skills.rest
catalystctl
catalystctlOFFICIAL
Explore all skills by this author →
WebsiteGitHub

diagnose-provider-fallback-models

Diagnose why a provider shows generic fallback models instead of its real models.

10|1|Updated Jul 7, 2026
Search Tags:#troubleshooting#diagnose#fallback models#model discovery#openai-compatible#provider configuration#lm studio
One-click install
npx skills add https://github.com/catalystctl/catcode --skill diagnose-provider-fallback-models-catalystctl
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: diagnose-provider-fallback-models
Source: https://github.com/catalystctl/catcode/tree/main/.catalyst-code/skills/diagnose-provider-fallback-models
Command: npx skills add https://github.com/catalystctl/catcode --skill diagnose-provider-fallback-models-catalystctl

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? A configured provider (such as a local LM Studio server) sometimes shows generic fallback models (Umans/Gemini/Codex/xAI) in the model picker instead of its real models. This Skill pinpoints the root cause — unreachable endpoint, auth mismatch, no model loaded, stale cache, or port mismatch — so the real model list appears again. ## Core Features & Use Cases - Live endpoint probing: Uses curl against /v1/models and /v1/models/info to interpret HTTP 000, 401/403, empty data, and healthy responses. - Cache inspection: Checks ~/.config/catalyst-code/models-cache.json for stale fresh entries that short-circuit discovery. - Configuration checks: Verifies base_url port alignment and activeProvider reachability. - Use Case: Your LM Studio provider at http://localhost:1234/v1 only shows Umans fallback models; the Skill guides you to discover the LM Studio server was never started, fix it, and get the real model list without clearing any cache. ## Quick Start Ask the agent to diagnose why my local provider only shows the generic Umans fallback models instead of its actual models.

Frequently Asked Questions about diagnose-provider-fallback-models

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
Why does my LM Studio provider only show generic fallback models?▼

Fallback models appear when model discovery cannot reach or parse the endpoint. For LM Studio, the most common cause is the server not being started — opening the app is not enough; you must click Start Server and load a model or enable JIT loading.

How do I diagnose a provider showing wrong models in the picker?▼

Probe the endpoint with curl against /v1/models and /v1/models/info. HTTP 000 means the server is not running, 401/403 indicates an API key mismatch, and a 2xx with empty data means no model is loaded.

Do I need to clear the models cache to fix fallback models?▼

Usually no. The fallback list is returned uncached, so once the endpoint serves real models the next fetch succeeds. Only delete ~/.config/catalyst-code/models-cache.json if a fresh cached entry holds wrong models.

Why does localhost model discovery fail with connection refused?▼

Connection failures often come from the server binding IPv4-only while localhost resolves to IPv6 ::1, or from a port mismatch. Try 127.0.0.1 instead of localhost and confirm base_url matches the server's actual port, such as 1234 for LM Studio.

Can an unreachable local provider break more than the model list?▼

Yes. If activeProvider points at an unreachable local provider, every turn fails, not just model discovery. Check the active provider configuration first when turns error out.

Related Skills

Explore More Software Engineering Skills→
XiaoyangyXiaoyangy
116

novel-check

Tests LLM provider connectivity with minimal real Generate calls per configured model.

Community
Basic
curtismu7curtismu7
1

agent-tool

Guides delegation of tasks to Claude Code subagents with prompt, model, and batching patterns.

Community
Intermediate
framersaiframersai

healthcheck

Probe endpoints to validate uptime, latency, DNS, and TLS validity.

Official
Advanced
IbIFACE-TechIbIFACE-Tech

provider-integration

Configure and switch between multiple LLM providers like OpenAI, Anthropic, Azure, and Ollama.

Official
Intermediate