preset-health

Preset Health Check (read-only)

A repeatable, read-only procedure for testing whether saved presets are healthy, classifying any failures, and producing an actionable report — without printing or mutating credentials.

Read-only contract. This reference enumerates, probes, and reports. It does not edit ~/.lingtai-tui/presets/saved/*.json, agent init.json, or any manifest.preset.allowed list. Remediation is described as a recommendation for the human to confirm and apply (or to explicitly authorize you to apply through the normal preset tooling). Never run a "fix" step automatically.

When to use

• The human asks "are my presets still working?" or "which preset is broken?"
• system(action="presets") surfaced a non-ok connectivity status (for example no_credentials, an auth failure, or an unreachable endpoint).
• A provider rotated keys, retired a model, or moved an endpoint and you need to find every saved preset affected.
• A new agent needs a clear procedure to triage presets before relying on them.

Where saved presets live

Saved (user-owned) presets are JSON files here:

~/.lingtai-tui/presets/saved/*.json

Built-in templates live under ~/.lingtai-tui/presets/templates/ and are rewritten on every lingtai-tui bootstrap — the directory itself is the marker distinguishing built-in from user-owned, so a health check focuses on saved/. The skill tree this reference ships in is extracted to ~/.lingtai-tui/utilities/swiss-knife/reference/preset-health/SKILL.md.

Each preset's relevant shape (other fields omitted):

{
  "manifest": {
    "llm": {
      "provider": "gemini",
      "model": "gemini-2.0-flash",
      "base_url": "https://...",
      "api_key_env": "GEMINI_API_KEY"
    }
  }
}

The credential is referenced indirectly through api_key_env (an environment variable name), not stored inline. That indirection is what lets a health check report "credential missing" without ever reading the secret value.