Is tmux required to run the tests?

Yes, tmux is required for non-interactive smoke tests that involve Terminal User Interface (TUI) components to simulate terminal environments.

Will running these tests incur API costs?

Yes, these tests make real API calls to your configured providers to ensure actual inference and tool-calling logic are functional.

How do I troubleshoot failed TUI submissions?

The skill includes guidance on capturing logs to verify EventPartialCompletionStart and EventFinal events, and notes that some TUIs require 'Tab' for submission.

What LLM providers are supported by these smoke tests?

The skill currently supports testing for OpenAI (including specialized 'thinking' responses) and Anthropic Claude for tool-calling and inference flows.

When should I use this skill?

Use it when refactoring core inference state, session management, engine building logic, or when adding support for new provider request formats.

Inference Smoke Tests

Name: Inference Smoke Tests
Author: go-go-golems

bygo-go-golems

•

Security & Testing

Executes automated and manual verification tests to ensure LLM inference and tool-calling logic are functioning correctly within the Geppetto ecosystem.

The Inference Smoke Tests skill provides a structured framework for verifying the health of LLM integrations within the Geppetto and Pinocchio ecosystems. It enables developers to run repeatable tests covering single-pass inference, streaming responses, complex tool-calling loops, and Bubble Tea TUI event flows. This skill is particularly valuable during refactors of core components like InferenceState or EngineBuilder, ensuring that changes to provider request formatting or event sinks do not break critical chat and agent functionalities across different providers like OpenAI and Anthropic.

Key Features

01Specialized tmux-driven testing for non-interactive Bubble Tea TUI validation

02Comprehensive coverage for OpenAI 'thinking' and Claude tool-calling logic

03Automated shell scripts for quick 'does-it-work' inference validation

0483 GitHub stars

05Validation of multi-turn chat state persistence and session management

06Detailed troubleshooting playbooks for API 400 errors and TUI failures

Use Cases

01Ensuring TUI event deltas are correctly rendered during refactors of the event sink system

02Performing a quick connectivity and credential health check for OpenAI and Claude APIs

03Verifying provider request formatting after updating LLM SDKs or API versions

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add go-go-golems/geppetto inference-smoke-tests

For use in Claude.ai and ChatGPT

Download Skill