Share via

Best practices for configuring your manifest

This article lists best practices for configuring your manifest YAML. For partners publishing to Security Store, see Publish to Store.

  1. There can only be one AgentDefinition in the Agent manifest yaml.

  2. FetchSkill and ProcessSkill must be namespaced by skillset. They must follow the format Skillset name.Skill name. See, FetchSkill and ProcessSkill.

  3. A few pointers on ChildSkills:

  • If a skill is defined in the same manifest YAML, it must be referenced under ChildSkills. From the following example, SkillA is a GPT skill defined in the manifest and added to the child skill.

  • ChildSkills can be defined in an OpenAPI specification manifest. For example, SkillB should have been defined in the OpenAPI spec manifest. For more information, see API tool sample.

  • It can be part of an existing plugin that is in Requiredskillsets in the Agentdefinition. For example, if SkillC is defined in another manifest YAML, it must be referenced under Requiredskillsets.

    
SkillGroups:

 - Format: Agent
    - Name:
      Settings:
      ChildSkills:
       - SkillA # part of the current manifest (plugin) yaml
       - SkillB # this should be part of the API OpenAPI skill group
       - SkillC # if it is in another manifest, the manifest must be included in the RequiredSkillSets of the AgentDefinition
  - Format: GPT
    - Name: SkillA
      template:
  - Format: API
    Settings:
      openApispecUrl: https://raw.githubusercontent.com/microsoftgraph/msgraph-metadata/main/v1.0/swagger.json
 AgentDefinitions:
  - Name: AgentDefinitionSample
   Description: This skill provides access to Microsoft Graph API functionalities.
   RequiredSkillSets:
    - MicrosoftGraphAPI # has SkillC

  1. The skillset name, which is the Descriptor.name should match the RequiredSkillsets value For example, from the manifest YAML sample Descriptor.name=Contoso.SecurityOperations.Samples-090925, which should be added to RequiredSkillsets.

  2. The Agent manifest supports only one authentication scheme per Agent. For more information, see Authentication.

  • Last updated on