credit-optimizer 精选

自动优化 Manus AI 额度消耗，降低使用成本。原文：Automatically optimize AI agent credit usage by routing tasks to the most cost-efficient execution path. Use when you want to reduce AI API costs by 30-75% without quality loss, classify task complexity before execution, route simple tasks to free or low-cost models, split complex tasks into optimized sub-tasks, or detect vague prompts before wasting credits.

仓库级简介：该 skill 尚无单独撰写的中文摘要

未分类 Codex CLI MCP 客户端 v5.2.0 MIT 51

在 GitHub 打开

安装方式

https://raw.githubusercontent.com/rafsilva85/credit-optimizer-v5/37eb57a8fb2506e8e1c1c5458748bac439c2f521/SKILL.md

下载后另存为 ~/.claude/skills/credit-optimizer/SKILL.md。链接锁定采集时的 commit，内容与本站展示一致。

整仓一次性获取：git clone --depth 1 https://github.com/rafsilva85/credit-optimizer-v5.git

依赖与权限

无需 API Key
需要联网访问外部服务
依赖外部命令：python
以上为按正文与 frontmatter 自动推断的信号，仅供参考；安装前建议先阅读原文。

SKILL.md 正文

来源 rafsilva85/credit-optimizer-v5@37eb57a · 许可 MIT · 内容未修改

Credit Optimizer v5

Automatically optimize AI agent credit/token usage by routing tasks to the most cost-efficient execution path — with zero quality loss.

Audited across 53 real-world scenarios. 30-75% cost savings. 0% quality degradation.

When to Use This Skill

Before executing any AI task that consumes credits or tokens
When you want to minimize API costs without sacrificing output quality
When processing batches of tasks with varying complexity
When you need to decide between different model tiers (free/standard/premium)

How It Works

Step 1: Task Classification

Analyze the incoming task and classify it into one of these categories:

Category	Examples	Typical Savings
Simple Q&A	Definitions, facts, conversions	90-100% (use free tier)
Code Generation	Scripts, functions, refactoring	40-60%
Research	Multi-source analysis, synthesis	20-40%
Creative Writing	Articles, stories, marketing copy	30-50%
Data Analysis	CSV processing, visualization	40-70%
Complex Reasoning	Multi-step logic, architecture	10-20%

Step 2: Prompt Quality Check

Before executing, evaluate the prompt:

Clarity Score (1-10): Is the request specific enough?

Score < 5: Ask for clarification BEFORE executing (saves wasted credits)
Score 5-7: Add reasonable assumptions and proceed
Score 8+: Execute directly

Scope Detection: Can this be split into smaller, cheaper sub-tasks?

If YES: Break into atomic tasks, route each independently
If NO: Route as single task

Data Requirement Check: Does this need real-time data?

If YES: Use tools/search first, then process with cheaper model
If NO: Use internal knowledge with appropriate model tier

Step 3: Model Routing

Route to the optimal execution path:

IF task is simple Q&A or formatting:
  → Use FREE tier / Chat mode (no credits)
  
IF task is medium complexity (code, writing, basic analysis):
  → Use STANDARD tier
  
IF task requires deep reasoning, multi-step logic, or creative excellence:
  → Use PREMIUM/MAX tier
  
IF task is mixed complexity:
  → SPLIT into sub-tasks and route each independently

Step 4: Execution Optimization

During execution, apply these optimizations:

Context Pruning: Only include relevant context, not entire conversation history
Output Scoping: Request specific output format to avoid verbose responses
Caching: Check if similar tasks were recently completed
Batch Processing: Group similar sub-tasks for efficient processing

Efficiency Directives

Never use premium models for tasks that standard can handle equally well
Always check if the task can be answered from cached/known information first
Split compound requests into atomic tasks before routing
Ask for clarification on vague prompts — it's cheaper than re-doing work
Use structured output formats to reduce token waste