Custom agent frontmatter `model` field rejects array syntax — incompatibility between Copilot CLI and VS Code Copilot Chat

[deyil]

Describe the bug

When a custom agent .agent.md file uses an array for the model frontmatter field (as supported by VS Code Copilot Chat), the GitHub Copilot CLI refuses to load the agent and throws a parse error.

Error message from Copilot CLI:

The following agents failed to load:
 .github/agents/Coordinator.agent.md: custom agent markdown frontmatter is malformed: model: Expected string, received array

Agent frontmatter that triggers the error:

---
name: Coordinator
description: Pure managerial agent that orchestrates implementation by delegating all technical work to specialized subagents.
argument-hint: Coordinate a plan, phase, feature, or code change
tools: [vscode, execute, read, agent, search, web, todo]
model: [GPT-5.4 (copilot), Claude Sonnet 4.6 (copilot)]
---

The model field being an array is intentional — it allows the agent to declare multiple model fallbacks or alternatives. VS Code Copilot Chat loads and uses this agent correctly. However, the Copilot CLI strictly expects model to be a string, causing a schema validation failure and preventing the agent from loading entirely.

Affected version

No response

Steps to reproduce the behavior

1. Create a custom agent file at .github/agents/MyAgent.agent.md with the following frontmatter:

---
name: MyAgent
description: An agent with multiple model options
model: [GPT-5.4 (copilot), Claude Sonnet 4.6 (copilot)]
---

2. Run any copilot CLI command that triggers agent loading (e.g. gh copilot suggest or open a session that scans .github/agents/)
3. Observe the error output on checking /agent

Expected behavior

The Copilot CLI should accept the model field as either a string or an array of strings, consistent with how VS Code Copilot Chat parses the same .agent.md frontmatter. When an array is provided, the CLI could use the first entry as the primary model, or support model fallback/selection semantics.

At minimum, the schema validation for model should align between Copilot CLI and VS Code Copilot Chat to avoid agents that work in one environment failing to load in the other.

Additional context

No response

[brolnickij]

Hit the same class of issue. In our case it was not array syntax but display names vs slug IDs.

CLI rejected all of these:

model: Claude Opus 4.6 (copilot)
model: GPT-5.4 (copilot)
model: Gemini 3 Flash (Preview) (copilot)

Fix: use slug IDs from ~/.copilot/config.json — claude-opus-4.6, gpt-5.4, claude-haiku-4.5, etc.

So there are at least 3 incompatibilities between CLI and VS Code for model::

1. Array syntax (this issue) — CLI rejects, VS Code accepts
2. Vendor suffix (copilot) — CLI rejects, VS Code uses for handoffs
3. Display names vs slugs — CLI wants slugs, VS Code wants display names

Copilot CLI v1.0.11, macOS.

[MetalHexx]

Yeah this is giving me trouble right now in my RadOrchestration project. I'm trying to make my system work cohesively between the vscode and cli and I'm running into a lot of friction aligning my agent definitions between both tools.