Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIDeepSeekDeepSeekMuseMuse+22 more

    model routing config auditor

    1

    Audit and patch existing LiteLLM, Portkey, and TensorZero routing configs for broken fallbacks and pricing drift.

    $13

    Secure checkout via Stripe

    Money-back guarantee

    Converts to your local currency at checkout

    Security scanned
    model routing config auditor

    model routing config auditor

    Example session with this skill installed

    Audit this LiteLLM config.yaml. I suspect some fallback chains are broken after we renamed our production models last week, and I want to know if our pricing comments are still accurate.

    • Read your context and instructions
    • Compiled the model routing config

    I have audited the provided LiteLLM config.yaml against the routing schema and the 2026-08-06 pricing snapshot. I found one broken fallback reference and two stale pricing comments.

    Audit Findings

    Route / KeyFindingSeverity
    gpt-4-fallback-chainFallback target gpt-4-0613-legacy does not resolve to any model in model_list.Critical
    ticket-classifierRoute pinned to claude-3-opus for simple label extraction. Over-provisioned.Medium
    summarizer-low-costComment # $0.01/1K tokens is stale. Current rate is $0.015/1K.Low

    Proposed Config Patch

    
    # Diff for fallback repairs and pricing updates
     model_list:
       - model_name: gpt-4-fallback-chain
         litellm_params:
           model: openai/gpt-4
    -      fallbacks: [gpt-4-0613-legacy]
    +      fallbacks: [gpt-4-turbo-preview] # Repointed to active model definition
    
       - model_name: summarizer-low-cost
         litellm_params:
           model: google_search
    -    # $0.01/1K tokens, cheapest reasoning model
    +    # $0.015/1K tokens per 2026-08-06 snapshot
    

    Tier-Swap Proposals

    1. Route ticket-classifier: Propose swap from claude-3-opus to claude-3-5-sonnet. Evidence: Task description indicates high-volume classification which Sonnet handles at lower cost.

    Next steps

    • Confirm acceptance of tier-swap proposal #1 to receive the final diff hunk.
    • Verify the gpt-4-turbo-preview fallback target matches your intended redundancy strategy.
    • Check OpenAI pricing directly for gpt-4-fallback-chain as the snapshot used secondary sources.

    Connects securely to your tools. The creator never sees your data.

    About this skill

    Routing configurations for LiteLLM, Portkey, and TensorZero are often treated as set-and-forget artifacts. Over time, these files drift as provider pricing changes, model tiers shift, and fallback chains break due to renamed identifiers or deprecated endpoints. This skill audits your existing declarative routing configs to identify stale pricing assumptions, broken fallback references, and misaligned model tiers.

    What it does

    • Tier alignment audit flags routes pinned to expensive frontier models for simple tasks like classification or extraction.
    • Fallback verification traces every retry and fallback target to ensure they resolve to valid, active model definitions.
    • Pricing drift detection compares hardcoded costs and thresholds against a dated pricing snapshot for Anthropic, OpenAI, and Google models.
    • Schema validation checks LiteLLM config.yaml, Portkey JSON, and TensorZero toml structures for platform-specific syntax errors.
    • Automated patching generates a diff to repair broken references and update stale pricing comments or cost thresholds.

    How it works

    1. Inventory routes by parsing the config file to map every caller-facing alias to its actual backend model string.
    2. Analyze patterns across three domains: wrong-tier assignments, broken fallback chains, and stale cost logic.
    3. Review findings in a ranked report highlighting production-outage risks first, followed by cost and quality drift.
    4. Approve changes via a per-item acceptance gate for model swaps, ensuring no tier changes happen without explicit confirmation.

    Frameworks & tools

    Supports LiteLLM proxy configurations, Portkey gateway configs, and TensorZero TOML files. Uses a regularly updated internal pricing snapshot for Anthropic, OpenAI, and Google Gemini.

    Why this beats prompting it yourself

    Generic prompts often miss silent no-op fallbacks caused by typoed string references that still look valid. This skill enforces a strict per-item approval workflow for tier swaps, preventing the agent from making autonomous cost-saving changes that might break complex reasoning tasks.

    Use cases

    • Auditing a LiteLLM config after a provider announces major pricing changes or new model releases.
    • Verifying that a complex Portkey fallback chain actually resolves to live models before a production deployment.
    • Identifying over-provisioned routes where legacy frontier models are still used for simple text formatting.
    • Cleaning up stale pricing comments and cost-based routing thresholds that no longer reflect market rates.

    Known limitations

    Does not build new routing logic from scratch or provide general cost-optimization advice based on live traffic data. Pricing for OpenAI models is secondary-sourced and requires manual verification.

    Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for LLM Ops.

    Audits existing configs only; does not design routing schemes. Pricing data requires manual confirmation for OpenAI. Does not instrument application call sites or add budget guards to source code.

    How to use

    1. 1

      Connect your agent to the Agensi MCP

      Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup

    2. 2

      Sign in to Agensi

      Your agent shows a sign-in link. Sign in and click Allow.

    3. 3

      Ask for what you need

      Your agent loads this skill and uses it on your task. It always asks before buying a skill.

    Reviews

    No reviews yet

    Be one of the first to try it. Every listed skill goes through the trust checks below.

    Security scanned

    Checked by our 8-point scan before listing

    Fresh listing

    Recently published to Agensi

    Money-back guarantee

    Doesn't do what its page promises? You get your money back.

    Trust & safety

    Security scan

    Security scan passed 1 month ago

    Permissions

    Environment Variables
    Listed1 month ago

    What's inside

    Frequently Asked Questions

    LiteLLM config.yaml, Portkey gateway JSON and TensorZero tensorzero.toml, or a custom router config. The free audit reports wrong-tier model assignments, broken or missing fallback chains and stale pricing assumptions. The paid tier produces a corrected config as a diff.

    No, the skill employs a per-item acceptance gate that requires user confirmation before applying any model swaps or tier changes, ensuring reasoning quality is preserved.

    It checks the pricing assumptions in your config (comments and cost-based thresholds) against a dated pricing reference that ships with the skill. The reference is a snapshot, not a live feed, and it does not update itself. It carries its own date and review-by date, and OpenAI figures in it are secondary-sourced and need manual confirmation. Treat any price finding as a prompt to verify against the provider's current pricing page.

    While it can fix syntax errors and broken references, it does not analyze live traffic data or suggest new routing logic; it is designed to maintain and repair existing configurations.

    The auditor generates a standard diff file, allowing you to review all proposed repairs and pricing updates before merging them into your production configuration.

    New in LLM Ops