LiteLLM

Call 100+ LLM providers through LiteLLM's unified API. Use when you need to call a different model than your primary (e.g., use GPT-4 for code review while running on Claude), compare outputs from multiple models, route to cheaper models for simple tasks, or access models your runtime doesn't natively support.

Security Scan

The skill's behavior (calling many external LLM providers) matches its description, but there are inconsistencies and small risks you should understand before installing — notably undeclared API key usage and an unpinned pip install instruction.

Purpose & Capability

The skill's code and SKILL.md implement a generic LiteLLM client for calling many LLM providers, which is coherent with the name/description. However, the registry metadata declares no required environment variables or primary credential while the documentation and code clearly expect API keys (e.g., LITELLM_API_KEY, OPENAI_API_KEY, ANTHROPIC_API_KEY or a proxy key). This is an inconsistency (likely sloppy metadata) but not itself evidence of malice.

Instruction Scope

Runtime instructions and the included script stay within the defined purpose: build messages and call litellm.completion (optionally via a proxy). The instructions do not ask the agent to read unrelated files or system secrets. They do instruct users to set provider API keys and an optional proxy endpoint, which will cause prompts to be transmitted to external LLM providers (expected for this skill).

Install Mechanism

There is no formal install spec in the registry (instruction-only), and SKILL.md suggests 'pip install litellm' with no version or source verification. Installing an unpinned package from PyPI is common but increases risk; you should verify the package origin and integrity (official project, expected maintainer) before installing.

Credentials

Although provider API keys are appropriate for a multi-provider LLM caller, the registry metadata lists no required env vars while SKILL.md and the script reference several sensitive variables (LITELLM_API_BASE, LITELLM_API_KEY, OPENAI_API_KEY, ANTHROPIC_API_KEY). This mismatch means the skill may access or require sensitive credentials that weren't declared up front. Also note that any prompts and related data will be sent to external services tied to those keys — consider data-sensitivity implications.

Persistence & Privilege

The skill is not always-enabled, doesn't request elevated or persistent system privileges, and does not modify other skills' configs. Autonomous invocation is allowed by default but is not combined here with other high-risk properties.

What to consider before installing

This skill is coherent with its stated purpose (calling many LLMs), but take these precautions before installing: - Verify the litellm package source/version on PyPI or the project's official docs before running 'pip install'. Prefer pinned versions or vetted packages. - Expect to provide API keys (OpenAI, Anthropic, or a LiteLLM proxy key). The registry omitted these declarations — confirm which keys you will supply and where they are stored. - Be aware that any prompt you send may be transmitted to third-party providers. Do not send sensitive secrets, credentials, or private data to this skill unless you trust the destination. - Consider deploying and pointing the skill to a trusted LiteLLM proxy (as suggested) to centralize auditing, caching, and to avoid sprinkling provider keys across environments. - If you need higher assurance, ask the publisher for: an authoritative homepage or repo, package signing or checksum, and explicit declared required env vars in the registry metadata.

LiteLLM - Multi-Model LLM Calls

Use LiteLLM when you need to call LLMs beyond your primary model.

When to Use

• Model comparison: Get outputs from multiple models and compare
• Specialized routing: Use code-optimized models for code, writing models for prose
• Cost optimization: Route simple queries to cheaper models
• Fallback access: Access models your runtime doesn't support

Quick Start

import litellm

# Call any model with unified API
response = litellm.completion(
    model="gpt-4o",
    messages=[{"role": "user", "content": "Explain this code"}]
)
print(response.choices[0].message.content)

Common Patterns

Compare Multiple Models

import litellm

prompt = [{"role": "user", "content": "What's the best approach to X?"}]

models = ["gpt-4o", "claude-sonnet-4-20250514", "gemini/gemini-1.5-pro"]
for model in models:
    resp = litellm.completion(model=model, messages=prompt)
    print(f"{model}: {resp.choices[0].message.content[:200]}...")

Route by Task Type

import litellm

def smart_call(task_type: str, prompt: str) -> str:
    model_map = {
        "code": "gpt-4o",           # Strong at code
        "writing": "claude-sonnet-4-20250514",  # Strong at prose
        "simple": "gpt-4o-mini",    # Cheap for simple tasks
        "reasoning": "o1-preview",  # Deep reasoning
    }
    model = model_map.get(task_type, "gpt-4o")
    resp = litellm.completion(
        model=model,
        messages=[{"role": "user", "content": prompt}]
    )
    return resp.choices[0].message.content

Use LiteLLM Proxy (Recommended)

If a LiteLLM proxy is available, point to it for caching, rate limiting, and observability:

import litellm

litellm.api_base = "https://your-litellm-proxy.com"
litellm.api_key = "sk-your-key"

response = litellm.completion(
    model="gpt-4o",  # Proxy routes to configured provider
    messages=[{"role": "user", "content": "Hello"}]
)

Environment Setup

Ensure litellm is installed and API keys are set:

pip install litellm

# Set provider keys (or configure in proxy)
export OPENAI_API_KEY="sk-..."
export ANTHROPIC_API_KEY="sk-..."

Model Reference

Common model identifiers:

• OpenAI: gpt-4o, gpt-4o-mini, o1-preview, o1-mini
• Anthropic: claude-sonnet-4-20250514, claude-opus-4-20250514
• Google: gemini/gemini-1.5-pro, gemini/gemini-1.5-flash
• Mistral: mistral/mistral-large-latest

Full list: https://docs.litellm.ai/docs/providers