Token导航 LogoToken导航TokenDH.com
运维权限需确认clawhub未标认证来源可访问clear审计通过

sql-optimizationSQL optimization 测试

Agent Skill

用于辅助数据库表结构、查询语句、迁移脚本和数据维护任务。它适合让 Agent 分析 schema、编写 SQL、排查查询问题、整理索引或生成迁移建议。使用时需要明确数据库类型、连接环境和目标表,区分只读分析与写入变更;涉及删除、更新、迁移和批量导入时,应优先 dry-run、备份或事务保护,避免误操作。

总安装

6,650

周安装

269

GitHub Stars

公开资料未说明

下载量

2,087
OpenClaw

安装说明

本站只整理中文说明和来源信息,不托管安装包,也不代用户安装。

GitHub

来源数

2

许可证

MIT-0

最后核验

2026-05-01

来源状态

来源可访问

安装方式

通过对话安装

复制提示词发给支持本地命令或 Skills 的 AI 助手,先确认命令和权限,再让它执行。

请帮我安装这个 Agent Skill:sql-optimization(SQL optimization 测试)
来源仓库:https://github.com/codekungfu/sql-optimization
安装命令:
openclaw skills install sql-optimization
安装前请先检查当前环境是否支持对应 CLI,并向我确认将要执行的命令、安装目录、联网范围和文件读写权限;确认后再执行。

命令行安装

复制命令到本机终端执行。该命令会通过 OpenClaw 从第三方来源获取 Skill;本站只展示命令,不托管安装包,也不自动执行。

ClawHubOpenClaw
openclaw skills install sql-optimization

简介

sql-optimization 辅助数据库查询分析与性能调优,支持索引策略建议。

  • 适合让 Agent 编写 SQL、排查慢查询或生成迁移脚本时使用。
  • 通过 clawhub 安装,需明确数据库类型与连接环境后再执行操作。
  • 涉及数据变更时应优先 dry-run 或事务保护,避免误删误改。
  • 适用于 OLTP 或数据分析系统中的 SQL 性能瓶颈处理。

SKILL.md

name
sql-optimization
description
Deep SQL performance workflow—symptom framing, execution plans, indexing strategy, query rewrite, locking/transaction behavior, statistics, partitioning, and verification. Use when queries time out, DB CPU spikes, or migrations change access patterns.

SQL Optimization (Deep Workflow)

Optimization without measurement is guesswork. Structure the work as observe → explain (plan) → change → verify, with explicit attention to correctness, locks, and write amplification from indexes.

When to Offer This Workflow

Trigger conditions:

  • Slow queries, growing P95/P99, replication lag, lock waits
  • ORM-generated SQL surprises; N+1 at DB layer
  • Index explosion, bloat, or “we added indexes everywhere”

Initial offer:

Use six stages: (1) frame the problem, (2) reproduce & measure, (3) read execution plans, (4) schema & indexes, (5) query & transaction tuning, (6) verify & guardrail. Confirm engine (PostgreSQL, MySQL, SQL Server, etc.) and environment (prod-like data volume).


Stage 1: Frame the Problem

Goal: Define SLO, scope, and non-goals.

Questions

  1. Which queries or endpoints are slow? User-facing vs batch?
  2. Regression—did deploy, data volume, or stats change?
  3. Isolation level and consistency requirements—can we read replicas?
  4. Write risk: is this table write-heavy? Index cost?

Exit condition: One-line problem statement with metric (e.g., “p95 2.4s on /reports at 10k RPS”).


Stage 2: Reproduce & Measure

Goal: Stable repro with representative cardinality and parameters.

Actions

  • Capture exact SQL, parameters, and frequency
  • Use EXPLAIN (ANALYZE, BUFFERS) or equivalent—engine-specific
  • Check buffer cache effects: cold vs warm cache; run twice when needed
  • Compare prod stats vs staging—row counts, histograms

Pitfalls

  • Optimizing on empty dev DB
  • Different parameter sniffing values changing plan choice

Exit condition: Baseline numbers + plan hash or saved plan for A/B.


Stage 3: Read Execution Plans

Goal: Name the dominant cost: seq scan, bad join order, sort, hash spill, nested loop explosion.

Interpret (adapt to engine)

  • Seq scan on large tables—filter selectivity? missing index? stats?
  • Index scan vs bitmap vs index only—covering indexes trade-offs
  • Joins: wrong order, missing stats, outdated NDV
  • Sort/hash spills to disk—work_mem / memory grants
  • Locks: FOR UPDATE, long transactions, hot row updates

Exit condition: Hypothesis tied to plan node(s), not generic “add index.”


Stage 4: Schema & Indexes

Goal: Right indexes for read paths without destroying writes.

Strategy

  • Composite index column order: equality → range; avoid redundant indexes
  • Partial indexes for hot subsets
  • Covering indexes vs table bloat—measure write cost
  • Foreign keys and constraints affecting plans
  • Statistics: ANALYZE, extended stats, histograms—when stale stats lie

Advanced (when relevant)

  • Partitioning for prune + maintenance
  • Materialized views / pre-aggregation for heavy reports

Exit condition: DDL proposal with rationale and rollback (drop index concurrently if supported).


Stage 5: Query & Transaction Tuning

Goal: Sometimes the fix is SQL rewrite, not hardware.

Techniques

  • Reduce rows touched early (CTEs vs inline—engine-dependent)
  • Pagination without OFFSET on huge pages (keyset)
  • Batch vs row-by-row; UNION ALL vs OR
  • N+1: batch queries, joins, data loader patterns
  • Transactions: shorten locks; avoid unnecessary SELECT FOR UPDATE
  • ORM: eager vs lazy loading discipline

Exit condition: New plan shows lower cost / measured latency; lock time acceptable.


Stage 6: Verify & Guardrail

Goal: Improvement holds under load and doesn’t regress neighbors.

Verify

  • Re-run EXPLAIN ANALYZE with production-like parameters
  • Load test or shadow traffic if available
  • Monitor: buffer hit ratio, index bloat, replication lag

Guardrails

  • Query timeouts and statement_timeout where safe
  • Alerts on sequential scans on large tables if observability supports

Final Review Checklist

  • [ ] Baseline and target metrics documented
  • [ ] Plan-based root cause, not guesswork
  • [ ] Index/DDL changes justified vs write load
  • [ ] Transaction/lock behavior considered
  • [ ] Verification on realistic data and load

Tips for Effective Guidance

  • Always mention parameter sniffing and stale statistics as frequent culprits.
  • Warn when adding indexes on very write-heavy tables without measuring bloat.
  • Prefer keyset pagination education for large lists.

Handling Deviations

  • No EXPLAIN access: infer from symptoms + ORM logs + index list; recommend safe staging repro.
  • Vendor DB: name that hints and features differ—avoid PostgreSQL-only advice on SQL Server without caveat.

适合场景

01

OpenClaw 用户查找和安装 Skill 时

02

用户想查找某类 Agent Skill 时

03

需要根据任务场景推荐可安装能力包时

04

需要对比不同来源的安装命令和来源信息时

能力概览

能力 1

按任务关键词查找相关 Skills

能力 2

展示可复制的安装命令

能力 3

保留来源站点、仓库和原始说明,方便继续核验

能力 4

补充不同宿主或平台的使用分布数据

能力 5

展示第三方安全扫描或审计结果

安装后应在对应宿主中按原始 README 的触发条件使用;具体调用方式请以来源页面和 README 为准。

平台分布

OpenClaw

81.52%
按下载量换算1,701

安全审计

VirusTotal

通过

ClawScan

通过

Static analysis

通过

权限和风险

权限需确认

当前来源未能明确判断权限范围,默认进入异常复核队列。

安装前确认

本站仅展示第三方公开信息,不托管安装包,不提供自动安装或运行环境。安装前应自行审查源码、依赖和命令行为。当前只有一个来源,正式发布前建议补源仓库或其他目录站核验。

来源信息

继续浏览同类 Skills