--agent code-review fails before inference with "Selected custom agent system prompt callback is unavailable" #5178

Describe the bug

Selecting the built-in code-review profile with copilot --agent code-review fails with an internal system-prompt callback error before inference. A normal CLI coordinator can successfully invoke code-review as a synchronous task in the same environment.

It is unclear whether direct selection of this built-in is supported. If it is intentionally subagent-only, the CLI should reject this selection with actionable guidance rather than an internal callback failure.

Affected version

GitHub Copilot CLI 1.0.98.

Steps to reproduce the behavior

Run a code-free prompt with the built-in selected:

env -u GH_TOKEN -u GITHUB_TOKEN copilot \
  --agent code-review \
  --model gpt-6.1-sol \
  --reasoning-effort xhigh \
  --silent --stream off --log-level error \
  --no-ask-user \
  -p 'Reply with exactly OK. Do not read files or call tools.'

The reproduction uses the existing Copilot login and does not require a pull request, source files, or an actual code review.

Observed result:

• Exit status: 1.
• Stdout: empty.
• Stderr:

Error executing prompt: Error: GenericFailure, host interaction call failed: Error: Selected custom agent system prompt callback is unavailable

Separately captured usage output reported:

{
  "totalUserRequests": 0,
  "totalApiDurationMs": 0,
  "modelMetrics": {},
  "agentMetrics": {},
  "lastCallInputTokens": 0,
  "lastCallOutputTokens": 0
}

These are selected fields, not the complete usage document.

Expected behavior

Either:

1. If direct built-in selection is supported, execute the selected profile with the supplied prompt.
2. If code-review is intentionally subagent-only, fail early with a clear message explaining the restriction and the supported invocation route.

Please clarify the intended support for built-in agents with --agent.

Additional context

• Operating system: macOS 26.6.2.
• Architecture: arm64.
• Requested model and reasoning effort: gpt-6.1-sol, xhigh.
• Successful comparison: a normal CLI coordinator invoked exactly one synchronous task with agent_type: code-review. The reviewer completed and its final response was returned to the coordinator.
• Direct selection and task delegation are distinct routes. The successful delegated run does not establish the cause of the direct-selection failure.
• The root cause is unknown. This report does not assume that a missing custom agent file is responsible.
• The failure also occurred in a separate review invocation; the code-free reproduction above removes the need to disclose that repository or its contents.
• No private source code, repository identifiers, credentials, session identifiers, or raw execution logs are included.