AgentSkillsCN

optimizing-gemini-models

每当您或用户编辑`utils/blogData.ts`文件，或对SEO元数据、URL别名等进行修改时，均可使用此技能。它会在部署前自动重新生成站点地图，并确保网站构建流程正确无误。

SKILL.md
--- frontmatter
name: optimizing-gemini-models
description: Monitors and updates the codebase to use the most optimal Gemini model versions for SaaS performance. Use when mentions "check latest models", "update gemini version", or "optimize ai performance".

Optimizing Gemini Models

Ensures the SaaS uses the most efficient, cost-effective, and powerful Gemini models based on the latest available versions and specific use cases.

When to Use This Skill

  • •When a new Gemini model is announced or released.
  • •When existing AI features require optimization (latency, cost, accuracy).
  • •When resolving production errors related to AI model availability (500 errors).

Prerequisites

  • •Read access to the official Gemini model documentation: https://ai.google.dev/gemini-api/docs/models.
  • •Write access to api/ai.ts and associated configuration files.

Workflow

Step 1: Research Latest Model Capabilities

Identify the available models and their specifically optimized use cases.

  • •Reference URL: Gemini Models
  • •Selection Criteria:
    • •Flash models: Best for high-volume, low-latency tasks like transcription and simple chat.
    • •Pro models: Best for complex reasoning, multi-step analysis, and high-precision chat.
    • •Stable tags: Always prefer model names with specific version suffixes (e.g., -001, -002) or the -latest alias if stability is confirmed.

Step 2: Analyze SaaS Requirements

Map the models to specific Diktalo actions in api/ai.ts:

  • •transcription: Accuracy vs. Speed (usually flash).
  • •summary: Context window size vs. Output quality.
  • •chat: Reasoning capabilities vs. Cost.

Step 3: Implement Configuration Update

Update the GEMINI_CONFIG object in api/ai.ts.

  • •Ensure modelPriorities are sorted correctly.
  • •Update preferredModel for each action.

Step 4: Verification & Smoke Test

  • •Build: Run npm run build to ensure no syntax errors.
  • •Test Request: Trigger a small AI request (Summary or Chat) and monitor the terminal logs for the used model name.

Validation

  • •The api/ai.ts file reflects the chosen model versions.
  • •The application successfully processes AI requests without 500 naming errors.
  • •Logs confirm the use of the intended model.

Error Handling

ErrorCauseResolution
404 (Model not found)Using a non-existent or deprecated model name.Cross-check with official docs and use the latest stable name.
429 (Rate Limit)The chosen model has lower quota on the current tier.Downgrade to a different model or implement retry logic.
500 (Internal Error)Temporary API instability or payload mismatch.Check runWithFallback logic and ensure payload follows model specs.

Resources

  • •examples/model-config-template.ts - Reference for a robust configuration setup.
  • •ADVANCED.md - Deep dive into model comparison for SaaS workflows.