mukt_skill

Overview

This skill is a zero-cost OpenRouter responder that auto-discovers the best currently free model, retries on failures, and returns clean, fast answers. It routes queries to the lowest-cost available OpenRouter model and falls back automatically when a candidate fails. Use it when you need inexpensive casual responses without exposing secrets or running heavy pipelines.

How this skill works

The skill queries OpenRouter for currently free model candidates, ranks them, and attempts them in order up to a configurable max. For each attempt it invokes the model with the supplied prompt and optional system instruction, captures the assistant text, and returns the first successful response. It prints a single JSON object containing selected_model, response, attempted_models, and free_model_candidates; if all attempts fail it exits non-zero and emits an error.

When to use it

• You want a free-model OpenRouter response for low-risk tasks.
• You need a quick draft, brainstorming, or second opinion without spending budget.
• You want automatic fallback across free models when calls fail.
• You need a lightweight, fast responder for non-sensitive content.
• You want to integrate cost-aware routing into scripts or automation.

Best practices

• Never provide secrets, credentials, or private data to free-model calls.
• Supply clear prompts and optional concise system instructions to improve quality.
• Set a reasonable --max-attempts to balance latency and success rate.
• Use low sampling temperature (default 0.3) for deterministic, cleaner answers.
• Enable --debug only during troubleshooting; it logs model ranking and fallbacks to stderr.

Example use cases

• Generate quick marketing copy drafts or content outlines without spending credits.
• Ask for coding examples or conceptual explanations where precision can be validated later.
• Produce alternative phrasings or brainstorming ideas for editing by a primary agent.
• Run casual Q&A or learning prompts for demonstrations and trials.