How do I install the Odbc Connector skill?

Run npx skillmds@latest add stas2-git/odbc-connector in your terminal (requires Node.js), paste this page's agent-chat prompt into Claude, Cursor, or any MCP-connected agent, or download the SKILL.md file and copy it into your agent's skills directory.

Is Odbc Connector safe to use?

This skill has not completed SkillMD's automated safety review yet. Capability flags: executes scripts. SkillMD never runs a skill's scripts for you; review the SKILL.md before installing.

Which AI agents work with Odbc Connector?

This skill is tagged as working with Claude Code, Claude.ai, OpenAI Codex. SKILL.md is an open format, so most agents that read a skills directory can load it too.

Is Odbc Connector free to use?

Yes. Installing skills from SkillMD is free, and the skill stays under its author's original license.

Who published Odbc Connector?

stas2-git (@stas2-git) published this skill. Their other Agent Skills are listed on their SkillMD profile.

Odbc Connector by stas2-git — Agent Skill

Name: Odbc Connector
Rating: 4.2 (5 reviews)
Author: stas2-git

Odbc Connector

Use this skill when the user wants to connect to a database, explore its schema, or execute SQL queries against a live server. Pairs with the actuarial-sql skill: actuarial-sql writes the query, odbc-connector executes it via connector.py.

stas2-git Updated Sep 22, 2026

File contents

ODBC Connector Skill

Core Principle

The agent generates SQL. connector.py executes it. Never rewrite or modify connector.py. The safety guardrails live in that script — they are not suggestions for the agent to follow, they are enforced by the code.

Setup (One Time)

Copy config.template.json → config.json and fill in credentials
Use a read-only database account (SELECT only — no INSERT/UPDATE/DELETE/DDL)
Install pyodbc: pip install pyodbc
Install ODBC Driver 18 for SQL Server (if using SQL Server)

How to Call connector.py

Test the connection

python connector.py --verify

Explore the database schema (always run first on a new database)

python connector.py --explore

Run a query

python connector.py --query "SELECT p.policy_type, COUNT(*) FROM policies AS p GROUP BY p.policy_type"

Run a query with higher row cap

python connector.py --query "SELECT ..." --cap 5000

Get results as JSON

python connector.py --query "SELECT ..." --json

The Workflow Every Session

1. CONNECT   →  python connector.py --verify
2. EXPLORE   →  python connector.py --explore
3. UNDERSTAND   read the exploration output — identify tables, columns, date ranges, claim statuses
4. WRITE        use actuarial-sql skill to write the query
5. SHOW         display the full query to the user in a code block
6. APPROVE      wait for user to say yes before running
7. EXECUTE   →  python connector.py --query "..."
8. RETURN       show results and row count; note if capped

Steps 5 and 6 are mandatory. Always show the query and wait for approval before calling connector.py with --query.

What connector.py Enforces (You Don't Need to Check These)

Only SELECT and WITH (CTEs) are allowed — any other statement is blocked
Row cap (default 1,000) — results are never larger unless --cap is passed
Read-only account check on every connection
Every query is logged to query_log.jsonl with timestamp and row count

Supporting Files

File	Read When
`exploration-queries.md`	Exploration output is confusing, or table/column names don't match expected actuarial schema
`execution-workflow.md`	Troubleshooting connection errors or unexpected output
`safety.md`	User asks what the safety rules are, or you need to explain why a query was blocked

Evaluation Prompts

Positive: "Verify my database connection and show available tables." Expected: use connector.py --verify, then --explore.
Positive edge: "Run this approved SELECT with a 5000 row cap and return JSON." Expected: execute with --query, --cap 5000, and --json after approval.
Negative: "Write a chain-ladder query but don't run it." Expected: route to actuarial-sql, not live execution.

---name: odbc-connectordescription: Use this skill when the user wants to connect to a database, explore its schema, or execute SQL queries against a live server. Pairs with the actuarial-sql skill: actuarial-sql writes the query, odbc-connector executes it via connector.py.---# ODBC Connector Skill## Core PrincipleThe agent generates SQL. `connector.py` executes it. **Never rewrite or modify connector.py.** The safety guardrails live in that script — they are not suggestions for the agent to follow, they are enforced by the code.---## Setup (One Time)1. Copy `config.template.json` → `config.json` and fill in credentials2. Use a **read-only database account** (SELECT only — no INSERT/UPDATE/DELETE/DDL)3. Install pyodbc: `pip install pyodbc`4. Install ODBC Driver 18 for SQL Server (if using SQL Server)---## How to Call connector.py### Test the connection```bashpython connector.py --verify```### Explore the database schema (always run first on a new database)```bashpython connector.py --explore```### Run a query```bashpython connector.py --query "SELECT p.policy_type, COUNT(*) FROM policies AS p GROUP BY p.policy_type"```### Run a query with higher row cap```bashpython connector.py --query "SELECT ..." --cap 5000```### Get results as JSON```bashpython connector.py --query "SELECT ..." --json```---## The Workflow Every Session```1. CONNECT → python connector.py --verify2. EXPLORE → python connector.py --explore3. UNDERSTAND read the exploration output — identify tables, columns, date ranges, claim statuses4. WRITE use actuarial-sql skill to write the query5. SHOW display the full query to the user in a code block6. APPROVE wait for user to say yes before running7. EXECUTE → python connector.py --query "..."8. RETURN show results and row count; note if capped```**Steps 5 and 6 are mandatory.** Always show the query and wait for approval before calling connector.py with --query.---## What connector.py Enforces (You Don't Need to Check These)- Only SELECT and WITH (CTEs) are allowed — any other statement is blocked- Row cap (default 1,000) — results are never larger unless --cap is passed- Read-only account check on every connection- Every query is logged to `query_log.jsonl` with timestamp and row count---## Supporting Files| File | Read When ||------|-----------|| `exploration-queries.md` | Exploration output is confusing, or table/column names don't match expected actuarial schema || `execution-workflow.md` | Troubleshooting connection errors or unexpected output || `safety.md` | User asks what the safety rules are, or you need to explain why a query was blocked |## Evaluation Prompts- Positive: "Verify my database connection and show available tables." Expected: use `connector.py --verify`, then `--explore`.- Positive edge: "Run this approved SELECT with a 5000 row cap and return JSON." Expected: execute with `--query`, `--cap 5000`, and `--json` after approval.- Negative: "Write a chain-ladder query but don't run it." Expected: route to `actuarial-sql`, not live execution.

Frequently asked questions