360 Secure SkillHub
360 安全技能中心

SQL Guard Copilot

通过命令行脚本检查表结构、把自然语言转为 SQL、执行只读查询，并提供 Lint、EXPLAIN、慢查询提示、结果列概要与可选审计日志；默认只读并拒绝破坏性语句。

gausszhu
gausszhu
热度 82.56
最近更新 2026-05-11 17:30:32
分类：数据依赖：代码执行 / 读写文件 / 第三方 API / 内部调用大模型应用场景：数据处理 / 代码生成 / 工作流

安装技能

复制下面的话术发给你的 Agent，它会通过 secure-skills 检索并安装该技能。

执行命令安装此 skill： npx secure-skills-cli add sql-guard-copilot-zhu
Metadata
name
sql-query-copilot
description
Simplify SQL querying and troubleshooting for MySQL, PostgreSQL, and SQLite. Use when users ask to inspect schema, convert natural language to SQL, debug SQL errors, run explain plans, lint risky SQL, or validate data with safe read-only execution.

SQL Query Copilot

Overview

Use this skill to turn plain-language requests into executable SQL with a predictable, low-risk workflow. Default to read-only execution and validate every query against schema before running.

Quick Start

Set SQL_DSN first (or pass --dsn each time).

# PowerShell
$env:SQL_DSN="mysql://user:password@127.0.0.1:3306/stock_monitor"
$env:SQL_DSN="postgres://user:password@127.0.0.1:5432/stock_monitor"
$env:SQL_DSN="sqlite:///d:/data/demo.db"

# Windows CMD
set SQL_DSN=mysql://user:password@127.0.0.1:3306/stock_monitor
set SQL_DSN=postgres://user:password@127.0.0.1:5432/stock_monitor
set SQL_DSN=sqlite:///d:/data/demo.db

# Bash / Zsh
export SQL_DSN="mysql://user:password@127.0.0.1:3306/stock_monitor"
export SQL_DSN="postgres://user:password@127.0.0.1:5432/stock_monitor"
export SQL_DSN="sqlite:///d:/data/demo.db"

Core commands:

python scripts/sql_easy.py tables
python scripts/sql_easy.py describe daily_kline
python scripts/sql_easy.py lint --sql "SELECT * FROM daily_kline"
python scripts/sql_easy.py explain --sql "SELECT code, close FROM daily_kline WHERE trade_date >= '2026-01-01'"
python scripts/sql_easy.py query --sql "SELECT code, close FROM daily_kline ORDER BY trade_date DESC" --limit 50
python scripts/sql_easy.py query --sql "SELECT code, close FROM daily_kline" --summary
python scripts/sql_easy.py ask --q "show symbols with old sell signals older than 20 days" --summary
python scripts/sql_easy.py profile

Set OPENAI_API_KEY (or pass --api-key) to use ask.

v0.2 Highlights

  • Multi-engine support: MySQL, PostgreSQL, SQLite.
  • SQL lint engine: catches high-risk patterns before execution.
  • Explain mode: quickly inspect query plan (EXPLAIN / EXPLAIN QUERY PLAN).
  • Natural-language mode: ask generates SQL from user intent.
  • Query summary: auto profile returned columns (null ratio, distinct count, min/max/avg).
  • Slow query warning: highlights expensive queries using --slow-ms.
  • Audit log: write command metadata to JSONL via --audit-log or SQL_EASY_AUDIT_LOG.

Workflow

  1. Clarify the metric and grain. Ask for time window, dimensions, and output columns before writing SQL.

  2. Discover schema first. Run tables, describe <table>, and profile before any complex SQL.

  3. Draft SQL in read-only mode. Use SELECT or WITH; keep columns explicit and add time filters.

  4. Execute with guardrails. Run via scripts/sql_easy.py query, keep --limit unless full export is explicitly needed.

  5. Validate results. Cross-check row count, null ratio, and edge dates; adjust query and rerun.

Guardrails

  • Default to read-only SQL.
  • Reject destructive statements (INSERT, UPDATE, DELETE, DROP, ALTER, TRUNCATE, etc.).
  • Prefer explicit columns over SELECT * for production/report queries.
  • Run lint before heavy or scheduled queries.
  • Run explain before approving complex joins/window queries.
  • Always quote identifiers when table/column names are uncertain.
  • For business decisions, provide both SQL and a short interpretation of returned data.

Query Patterns

Read references/query_patterns.md when creating:

  • Top-N and ranking queries
  • Time-window aggregation
  • Dedup with window functions
  • Funnel-style conditional counts
  • Data quality checks (null/duplicate/outlier)

Read references/chanquant_templates.md for Chanquant-specific query templates.