[DRAFT] RFC: DSQL Agent Plugin

[anwesham-lab]

Is this related to an existing feature request or issue?

No

Summary

DSQL already has an active Agent Skill hosted in the MCP repository. We'd likely

Use case

Anyone developing or exploring agentic development with Aurora DSQL.

Proposal

Add the DSQL plugin on the marketplace. This plugin will offer the DSQL skill and MCP server to start at minimum, to mirror the user experience of the Kiro Power for DSQL and offer comprehensive development guides, best practices, and interactive onboarding experiences with Aurora DSQL.

Will additionally include tools that execute the detailed workflows in the top level SKILL.md and more as tools with hook-based invocation for common migration patterns from standard PostgreSQL that we support with alternative patterns of change in DSQL.

Current Skill layout:

mcp/src/aurora-dsql-mcp-server/skills/dsql-skill]  
dev-dsk-anwemuk-2c-3fcca56d % tree -aL 4
.
├── mcp
│   ├── .mcp.json
│   ├── mcp-setup.md
│   └── mcp-tools.md
├── references
│   ├── access-control.md
│   ├── ddl-migrations.md
│   ├── development-guide.md
│   ├── dsql-examples.md
│   ├── language.md
│   ├── mysql-to-dsql-migrations.md
│   ├── onboarding.md
│   └── troubleshooting.md
├── scripts
│   ├── cluster-info.sh
│   ├── create-cluster.sh
│   ├── delete-cluster.sh
│   ├── list-clusters.sh
│   ├── loader.sh
│   ├── psql-connect.sh
│   └── README.md
└── SKILL.md

3 directories, 19 files

Expect some scripts to be transformed into tools and hooks.

Out of scope

Complete and thorough contents of additional tools and hooks not yet anticipated. We expect the plugin to evolve as we discover additional wants for an optimal user experience.

Potential challenges

Keeping source of truth synced between the skill that's already in use in the MCP repository and the skill we would like packaged as part of a DSQL plugin.

Likely will require a github action for PRs to be created in the other repository when a change is pushed to one.

Dependencies and Integrations

re: potential challenges, we'll want to ensure we're consistent with the existing DSQL skill

Alternative solutions

[krokoko]

Thanks @anwesham-lab for this RFC !

Strengths

Focused plugin: Single domain (Aurora DSQL, agentic development, migrations) fits “do one thing well.”
Skill + MCP: Bundling the DSQL skill with the MCP server matches the deploy-on-aws pattern and is appropriate.
References: Existing reference set (access-control, ddl-migrations, development-guide, etc.) fits the “reference files” pattern; they must be checked for length and single-topic.
Starting minimal: “Skill and MCP server to start at minimum” is aligned with incremental delivery.

A few questions:

• Which MCP server(s) will the plugin bundle? Is it the existing aurora-dsql MCP server from the MCP repo, and will it be referenced the same way as in that repo?
• Can you share the current SKILL.md frontmatter (name, description, tags, ...) and approximate line counts for SKILL.md and each file in references/ ? Do you plan to trim or split any to meet the ~200–300 (SKILL) and ~50–100 (references) line guidance?
• Will the plugin expose one skill (e.g. dsql) or multiple (e.g. dsql, dsql-migrate)?
• For v1, which migration patterns will be covered by tools/hooks, and which will remain documentation-only?
• Who will be the plugin team (CODEOWNERS) for plugins//?

Thank you !