| name | agent-creator |
| description | Create a new Microsoft Agent Framework agent from a natural-language description or manual instruction edits. Use when the user wants to build a new AI agent, modify an existing agent's instructions, or generate an agent configuration. Outputs a validated agent definition (YAML + Python code) that can be deployed to Azure AI Foundry.
|
| license | MIT |
| compatibility | Requires python3, agent-framework SDK |
| metadata | {"author":"poc-team","version":"1.0"} |
| allowed-tools | load_skill read_skill_resource run_skill_script |
Agent Creator Skill
Purpose
This skill enables an orchestrator agent to create and validate Microsoft
Agent Framework agent definitions from either:
- Natural-language prompts – The user describes the agent they want in plain
language, analogous to creating a GPT in ChatGPT.
- Manual instruction editing – The user provides explicit instruction text,
tool lists, and model parameters.
Workflow
Step 1 – Gather Requirements
Collect the following from the user or from the orchestrator context:
| Field | Required | Default | Description |
|---|
name | yes | — | Unique snake_case agent name (max 64 chars) |
description | yes | — | One-line description of the agent |
instructions | yes | — | System prompt / instruction text for the agent |
model | no | gpt-4o | Azure OpenAI deployment name |
tools | no | [] | List of tools the agent may call |
context_providers | no | [] | Context providers (e.g., SkillsProvider) |
temperature | no | 0.7 | Sampling temperature |
Step 2 – Generate Agent Definition
Produce two artefacts:
agent.yaml – declarative agent configuration:
name: <name>
description: <description>
model: <model>
instructions: |
<instructions>
tools: <tools>
temperature: <temperature>
agent.py – executable Python code using agent_framework:
from agent_framework import AzureOpenAIChatClient
from azure.identity.aio import DefaultAzureCredential
agent = AzureOpenAIChatClient(
credential=DefaultAzureCredential(),
endpoint="<AZURE_OPENAI_ENDPOINT>",
deployment="<model>",
).as_agent(
name="<name>",
instructions="""<instructions>""",
)
Step 3 – Validate
Run the validate_agent script to check:
- YAML is well-formed
- Required fields present
agent.py has no syntax errors- Instructions length < 32,000 chars
If validation fails, report the errors and suggest fixes.
Step 4 – Persist
Save both files to generated/agents/<name>/.
Natural Language → Agent Generation
When the user provides a natural-language description like "I want an agent that
helps with expense reports", generate appropriate:
- Name – derive from the description (e.g.,
expense_report_assistant)
- Instructions – craft detailed system prompt capturing expertise, tone,
constraints, and response format
- Tools – suggest relevant tools based on the domain
- Description – concise summary
Quality Guidelines
- Instructions should be 200–2000 characters
- Be specific about the agent's domain and boundaries
- Include output format guidance where applicable
- Add safety guardrails (out-of-scope handling, refusal policies)
