Rediger

Del via


Agent manifests for Foundry Agent Service

Note

Agent Manifests are for educational and experimentation purposes. Resulting agents are not production ready. Review all provided resources and carefully test agent behavior in the context of your use case. Agents you create may be subject to legal and regulatory requirements, may require licenses, or may not be suitable for all industries, scenarios, or use cases. By using any template, you are acknowledging that resulting agents and other output are solely your responsibility, and that you will comply with all applicable laws, regulations, and relevant safety standards, terms of service, and codes of conduct. See the Transparency note for Azure Agent Service for more information.

Microsoft Foundry Agent Service provides a collection of pre-built agent manifests that help you easily jumpstart building and deploying agents with just a few clicks. Each manifest is an educational sample of a business automation agent and a prompt engineering learning resource. Browse manifests to find one that matches your scenario, select Create Agent, and start running it in minutes.

Manifests serve two purposes:

  • Deploy immediately — Each manifest helps you create a simple, experimental working agent that you can run without writing prompt instructions from scratch.
  • Learn prompt patterns — Each manifest demonstrates a specific prompt engineering technique, such as XML-structured instructions or autonomous decision logic, that you can study and adapt for your own agents.

Prerequisites

Browse the manifests

To browse the agent manifests:

  1. Open the Foundry portal.

  2. Navigate to your project.

  3. Select Discover > Agents.

    Screenshot of the Foundry agent manifests showing 30 agent manifests with search, sort, and tool icons for each manifest.

You can search for manifests by name or sort by Featured to find the right starting point for your scenario. Each manifest card shows the agent name, publisher, and the tools it uses.

Manifest components

Every agent manifest includes these components:

Component Description
Tools The tools the agent uses, such as web search, code interpreter, or SharePoint. Most manifests use one or two tools.
Industry Whether the manifest is cross-industry (generic) or vertical-specific, such as marketing or manufacturing.
Tone The communication style: formal, technical, coaching, conversational, or concise.
Output format What the agent produces: reports, dashboards, code, emails, narratives, presentations, or tables.
Interaction style How the agent engages: autonomous (no mid-flow input), single-shot (one request, one response), or multi-turn (ongoing conversation).
Complexity The decision-making depth: simple (linear steps), multi-step (phased execution), or decision-tree (branching logic).
Prompt structure The instruction format: XML tags, Markdown headers, natural language, or numbered rules.
Memory Whether the agent retains context across conversations (on) or treats each conversation independently (off).

Tools used by manifests

Manifests use the most common enterprise tools available in Agent Service. The following table lists each tool and its purpose.

Tool Type Description
Web search Built-in Retrieves real-time information from the public web. Most versatile tool with the lowest setup friction.
Code Interpreter Built-in Writes and runs Python code for data analysis, calculations, and chart generation.
SharePoint (preview) Built-in Searches and retrieves documents from your SharePoint sites.
GitHub (via MCP) Custom Accesses repositories, pull requests, issues, and code for developer workflows. Connected through an MCP server.
Grounding with Bing Search Built-in Retrieves cited, verifiable facts from the web with source attribution.
Azure AI Search Built-in Queries knowledge indexes for grounded, domain-specific answers.
Microsoft Fabric (preview) Built-in Connects to your enterprise data warehouse for analytics.
File Search Built-in Analyzes uploaded documents using vector search.
Browser Automation (preview) Built-in Interacts with web UIs through natural language prompts.
OpenAPI tool Custom Calls external APIs using an OpenAPI specification.

Most manifests use two tools. A few single-tool manifests show that well-crafted instructions can handle complex workflows without additional tools.

For more information about configuring tools, see Agent tools overview.

Industries and use cases

The manifests cover both cross-industry and vertical-specific scenarios:

Category Coverage Examples
Generic Cross-industry manifests for common business workflows Competitive research, status reporting, incident analysis, meeting prep
Marketing Campaign analysis, brand monitoring, content planning A/B test analysis, social campaign performance, content calendars
Manufacturing Supply chain and operations Supplier qualification, procurement automation
Retail Customer insights and store operations Review synthesis, handbook assistants
Travel and hospitality Planning and booking Trip itinerary design
Non-profit Donor management and engagement Donor engagement strategy
E-commerce Product optimization SEO optimization for product listings

Tip

Cross-industry manifests work in any vertical. A manifest like the "Competitive Landscape Researcher" applies equally to technology, retail, or manufacturing.

Available manifests

The following tables list all agent manifests currently available, grouped by industry.

Generic

Manifest Tools Interaction Complexity
Competitive Landscape Researcher Web search, Code Interpreter Autonomous Multi-step
Weekly Team Status Reporter GitHub, SharePoint Autonomous Simple
Release Notes Generator GitHub, Code Interpreter Single-shot Multi-step
Meeting Prep Briefing Web search, SharePoint, Grounding with Bing Search Multi-turn Multi-step
Data Quality Auditor Microsoft Fabric, Code Interpreter Autonomous Multi-step
Internal Policy Q&A SharePoint Single-shot Simple
Sales Metrics Dashboard Builder Microsoft Fabric, Code Interpreter Multi-turn Multi-step
Blog Post Drafter Web search, File Search Multi-turn Multi-step
Incident Postmortem Writer GitHub, SharePoint Single-shot Multi-step
Internal App Test Runner Browser Automation, GitHub Multi-turn Multi-step
Executive Weekly Digest Microsoft Fabric, SharePoint, Grounding with Bing Search Single-shot Multi-step
API Integration Troubleshooter OpenAPI tool, Code Interpreter Multi-turn Decision-tree
Document Standards Reviewer File Search, SharePoint Multi-turn Decision-tree
PR Review & Merge Assistant GitHub, File Search Single-shot Decision-tree
RFP Response Drafter Azure AI Search, SharePoint Single-shot Decision-tree
Event Venue Research & Booking Web search, Browser Automation Multi-turn Multi-step
Codebase Documentation Generator GitHub, Code Interpreter Autonomous Multi-step
Workspace Utilization Reporter Microsoft Fabric, SharePoint Multi-turn Multi-step
Industry News & Trend Scanner Grounding with Bing Search Autonomous Simple

Marketing

Manifest Tools Interaction Complexity
Content Calendar Planner Web search, SharePoint Multi-turn Decision-tree
Brand Mention Monitor Grounding with Bing Search, Code Interpreter Autonomous Decision-tree
Campaign A/B Test Analyzer Azure AI Search, Code Interpreter Multi-turn Multi-step
Social Campaign Performance Analyzer Code Interpreter Autonomous Multi-step

Manufacturing

Manifest Tools Interaction Complexity
Supplier Qualification Checker Web search, File Search Multi-turn Decision-tree
Internal Procurement Portal Browser Automation, Microsoft Fabric, OpenAPI tool Autonomous Decision-tree

Retail

Manifest Tools Interaction Complexity
Customer Review Synthesizer Code Interpreter Autonomous Multi-step
Store Operations Handbook Assistant SharePoint, Azure AI Search Multi-turn Decision-tree

Travel and hospitality

Manifest Tools Interaction Complexity
Trip Itinerary Designer Web search, File Search Multi-turn Decision-tree

Non-profit

Manifest Tools Interaction Complexity
Donor Engagement Strategist Azure AI Search, Grounding with Bing Search, OpenAPI tool Multi-turn Multi-step

E-commerce

Manifest Tools Interaction Complexity
Product Listing SEO Optimizer Web search, Browser Automation Multi-turn Multi-step

Prompt engineering patterns

Each manifest uses one of four prompt structure styles. You can study these patterns to improve your own agent instructions:

Pattern Best for Example
XML tags Complex agents with clear separation of concerns. Sections like <role>, <scope>, and <tool_strategy> make instructions easy to parse. Meeting Prep Briefing
Markdown headers Technical workflows with hierarchical organization. Uses # Role, ## Step 1 structure for clear step ordering. Codebase Documentation Generator
Natural language Coaching personas and conversational agents. Reads like a paragraph-style briefing with implicit structure. Trip Itinerary Designer
Numbered rules Strict execution contracts and autonomous decision logic. Each rule is a discrete, enforceable instruction. Brand Mention Monitor

No single pattern is universally best. Choose a structure based on how your agent needs to process its instructions: strict compliance favors numbered rules, complex multi-phase workflows benefit from XML tags, and conversational agents work well with natural language.

Create an agent from a manifest

To create an agent from a manifest:

  1. Open the agent manifests in the Foundry portal.

  2. Select a manifest that matches your use case.

  3. Select Create Agent.

  4. Configure your agent:

    • Select a model deployment (for example, gpt-5-mini).
    • Connect the required tools (for example, add a SharePoint connection or configure web search).
    • Optionally customize the instructions to match your specific requirements.

    After creation, you can continue to modify the instructions, swap tools, or change the model deployment at any time.

    Note

    Agent manifests are for educational and experimentation purposes only. Resulting agents are not production ready.

  5. Test your agent in the agents playground.

  6. When you're satisfied with the results, publish your agent.

Tip

Start by running a manifest with its default instructions before customizing. This helps you understand the agent's workflow and identify which parts to adjust for your scenario.