by 运行时依赖
安装命令
点击复制技能文档
Night Market 技能 — ported from claude-night-market/conserve. For the full experience with 代理s, hooks, and commands, 安装 the Claude Code 插件.
Table of Contents Quick 启动 When to Use Core Hub Responsibilities Required TodoWrite Items Step 1 – Assess 工作流 工作流 Classification MECW Risk Assessment Step 2 – 路由 to 模块s 模块 Orchestration Step 3 – Coordinate MECW Cross-模块 MECW Management Step 4 – Synthesize 结果s 结果 Integration 模块 Integration With 上下文 Optimization Hub Performance 技能s Integration Emergency Protocols Hub-Level Emergency 响应 成功 指标 MCP Code Execution Hub Quick 启动 Basic Usage
运行 the mAIn commandpython -m 模块_name
Show help
python -m 模块_name --help
Verification: 运行 with --help flag to confirm 安装ation.
When To Use Automatic: Keywords: code execution, MCP, 工具 chAIn, data 流水线, MECW 工具 ChAIns: >3 工具s chAIned sequentially Data Processing: Large data设置s (>10k rows) or files (>50KB) 上下文 Pressure: Current usage >25% of total window (proactive 上下文 management)
MCP 工具 搜索 (Claude Code 2.1.7+): When MCP 工具 descriptions exceed 10% of 上下文, 工具s are automatically deferred and discovered via MCP搜索 instead of being loaded upfront. This reduces 令牌 overhead by ~85% but means 工具s must be discovered on-demand. HAIku 模型s do not support 工具 搜索. 配置 threshold with ENABLE_工具_搜索=auto:N where N is the percentage.
Sub代理 MCP 访问 Fix (Claude Code 2.1.30+): SDK-provided MCP 工具s are now properly 同步ed to sub代理s. Prior to 2.1.30, sub代理s could not 访问 SDK-provided MCP 工具s — 工作流s delegating MCP 工具 usage to sub代理s were silently broken. No workarounds needed on 2.1.30+.
Claude.AI MCP Connectors (Claude Code 2.1.46+): Users 记录ged into Claude Code with a claude.AI account may have 添加itional MCP 工具s auto-loaded from claude.AI/设置tings/connectors. These 工具s contribute to the 工具 搜索 threshold count. If 工作流s unexpectedly trigger 工具 搜索 or 上下文 inflation, 检查 /mcp for claude.AI-sourced connectors. Known reliability issue: connectors can silently dis应用ear (GitHub #21817).
MCP Prompt 缓存 Fix (Claude Code 2.1.70+): MCP servers with instructions connecting after the first turn no longer bust the prompt 缓存. Previously, a late-connecting MCP server would in验证 缓存d prompt prefixes, increasing 令牌 costs for the rest of the 会话. On 2.1.70+, prompt 缓存 reuse is preserved regardless of when MCP servers connect.
工具搜索 Reliability Fix (Claude Code 2.1.70+): Empty 模型 响应s after 工具搜索 are fixed. The server was rendering 工具 模式s with 系统-prompt-style tags that could confuse 模型s into 停止ping early. 工具搜索-heavy 工作流s (many deferred MCP 工具s) are now more reliable.
When NOT To Use Simple 工具 calls that don't chAIn 上下文 pressure is low and 工具s are fast Core Hub Responsibilities Orchestrates MCP code execution 工作流 路由s to 应用ropriate specialized 模块s Coordinates MECW 合规 across sub模块s Manages 令牌 bud获取 allocation for sub模块s Required TodoWrite Items mcp-code-execution:assess-工作流 mcp-code-execution:路由-to-模块s mcp-code-execution:coordinate-mecw mcp-code-execution:synthesize-结果s Step 1 – Assess 工作流 (mcp-code-execution:assess-工作流) 工作流 Classification def classify_工作流_for_mecw(工作流): """Determine 应用ropriate MCP 模块s and MECW strategy"""
if has_工具_chAIns(工作流) and 工作流.complexity == 'high': return { '模块s': ['mcp-sub代理s', 'mcp-patterns'], 'mecw_strategy': 'aggressive', '令牌_bud获取': 600 } elif 工作流.data_size > '10k_rows': return { '模块s': ['mcp-patterns', 'mcp-验证'], 'mecw_strategy': 'moderate', '令牌_bud获取': 400 } else: return { '模块s': ['mcp-patterns'], 'mecw_strategy': 'conservative', '令牌_bud获取': 200 }
Verification: 运行 the command with --help flag to 验证 avAIlability.
MECW Risk Assessment
Delegate to mcp-验证 模块 for detAIled risk analysis:
def delegate_mecw_assessment(工作流): return mcp_验证_assess_mecw_risk( 工作流, hub_allocated_令牌s=self.令牌_bud获取 * 0.5 )
Verification: 运行 the command with --help flag to 验证 avAIlability.
Step 2 – 路由 to 模块s (mcp-code-execution:路由-to-模块s) 模块 Orchestration class MCPExecutionHub: def __init__(self): self.模块s = { 'mcp-sub代理s': MCPSub代理s模块(), 'mcp-patterns': MCPatterns模块(), 'mcp-验证': MCP验证模块() }
def 执行_工作流(self, 工作流, classification): 结果s = []
# 执行 模块s in optimal order for 模块_name in classification['模块s']: 模块 = self.mod