antonbabenko/agent-plugins

★ 15Audience · developerComplexity · 2/5ActiveSetup · easy

Mindmap

mindmap
  root((agent-plugins))
    Inputs
      Agent host
      Project files
      Language server
    Outputs
      Safer refactors
      Honest failure messages
      Terraform guidance
    Use Cases
      Stop blind rename
      Avoid Terraform destroys
      Delegate to subagents
    Tech Stack
      Markdown
      npx skills
      Terraform
      LSP

mindmap root((agent-plugins)) Inputs Agent host Project files Language server Outputs Safer refactors Honest failure messages Terraform guidance Use Cases Stop blind rename Avoid Terraform destroys Delegate to subagents Tech Stack Markdown npx skills Terraform LSP

Things people build with this

USE CASE 1

Stop an AI agent from doing a blind text rename by routing through a language server when one is available.

USE CASE 2

Force the agent to declare on line one when it falls back to regex because the right tool is missing.

USE CASE 3

Guide an agent writing Terraform toward moved blocks instead of resource renames that would destroy and recreate.

USE CASE 4

Install the same skill set across Claude Code, Codex, Cursor, Copilot, Gemini CLI, and OpenCode from one repo.

Tech stack

MarkdownnpxTerraformLSP

Getting it running

Difficulty · easy Time to first run · 5min

Each agent host has its own install path, so picking the right marketplace command for Claude Code, Codex, Cursor, or Gemini CLI is the only real gotcha.

In plain English

Agent Plugins is a collection of small add-ons that aim to make AI coding assistants behave more carefully when they edit code. The author, Anton Babenko, packages them as a marketplace for tools like Claude Code and OpenAI Codex, with installation paths also documented for Cursor, Copilot, Gemini CLI, OpenCode, and a few other agent hosts. The pitch is that these are skills the agent loads on demand and applies while it works, rather than long prose guides that the agent reads once and ignores. The headline plugin is called code-intelligence. Its job is to stop an AI agent from doing a blind text search and replace when it tries to rename a variable or find every reference to a piece of code. With the plugin in place, the agent first checks whether a language server is available for the project, with terraform-ls given as the worked example, and uses position-anchored reference lookups when it can. When the right tool is missing, the plugin makes the agent say so on the first line of its reply, instead of quietly falling back to a regex and claiming the result is complete. A second plugin, terraform-skill, lives in a separate repository but installs the same way. It tries to guide the agent toward Terraform's real failure modes, such as identity churn, secret exposure, blast radius, and state corruption, before any HCL code is written. The README lists a few before and after comparisons, for example using a moved block to change a resource address instead of a text rename that would trigger a destroy and recreate on apply. A third plugin called claude-delegator is mentioned in the install commands but not described in detail. Installation is the bulk of the README. The recommended path is an npx skills command that works across any compatible agent host. There are also dedicated marketplace commands for Claude Code and Codex, plus collapsible sections with clone and symlink instructions for Gemini CLI, Cursor, Copilot, OpenCode, the Codex clone fallback, Antigravity, and a manual Claude Code setup. The repository has fifteen stars and a continuous integration badge for a validation workflow.

Copy-paste prompts

Prompt 1

Install antonbabenko/agent-plugins into Claude Code using the npx skills path. Show the exact command and where the code-intelligence skill ends up on disk.

Prompt 2

Walk through what code-intelligence in agent-plugins does when terraform-ls is missing from the project. Show the first-line message the agent must print.

Prompt 3

Add the terraform-skill plugin from agent-plugins to Cursor. Show the symlink layout and confirm Cursor picks it up.

Prompt 4

Use agent-plugins to refactor a Terraform module so a resource address change uses a moved block instead of a destroy and recreate.

Prompt 5

Compare agent-plugins to a plain CLAUDE.md prose guide for the same goal. Explain why a loaded skill behaves differently from prose.

Open on GitHub → Explain another repo

Generated 2026-05-22 · Model: sonnet-4-6 · Verify against the repo before relying on details.