5. Steps | History, current events, likely needs: three questions to ask |
| 6. Exit | One page, four fixed sections, cited sources |
| 7. Safeguards | Never enter data into the CRM, contact the client, or invent a figure. |
| 8. Human validation | The salesperson rereads before the appointment |
In the binder, the completed form automatically generates the agent's instructions, ready to be copied into a project or task. Cowork.
5. Twenty agent ideas by function
| Function | Agents |
|---|---|
| Board | Weekly summary note of indicators (Cowork planned), preparation of committee files (Cowork) |
| HR | Frequently asked questions about agreements and procedures (project), in-house format job descriptions (skills) |
| Finance | Monthly commentary on budget variances (Cowork planned), consistency check before broadcast (skill) |
| Legal | Comparison of two versions of a contract (draft), weekly regulatory monitoring (Cowork planned) |
| Buying and selling | Analysis of responses to a call for tenders (Cowork), preparing client meetings (project and connectors), following up on draft proposals (Cowork planned) |
| Marketing and communication | Monday's competitive intelligence (Cowork planned), adapting a message for each channel (skill) |
| Project, support and quality | Reporting and monitoring of actions (skill and Claude Tag), team progress update (Claude Tag), answers from the knowledge base (project), written procedure based on an oral operating procedure (skill) |
| IT and data | Code review and missing tests (Claude Code sub-agents), large-scale processing of incoming documents (Agent SDK or Managed Agents) |
6. Test and provide guidance before sharing
An agent who answers correctly three times is not tested. The kit offers a set of twenty scenarios in four categories: nominal, boundary, out-of-bounds, and traps. The traps verify that the agent is unaware of instructions hidden in a web page or document. Recommended threshold: 100% on the traps and at least 85% overall.
- An owner named by agent, and one alternate.
- The slightest privilege : only the necessary connectors, read-only whenever possible.
- No irreversible action without human validation : the agent prepares, a human sends, publishes or deletes.
- Sources cited in each deliverable.
- A spending cap and a justified frequency for planned tasks.
- A register of agents and a quarterly magazine.
Most business agents fall under the minimal risk of the AI Act, but some uses tip into high risk, such as application sorting or employee evaluation: these require specific analysis before any deployment.
7. What the kit contains
The Word document has 12 pages in 8 chapters: what is called a Claude agent; the six ways to create one; choosing the right type; the design sheet and instruction template; twenty agent ideas per function; testing before opening; framing agents; the 30-day pilot; the sources.
The Excel workbook "Claude Agents Kit Tools" contains seven tools: the choice of agent type in five questions; the design sheet which makes up the instructions; the filterable catalogue of 20 ideas with building block, trigger, connectors, risk and estimated gain; the set of 20 tests with success rate and commissioning decision; the risk grid; the agent register with review alert; the four-week pilot and the measurement of time saved.
Download the Claude Agents kit (Word and Excel, free)
12 pages in 8 chapters and a binder of 7 tools. Written by NextStart.AI based on Anthropic documentation and our agent creation workshops for large organizations.
- The six ways to create an agent and how to choose
- The design sheet that generates the instructions
- Twenty agent ideas by function
- The set of 20 tests, the risk grid and the register
8. Frequently Asked Questions
What is an Agent Claude?
This is a configuration for Claude, who is given a goal rather than a question: he plans his steps, uses tools like files, web search, or connectors to your applications, and delivers a result. It can take the form of a project with instructions, a skill, or a task. Cowork, from Claude Tag in Slack, from a Claude Code sub-agent or from an agent built with the Agent SDK.
Do you need to know how to code to create an agent with Claude?
No, not for most business needs: projects, skills, tasks Cowork Claude Tag and Claude Code are configured using natural language. Code is only required for the Claude Code sub-agents, the Agent SDK, and Managed Agents, which are handled by the technical teams.
Can an agent named Claude work alone each week?
Yes, with a scheduled task in Cowork : schedule, daily, weekly, or weekdays. Since October 6, 2026, the new sessions Cowork Pro and Max offers run in the cloud, even when the computer is turned off; access to local files, however, requires the desktop application to be open.
What is Claude Tag?
This is Claude's team agent in Slack, launched in beta on June 23, 2026 for Team and Enterprise plans. It is addressed as @Claude in a channel; it maintains the team context, and administrators choose its channels, tools, data, and spending limits.
What is the difference between Agent SDK and Managed Agents?
The Agent SDK is a Python or TypeScript library that integrates the Claude Code mechanics into your own applications, on your infrastructure. Managed Agents hosts the agent with Anthropic in an isolated environment, which has been in public beta since April 2026 and is billed on a pay-as-you-go basis.
How can we prevent an agent from making a serious mistake?
Through four rules: rights limited to the bare minimum, read-only initially; no irreversible actions without human validation; sources cited in each deliverable; a set of tests passed before opening and after each modification.
Is the kit free?
Yes. The Word document and Excel workbook are free, in exchange for a professional email address which will only be used to send you the download link. No unsolicited emails.
9. Going further
To provide your agents with reusable skills, the Claude Skills kit provides ten, and the Claude Plugins Kit explains how to distribute them to a team. To delegate entire tasks, the Cowork kit ; to establish the general framework, the Deploying Claude in the Workplace KitOn the page AI AgentsDiscover our workshops and courses, from theAgent Party withAgent Studio.
Read also : Claude Cowork, Copilot Cowork, ChatGPT Work: the comparison for businesses et Claude Code in business: the guide to getting started and managing it.
The entire series of free kits. To drive adoption: AI charter template, AI Ambassadors Kit, Manager's toolkit for dealing with AI, AI-powered time-saving kit, Adoption Roadmap Kit, AI maturity grid, AI Measurement and ROI Kit, AI Governance Kit, Cowork kit, 50 Copilot M365 prompts, Copilot Studio kit, Deploying Claude in the Workplace Kit, Claude Agents Kit, Claude Skills kit, Claude Plugins Kit, Claude Code kit, Claude Design kit et Vibe coding kitFor the AI project sponsor: AI Act kit, GDPR kit, Security and data kit, Shadow AI kit, AI and Social Dialogue Kit, frugal AI kit, AI Sovereignty Kit et responsible AI kitBy profession: AI HR kit, AI marketing kit, AI finance kit, AI kit for lawyers, AI kit for salespeople, AI toolkit for project managers, AI innovator kit et communicator's AI kit.
10. Sources
The information cited on this page and in the kit is based on the following sources, verified on October 25, 2026.
- Anthropic, Use Claude Cowork on web, desktop, and mobile.
- Anthropic, Schedule recurring tasks in Claude Cowork.
- Anthropic, Subagents in Claude Code.
- Anthropic, Agent SDK overview.
- Help Net Security, Claude Managed Agents (April 9, 2026).
- VentureBeat, Anthropic launches Claude Tag (June 23, 2026).