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 !

[anwesham-lab]

Thanks for the response! I'll try to cover the current thoughts for each question, but if you're available during the meeting time for the invite I sent, we can follow up more during.

---

* Which MCP server(s) will the plugin bundle? Is it the existing [aurora-dsql MCP server](https://github.com/awslabs/mcp/tree/main/src/aurora-dsql-mcp-server) from the MCP repo, and will it be referenced the same way as in that repo?

Yes and mainly; I'll leverage plugin structure so that the MCP is actively initialized (but the MCP steering files remain the same).

---

* Can you share the current SKILL.md frontmatter (name, description, tags, ...) and 

Here's the link to SKILL.md

XML Header:

---
name: dsql
description: Build with Aurora DSQL - manage schemas, execute queries, and handle migrations with DSQL-specific requirements. Use when developing a scalable or distributed database/application or user requests DSQL.
---

---

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?

Line Count:
SKILL.md: 274

References (we wouldn't initially plan to trim and some of the files are significantly longer than the 50-100 line guidance, curious what the benefits of this have been, we've seen success with the skill's reference files at full length in practice and benchmark evaluation tests):

development-guide.md: 415 lines, 317 loc

languages.md: 149 lines, 125 loc

onboarding.md: 324 lines, 239 loc

• there is intent to trim here, but likely to at best 150 loc

troubleshooting.md: 113 lines, 85 loc

dsql-examples.md: 470 lines, 358 loc

ddl-migrations.md: 635 lines, 486 loc

• Can be split/transformed into folder with multiple workflows as their own files
• possible transition into its own skill, but for existing user convenience, this change wouldn't be immediate

mysql-to-dsql-migraitons.md: 1073 lines, 879 loc

• Same as above:
  • Can be split/transformed into folder with multiple workflows as their own files
  • possible transition into its own skill, but for existing user convenience, this change wouldn't be immediate

Worth noting that significant amounts of content are tabular and brief in some of the longer reference files, so they're not context rich in a way that tends to explode context windows. They also have clear heading structures and are referenced with relevant headings in top-level docs when necessary to help.

---

* Will the plugin expose one skill (e.g. dsql) or multiple (e.g. dsql, dsql-migrate)?

To start: Just the one dsql skill
Eventually: we'd likely move to multiple specialized skills for better/more detailed paths on things like migrations as you mention

---

* For v1, which migration patterns will be covered by tools/hooks, and which will remain documentation-only?

v1: I want to get our skill as is, on the plugin marketplace.

All scripts in dsql-skill/scripts can immediately be turned into tools as desired/treated as tools in their as-is state. The scripts already employ them as standard executable bash.

Migrations and other workflows as tools/hooks would be follow-ups.

---

* Who will be the plugin team (CODEOWNERS) for plugins//?

SAME CODEOWNERS as DSQL MCP or we can use aws-aurora-dsql-maintainers for simplicity.

[scottschreckengaust]

A request @awslabs/agent-plugins-writers as the parent team to @awslabs/aws-aurora-dsql-maintainers would be the first step, then use the team in the .github/CODEOWNERS.

[anwesham-lab]

Request made for new pending team request. Team: https://github.com/orgs/awslabs/teams/agent-plugins-dsql

[scottschreckengaust]

Request made for new pending team request. Team: https://github.com/orgs/awslabs/teams/agent-plugins-dsql

approved 👍🏼

[theagenticguy]

lgtm