Decision Economic Optimizer

核心用法

Decision Economic Optimizer 是 Which-LLM 提供的确定性决策优化服务，通过 REST API 帮助用户在预算和质量约束下选择最优 LLM 模型。用户向 /decision/optimize 端点发送自然语言目标和约束条件（如最小质量分数、最大成本），API 返回推荐模型及有序回退计划。该服务采用 HTTP 402 支付模式，每次请求需支付 0.01 USDC，用户可通过向 /decision/outcome 报告实际执行结果获得积分令牌，用于抵扣后续请求费用（通常可获 50% 折扣）。支持 Base、Ethereum、Arbitrum、Optimism、Avalanche 等多条 EVM 链支付。

显著优点

首先，算法具有确定性，在相同输入下始终返回一致结果，确保决策可重现。其次，经济模型设计精巧，微额支付（$0.01/次）降低了使用门槛，而结果反馈机制形成的积分体系有效激励用户参与数据回流，形成正向循环。第三，安全架构严谨，强制要求使用专用隔离钱包（非主钱包），并通过多源地址验证（well-known 端点、API 文档、ENS 记录）防止地址篡改攻击。第四，提供完整的回退计划（fallback plan），当首选模型失败时自动提供次优选择，增强系统韧性。最后，透明度高，所有支付记录在链上可审计，API 返回详细的可解释性评分（explainability score）帮助理解决策逻辑。

潜在缺点与局限性

该 Skill 要求用户向 Agent 提供钱包私钥（或助记词、密钥库文件），尽管通过"专用钱包+限制资金"（建议 $2-10 USDC）模式缓解风险，但仍存在私钥泄露的理论可能。此外，自主支付设计意味着 Agent 可在无需逐笔人工确认的情况下自动签名交易，虽然单次金额极小，但若未正确配置钱包隔离，仍存在资金损失风险。加密货币支付本身也构成了使用门槛，不熟悉区块链操作的用户需要学习曲线。作为相对较新的项目（v0.1.0），其长期安全记录和 API 稳定性尚待时间验证。最后，积分存在有效期（90 天）和衰减机制（30 天后 50% 衰减），频繁使用者需留意积分管理。

适合的目标群体

主要面向 AI 应用开发者、LLM 基础设施工程师、以及需要批量优化模型选择成本的企业技术团队。特别适合构建 LLM 代理（Agent）系统的开发者，需要在代码执行、文本摘要、实体提取等不同任务类型中动态选择性价比最优的模型。对于关注可审计性和透明度的金融科技场景，以及希望利用微支付实现精准成本控制的 SaaS 产品团队也具有吸引力。不适合完全不接受加密货币支付，或无法承担私钥管理责任的普通终端用户。

使用风险

除前述私钥管理风险外，还包括区块链网络拥堵导致的交易确认延迟（可能影响 API 响应时间）、gas 费波动（尽管 Base 等 L2 网络费用较低）、以及智能合约或 API 端的潜在漏洞。用户必须严格遵循"多源验证"流程确认支付地址，否则可能遭遇地址投毒攻击。建议定期通过区块链浏览器（如 Basescan）监控交易记录，设置钱包余额变动通知，并每月轮换专用钱包以降低长期风险敞口。

Overview

Decision Economic Optimizer is a specialized API service for programmatic LLM model selection under budget and quality constraints. It provides deterministic recommendations with fallback plans for resilience, using a novel pay-per-use model with outcome-driven discounts.

Core Functionality

The skill operates through two primary endpoints:

• POST /decision/optimize: Returns a recommended LLM model based on natural language goals, cost constraints, and quality thresholds
• POST /decision/outcome: Reports execution results to earn credit tokens (~50% refund on future requests)

The pricing model uses HTTP 402 Payment Required responses, requiring exact USDC transfers ($0.01 per request) verified on-chain across multiple EVM networks (Base, Ethereum, Arbitrum, Optimism, Avalanche).

Notable Advantages

1. Deterministic, auditable decisions: Each recommendation includes a decision hash and scoring breakdown, enabling reproducibility and debugging
2. Fallback planning: Automatically generates ordered backup options when primary choices fail
3. Outcome-driven economics: Reporting actual execution results earns credits that reduce future costs, creating incentive alignment
4. Multi-chain support: Flexible payment across 5 major L1/L2 networks
5. Comprehensive security documentation: Extensive guidance on wallet isolation, address verification, and risk mitigation

Significant Limitations & Risks

1. Credential sensitivity: Requires WALLET_PRIVATE_KEY or equivalent, creating irreversible exposure if mishandled
2. Autonomous payment authority: No per-transaction approval; agent can spend up to wallet balance without confirmation
3. T3 source credibility: Personal developer (zapkid) without organizational backing or established track record
4. Fixed cost per request: $0.01 may be expensive for high-volume use compared to aggregated API pricing
5. No code auditability: Pure Markdown skill with no executable code to inspect (T-MD classification)

Suitable Users

• AI infrastructure teams building cost-optimization layers over multiple LLM providers
• Developers needing deterministic, explainable model selection with fallback resilience
• Organizations comfortable with cryptocurrency micropayments and self-custody wallet management
• Experimenters willing to accept T3 risk level for novel economic mechanisms

Not suitable for: Users requiring enterprise SLAs, those uncomfortable with crypto payments, or high-volume applications where $0.01/request exceeds alternative pricing.

Risk Assessment

| Factor | Level | Notes |

|--------|-------|-------|

| Financial exposure | Medium | Bounded by dedicated wallet limits, but irreversible if compromised |

| Source trust | Medium | T3 developer, no organizational backing |

| Operational complexity | Medium | Requires EVM wallet management, gas tokens, multi-source verification |

| Privacy | Low | Minimal PII exposure; payment addresses are public |

The skill implements mandatory multi-source verification (.well-known endpoint, API documentation, ENS) for payment addresses, mitigating a critical attack vector. However, the fundamental trade-off remains: convenience of autonomous operation versus trust in the agent with signing capability.