fable-prompt-cache

name	fable-prompt-cache
description	Use to execute Claude Fable 5 prompts on Google Cloud Vertex AI using prompt caching breakpoints to minimize token usage costs by up to 90%.

Claude Fable 5 Prompt Caching Workflow

Use this skill when running prompts against Anthropic's Claude Fable 5 model on Google Cloud Vertex AI. This workflow leverages Vertex AI's support for ephemeral prompt caching to optimize token usage.

Prerequisites

1. Vertex AI API Enablement: Ensure the Generative AI / Vertex AI API is enabled in your active project.
2. Quota Allocation: Verify that your project has sufficient quota for:
   • Metric: aiplatform.googleapis.com/online_prediction_input_tokens_per_minute_per_base_model
   • Base Model: anthropic-claude-fable-5
   • Region: us-east5
   • Minimum Quota Target: 100,000 tokens (default of 1,024 is too small since the stable cached system prompt exceeds 1,024 tokens).
3. Authentication: Obtain default credentials:

   gcloud auth application-default login
   

Prompt Caching Implementation

To trigger prompt caching, the cached portion of your prompt (e.g., system instructions or static context) must exceed the minimum size threshold (1,024 tokens).

1. Request Structure

Format the request as an array of content blocks and attach the cache_control parameter to the stable block you want to cache.

Endpoint: POST https://us-east5-aiplatform.googleapis.com/v1/projects/<PROJECT_ID>/locations/us-east5/publishers/anthropic/models/claude-fable-5:rawPredict

Payload Example:

{
  "anthropic_version": "vertex-2024-10-22",
  "messages": [
    {
      "role": "user",
      "content": "Perform the task based on the system instructions."
    }
  ],
  "system": [
    {
      "type": "text",
      "text": "... [Stable instructions (>1,024 tokens)] ...",
      "cache_control": {"type": "ephemeral"}
    }
  ],
  "max_tokens": 4096,
  "stream": false
}

2. Verify Cache Hits

In the API response, inspect the usage block:

• Cache Write: cache_creation_input_tokens > 0 (first run / cache miss).
• Cache Hit: cache_read_input_tokens > 0 and cache_creation_input_tokens == 0 (subsequent runs within the 5-minute TTL).

Executing the Diagnostic/Interactive Script

A pre-packaged script is located in the conversation's scratch directory (scratch/fable_prompt_workflow.py).

• Run a single test query:

  conda run -n goodq_core python scratch/fable_prompt_workflow.py "Hello, verify model status."
  

• Run the interactive chat loop:

  conda run -n goodq_core python scratch/fable_prompt_workflow.py