ferdinandobons/awsbedrockagentcoreskill

AWS Bedrock AgentCore Skill is a plugin for Claude Code, the AI coding assistant, that gives it a pre-researched playbook for building AI agents on Amazon Web Services. Without something like this, an AI coding agent asked to build an agent on AWS would need to search across scattered documentation, encounter outdated examples, and make version-specific mistakes that only show up when you try to deploy. This plugin front-loads that research so the coding agent can go straight to the correct approach. The plugin covers three main AWS products: Strands Agents (a framework for building agents in Python), Amazon Bedrock (AWS's managed service for running AI models, with features like knowledge bases for searching documents and guardrails for safety filtering), and Bedrock AgentCore (a newer set of managed services for running agents in production, handling things like memory across sessions, secure tool access, and built-in browser and code-execution capabilities). The plugin is structured as a decision tree plus a reference library. When you describe what you want to build, a routing file called SKILL.md maps your use case to a recommended pattern and points to the relevant reference files. There are 20 reference files in total, covering roughly 19,000 lines of guidance. Each best practice and code snippet includes an inline link to its official AWS documentation source, so you can verify any recommendation rather than taking it on trust. The project notes 636 such citations across the files. The plugin also documents specific mistakes that are easy to make when building on these services: using older API patterns instead of the current ones, setting cache durations that a model does not support, or getting memory size wrong in ways that cause silent failures. These corrections are embedded alongside the correct approach. Installing the plugin takes two commands in Claude Code. Once installed, it activates automatically when a task involves building or configuring agents on AWS, without requiring an explicit invocation. The project is licensed under MIT.