docker

Docker Agent Plugin

Inspect, diagnose, and review local Docker environments and Dockerfiles: containers, Compose projects, and best practices.

An Agent Plugin (spec 1.0.0).

What is this?

A skills-only plugin that teaches agents to work with the docker and docker compose CLIs and review Dockerfiles — which commands to run, in what order, and what the output means. No MCP server, no external dependencies.

Why would I install it?

• Lightweight and dependency-free (agents already have docker CLI access).
• Read/inspect before mutation; destructive operations are never automatic.
• Dockerfile review adds a distinct, valuable workflow.

What MCP/upstream does it use?

None — deliberately skills-only. Docker's official MCP (docker/hub-mcp) targets Docker Hub, not local daemons; Docker MCP Gateway manages MCP server lifecycles, not local container diagnostics. A separate future docker-hub plugin may use the Hub MCP — it is a different product and is not combined here.

Does it need authentication?

No. Docker's own auth (~/.docker/config.json for registries) is used by the CLI; this plugin never reads or stores it.

Is it read-only?

Skills default to read-only. All skills prefer docker ps, inspect, logs, compose config, compose ps before any mutation. Destructive operations (rm, prune, down -v, system prune, volume delete) are gated behind explicit user intent and never run automatically.

What clients were actually tested?

Any client that gives the agent shell access to docker/docker compose. The skills are CLI-command workflows; verify in your client that the agent can run shell commands.

Requirements

• Docker CLI + daemon (docker), and the compose plugin for diagnose-compose.
• Shell access for the agent.

Installation

Copy the plugin's runtime files (plugin.json, skills/, README.md, LICENSE — this plugin has no mcp.json) into your Agent Plugins client's plugin folder, or use your client's plugin install flow. Do not copy the authoring sources (plugin.yml, skills-src/, docs/) — see the collection README for the full boundary. No build step is required, and no external packages are fetched.

Examples

1. "What's running in Docker on this machine?" → inspect-docker-environment.
2. "My api container keeps restarting — why?" → diagnose-container.
3. "Explain what this compose file does and how the services connect" → diagnose-compose.
4. "Review this Dockerfile before we ship it" → review-dockerfile.

Upstream dependencies

None (skills only). The skills reference the standard docker and docker compose command-line interfaces.

Agent Plugins version

1.0.0 (Working Draft) — https://agent-plugins.org/specification

Runtime requirements

Docker CLI + daemon; compose plugin for compose workflows.

Security / default behavior

• All skills default to read-only commands.
• Destructive operations are explicitly gated behind user authorization.
• Agents should not dump full container inspect JSON into responses; the skills favor --format projections.
• No credentials are stored in the plugin.

Known limitations

• Requires the agent to have shell access to the docker CLI.
• Does not manage Docker Hub (separate product, future docker-hub plugin).

License

MIT. See LICENSE.

CHANGELOG

See CHANGELOG.md.