| name | validate-agent |
| description | Validate agent definitions for consistency, model availability, handoff integrity, and tool existence. Use this when creating or modifying agents. |
Validate Agent Skill
Purpose
Ensure that agent definitions are technically sound, follow project standards, and are compatible with the current environment and other agents.
When to Use
- After creating a new agent.
- After modifying an existing agent's frontmatter (model, tools, handoffs).
- When troubleshooting "dead-end" handoffs or tool execution failures.
Workflow
1. Run Automated Validation
Execute the validation script to check for common errors in frontmatter and structure:
./scripts/validate-agents.py
This script checks:
- Model Availability: Verifies the model exists in
docs/ai-model-reference.md.
- Handoff Integrity: Verifies that handoff targets exist.
- Required Sections: Ensures all mandatory headers are present.
- Snake Case Tools: Warns about potential invalid tool names.
2. Validate Tool Existence
As the Workflow Engineer, you have access to all available tools in the workspace. You must manually verify that every tool listed in an agent's tools: array exists in your own tool list.
- Check: Compare the agent's
tools: list against your own available tools.
- Wildcards: If an agent uses a wildcard (e.g.,
github/*), ensure you have tools with that prefix.
- Naming: Ensure the tool IDs match exactly (case-sensitive). VS Code tools typically use
category/toolName or publisher.extension/toolName format.
3. Verify Boundary Consistency
Check that the agent's boundaries (✅ Always Do, ⚠️ Ask First, 🚫 Never Do) are:
- Specific: Use concrete commands and file paths.
- Actionable: The agent can actually perform the "Always Do" actions with its tools.
- Consistent: They do not contradict project-wide instructions in
.github/copilot-instructions.md.
4. Check Handoff Logic
Verify that the handoffs: section makes sense for the agent's role:
- Does the
prompt provide enough context for the next agent?
- Is
send: false used for handoffs that require user review before proceeding? (Standard for this project).
Output
- A summary of validation results.
- Fixes for any identified issues.
- Confirmation that the agent is ready for use.
