Agent skills

The CamusDB agent skills give a coding agent product-specific guidance for designing schemas, writing SQL, handling transactions, tuning queries, and operating CamusDB. They work with Claude Code, Codex, and Cursor.

The skills are a companion to the CamusDB MCP server:

• The skills teach the agent CamusDB syntax, behavior, workflows, and safety rules.
• The MCP server gives the agent typed tools for inspecting and working with a running CamusDB database.

You can use either one on its own. Used together, they let an agent apply CamusDB guidance while it examines schemas, explains plans, runs read-only queries, or performs the database operations its MCP credentials allow.

Available skills

Skill	Use it to
camusdb-schema-design	Review keys, data types, JSONB and GIN indexes, constraints, sequences, TTL, and sharding placement.
camusdb-sql	Write CamusSQL, including upserts, set operations, window functions, and JSONB queries; port PostgreSQL or MySQL SQL.
camusdb-transactions	Choose transaction modes, use locking reads and upserts, handle failed statements, and retry conflicts.
camusdb-query-performance	Diagnose plans and distributed queries using statistics, index usage, covering indexes, and the result cache.
camusdb-troubleshooting	Diagnose CADBxxxx errors, request limits, statement statistics, and slow or stuck workloads.
camusdb-dotnet	Use ADO.NET, Dapper, EF Core, JSONB mapping, pipelines, and OpenTelemetry.
camusdb-typescript	Use the TypeScript connector or call the HTTP API.
camusdb-vector-search	Store embeddings and write nearest-neighbor queries.
camusdb-migrations-and-recovery	Rehearse schema changes, recover data, and back up or restore a database.
camusdb-local-dev	Configure local nodes and clusters, choose storage backends including Solarium, use camus-cli, and connect MCP.

An agent reads the short description of each installed skill and loads the full instructions when a task matches. You do not need to load every skill into the conversation.

The 0.2.0 plugin refreshes all ten skills for the engine and client changes through November 4, 2026. Detailed references now cover JSONB/GIN, set and window queries, locking reads and upserts, planner behavior, and current MCP tools. For example, ask the agent to review a JSONB index or fix code that catches a statement error and continues an already-aborted transaction.

Install for Claude Code

Install the repository as a Claude Code plugin:

/plugin marketplace add camusdb/camusdb-skills
/plugin install camusdb@camusdb

You can instead clone the repository and link its skills into ~/.claude/skills:

git clone https://github.com/camusdb/camusdb-skills.git
cd camusdb-skills
./install.sh --agent claude

Install for Codex or Cursor

Codex and Cursor discover skills in ~/.agents/skills:

git clone https://github.com/camusdb/camusdb-skills.git
cd camusdb-skills
./install.sh --agent agents

Run ./install.sh without an --agent argument to install for all supported agents at once.

Install in a project

Install a copy in a project when the whole team should get the same skill files from version control:

./install.sh --project ~/code/my-app --copy

This writes the skills to .claude/skills and .agents/skills in the project. Use --copy for a project installation. A symbolic link points to your local clone and does not work on another developer's machine.

Use the skills with MCP

Install the skills and configure the MCP server in the same agent. Then ask for the task in ordinary language. For example:

• Review migrations/002_orders.sql against the CamusDB best practices.
• Explain why this CamusDB query is slow and verify the plan.
• Try this migration on a branch of the prod database first.
• Diagnose CADB0502 and fix the transaction retry logic.

You can also invoke a skill explicitly. Use /camusdb-schema-design in Claude Code and Cursor, or $camusdb-schema-design in Codex.

The skills are instructions, not a security boundary. The MCP server uses the CamusDB user and grants in its configuration. Give that user only the database privileges the agent needs, and prefer a database branch for experiments.

Update or remove

A symbolic-link installation follows the local clone, so git pull updates it. Run the installer again after a copy installation. To remove the installed skills, run:

./install.sh --uninstall