Huntable-CTI-Studio: Skill for Claude Code

Add Cloud Model

This skill registers a new OpenAI or Anthropic model so it shows up in the Workflow agent configuration dropdowns.

How the model list reaches the UI

The dropdown is built from config/provider_model_catalog.json. Every time load_catalog() runs (on each page load), it applies filters before returning the list:

• OpenAI: strips dated snapshots, non-chat models (audio/TTS/image/realtime/codex), then further narrows to an explicit project allowlist. Only models in the allowlist or matching gpt-5* appear in the dropdown.
• Anthropic: keeps one representative per model family — bare name preferred over -latest preferred over dated snapshot.

Important: When you save an API key in the Settings page, the app fetches the live model list from the provider's API and overwrites config/provider_model_catalog.json. The same filters run during that write. So:

• A manually-added OpenAI model that OpenAI's API doesn't return will be removed on next save.
• The allowlist is the durable gate — models in the allowlist survive refreshes because they're also returned by the API.
• Anthropic models behave the same way: survive if the API returns them, removed if not.

---

Quick decision guide

Adding an OpenAI model:

Model type	Files to change
Any gpt-5* variant	Catalog JSON + fallback catalog + context tokens (3 files)
o-series, gpt-4x, or anything else	Same 3 files + allowlist (4 files)
Reasoning model (no temperature support)	Same as above + reasoning prefixes (4-5 files)

Adding an Anthropic model:

Always: catalog JSON + fallback catalog + context tokens (3 files). The family deduplication filter runs automatically — no extra code needed.

---

Step 1 — Add to the live catalog

File: config/provider_model_catalog.json

Add the model ID string to the correct provider array. Keep it sorted.

{
  "openai": [
    "gpt-4.1",
    "gpt-4.1-mini",
    "new-model-id-here",
    ...
  ],
  "anthropic": [
    "claude-sonnet-4-6",
    "new-model-id-here",
    ...
  ]
}