Short SKILL.md briefings for AI coding agents — so they get right what they often get wrong about MariaDB (MySQL compatibility, vectors, replication, Oracle migration, and more). Skills install on the AI agent, not on your MariaDB server.
A skill is a curated markdown file the agent reads when a topic matches. Think of it as a briefing note: MariaDB-specific guidance, common LLM mistakes, and pointers to mariadb.com/docs.
Skills work with Claude Code, GitHub Copilot, Cursor, OpenAI Codex, and 20+ other tools. Learn more at agentskills.io and browse community skills at skills.sh.
We have created an initial set of eight skills at github.com/mariadb/skills — each for a different situation where agents need MariaDB-specific guidance. Install instructions and how to suggest improvements follow below.
Skills use wrong/right pairs, version annotations, and links to official documentation so agents can tailor advice to the MariaDB version you run.
For developers migrating from MySQL, or where MySQL habits cause unexpected behavior in MariaDB. Covers authentication, JSON differences, GTID incompatibility with MySQL, and features MariaDB has that MySQL does not.
For developers who want to use MariaDB well or evaluate a migration. System-versioned tables, RETURNING, sequences, instant DDL, native IP types, Oracle compatibility mode, Galera, ColumnStore, and more — the things agents rarely suggest unprompted.
For developers migrating from Oracle Database. sql_mode=ORACLE, PL/SQL compatibility, data type mapping, what needs manual rewriting, and migration tools.
For diagnosing slow queries and designing indexes. EXPLAIN, composite index rules, cursor-based pagination, histograms, and MariaDB-specific optimizer behavior.
For replicated and HA setups — async replication, GTID, semi-sync, parallel replication, Galera Cluster, and application constraints that matter in production.
For automatic row history without triggers or audit tables — FOR SYSTEM_TIME queries, partitioning, and gotchas.
For AI applications, RAG, and semantic search. Native vector support since MariaDB 11.7 — no extension or plugin to install.
For connecting agents to a MariaDB database via the Model Context Protocol — schemas, read-only SQL, optional semantic search, and secure setup.
Use npx skills for an interactive install — choose which MariaDB skills to install and which agents on your system (Claude, Cursor, Codex, etc.):
npx skills add mariadb/skills
Or install one skill:
npx skills add mariadb/skills/<skill-name>
Replace <skill-name> with e.g. mariadb-vector or mysql-to-mariadb. Installs are counted on skills.sh.
Download SKILL.md into the agent's skill folder, for example:
mkdir -p ~/.claude/skills/<skill-name>
curl -o ~/.claude/skills/<skill-name>/SKILL.md \
https://raw.githubusercontent.com/mariadb/skills/main/<skill-name>/SKILL.md- Claude Code and Claude Desktop:
~/.claude/skills/<skill-name>/SKILL.md - OpenAI Codex:
~/.agents/skills/<skill-name>/SKILL.md
Once installed, skills activate automatically. When your question matches a skill's topic, the agent reads it and applies the guidance for that session. No manual activation is needed.
Open for improvements and more skills. Contributions and feedback welcome via pull requests.