amazon-bedrock Skill (AWS's own repo) for Claude Code: What It Does and How to Install It
Teaches your assistant to build generative AI apps on Amazon Bedrock, from model calls to Knowledge Bases, Guardrails and AgentCore.
- Installs
- 8.8K
- on skills.sh
- Stars
- 3.4K
- on GitHub
- Files
- 46
- besides SKILL.md
- Updated
- Oct 2026
- last repo commit
This skill, from AWS's own agent-toolkit repo, gives your assistant working knowledge of Amazon Bedrock. It routes each task to a reference file: model calls with the Converse API, Knowledge Bases, Agents, Guardrails, the AgentCore services, prompt caching, quotas, costs and model migrations.
It runs AWS CLI, boto3 and Python commands with your credentials and can create resources, IAM roles and payment setup in your account. The folder has 47 files. Use a development account and a limited role.
Good for
- Developers building generative AI apps on AWS
- Teams setting up RAG with Bedrock Knowledge Bases
- Engineers deploying agents to AgentCore or migrating classic Bedrock Agents
- Anyone debugging Bedrock throttling, access or cost problems
Install amazon-bedrock
Claude Code
- 1
Copy this prompt.
Install the amazon-bedrock skill in Claude Code. Run this command: /plugin install aws-core@claude-plugins-official Then check that it loaded and tell me how to use it.
- 2
Paste it into Claude Code. It shows the change and asks before running it.
Requires AWS CLI v2 with credentials, boto3 1.34 or newer.
Pick one route. The aws-core plugin installs 25 skills, an AWS MCP server config (.mcp.json) and a Claude Code hook (secret-safety.py), not just this skill. The skills CLI command in the README installs the repo's skills folder; the folder and generated skills CLI routes install only amazon-bedrock. The skill runs AWS CLI, boto3 and Python commands, so there is no Claude app upload route. Run it with a least-privilege role in a development account. The README says you need uv installed, and an AWS account with credentials configured locally for API calls and script execution (not for documentation search or skill discovery).
From the project README. Source.
What's in the skill
When it loads
Builds generative AI applications on Amazon Bedrock. Covers model invocation (Converse API, InvokeModel), RAG with Knowledge Bases, Bedrock Agents, Guardrails, and AgentCore (including the Harness managed agent loop). Applies when invoking models, setting up Knowledge Bases, deploying to AgentCore, migrating/porting/converting a Bedrock Agent (including inline agents) to an AgentCore Harness, troubleshooting Bedrock errors (ThrottlingException, AccessDeniedException), or choosing Bedrock models (Claude, Llama, Nova, Titan). Also for KB migration to Managed, prompt caching, quota and throttling diagnosis, cost tracking, migrating between Claude model generations, chunking strategies, and Bedrock model selection. Covers AgentCore Payments (x402, microtransactions, Payment Manager, Connector, Instrument, Coinbase CDP, Stripe Privy, paid endpoints, agent payments). NOT for custom model training, Rekognition, or Comprehend. For SageMaker model discovery, customization, or endpoints, use aws-ai-ml.
Sections in SKILL.md
- Guardrail — where this skill's own files live (MCP vs local install)
- Table of Contents
- Overview
- Bedrock API Landscape
- Critical Warnings
- Security Considerations
- Converse API vs InvokeModel
- Which Bedrock Capability Do You Need?
- Knowledge Bases (retrieval, agentic retrieval and RAG)
- Common Workflows
- Verify Dependencies
- Examples — mapping user intent to workflows
- Invoke a model
- Create a Knowledge Base
- Query a Knowledge Base
- Create an Agent with action groups
Key facts
- Installs on skills.sh
- 8,750 (Nov 2, 2026)
- Last repo commit
- Oct 30, 2026
- Audits on skills.sh
- Gen Agent Trust Hub: pass, Socket: warn, Snyk: warn
- What it includes
- 47 files: SKILL.md, 42 reference files, 3 asset templates and 1 Python script (fetch_bedrock_agent.py)
- Skill version
- 8 (SKILL.md metadata)
What it includes
Which of 5 Bedrock endpoints and 3 AgentCore endpoints to call for each job.
The 5 are bedrock, bedrock-runtime, bedrock-mantle, bedrock-agent and bedrock-agent-runtime. Using the wrong one is a common error.
Bedrock API landscapeFour: set maxTokens, PII still reaches CloudWatch Logs, SDK versions, Agents classic.
Unset maxTokens reserves far more quota than needed. Bedrock Agents classic is in maintenance mode and closed to new customers.
Critical warningsUse the Converse API; check model access first, then call it with maxTokens set.
The AWS CLI cannot stream, so ConverseStream needs an SDK.
Runs code ·
Invoke a modelCreate, query and migrate; Managed Knowledge Base by default for new RAG work.
Creating one runs create-knowledge-base, a data source and an ingestion job. Customer-managed KBs only if you ask.
Runs code ·
Knowledge BasesCreate an agent with action groups; always run prepare-agent after any change.
The skill notes that agents often skip prepare-agent. Do not use double underscores in action group names.
Runs code ·
Agents and action groupsThree integration modes, with a PII logging compliance warning shown first.
Masking applies to the response only; the original text is still logged in CloudWatch.
Runs code ·
Guardrails13 rows: Harness, Gateway, Runtime, Memory, Identity, Policy, Payments, Registry and more.
Each row points to a reference file the assistant must read first. Harness is the preferred route for new agent builds.
Runs code ·
AgentCore servicesLets an agent pay x402 endpoints in USDC through Coinbase CDP or Stripe Privy.
See the permissions section: setup creates IAM roles, a payment manager and a wallet.
Runs code ·
AgentCore PaymentsThree diagnostic workflows with CLI and CloudWatch checks.
Quota health compares limits to peak use. Cost tracking covers inference profile tags, CUR 2.0 and AWS Budgets.
Runs code ·
Prompt caching, quota health, cost trackingCustomer-managed KB to Managed KB, Bedrock Agent to AgentCore Harness, Claude versions.
The KB migration builds a new Managed KB beside the old one. The agent migration drives the AgentCore CLI and has approval gates.
Runs code ·
MigrationsModel ID formats, quick defaults, and 18 common errors with fixes.
Covers AccessDenied, Malformed input request, ThrottlingException, empty KB results and more.
Model selection and troubleshootingSnapshots a Bedrock Agent into one JSON manifest for migration planning.
Run as python fetch_bedrock_agent.py --agent-id ID --region REGION --out manifest.json. The script file says it needs boto3.
Runs code ·
fetch_bedrock_agent.py
From the official docs
Permissions
High risk
It can create and change AWS resources, IAM roles and payment setup in your account, and the plugin route adds an MCP server and a hook. Use a development account and a limited role, and read each command first.
What it can do
- Read your data
- Create and change things
- Delete things
- Read-only mode
Watch out for
It creates and changes resources in your AWS account
References run create-knowledge-base, create-agent, create-role, put-role-policy, create-gateway-target, create-budget and more; one mentions delete-knowledge-base and delete-stack. The SKILL.md says to confirm before destructive steps. Use a dev account and a limited role. github.com
AgentCore Payments lets an agent pay in USDC
The payments reference sets up an IAM role, a payment manager, a connector and a wallet, and agents then pay x402 endpoints from an end user's wallet. It suggests Base Sepolia (testnet) first and says mainnet wallets are funded by the end user. Set session budgets. github.com
The aws-core plugin adds an MCP server and a hook
Its .mcp.json runs uvx mcp-proxy-for-aws-cli==1.7.0 against https://aws-mcp.us-east-1.api.aws/mcp with --skip-auth. A PreToolUse hook runs secret-safety.py on Bash and AWS tool calls; it reads the tool call from stdin and denies direct Secrets Manager value fetches. The hook has a 5 second timeout, and the script's own comment says a hang would let the tool call through. github.com
It can install the AgentCore CLI globally
references/agentcore-harness.md says to run npm install -g @aws/agentcore@preview. That is a preview npm channel. github.com
One script writes an agent manifest file
fetch_bedrock_agent.py reads a Bedrock Agent with boto3 and writes JSON to the --out path. The manifest can hold prompts and schemas, so keep it out of version control. github.com
KB migration provisions only
safety-and-iam.md says the migration builds a new Managed KB beside the old one and never ingests data, cuts over traffic, decommissions or mutates the original. github.com
Technical details
| Turning on read-only | No switch. fetch_bedrock_agent.py is documented as needing only read-only permissions; the creation workflows are not. |
|---|---|
| Talks to | AWS service endpoints for your account and region. https://aws-mcp.us-east-1.api.aws/mcp (the aws-core plugin's MCP config). npm registry (npm install -g @aws/agentcore@preview). https://sandbox.node4all.com/v1/x402-test (the payments test example) |
| Sign-in and keys | The payments reference has you put Coinbase CDP or Stripe Privy credentials in a .env.payments file and load them into your shell. The provider credentials are then stored with bedrock-agentcore create-payment-credential-provider. |
| Scopes |
|
Questions
What does the amazon-bedrock skill do?
It teaches your assistant to build generative AI apps on Amazon Bedrock. That covers calling models with the Converse API, Knowledge Bases for RAG, Bedrock Agents, Guardrails and the AgentCore services, plus quota, caching, cost and model-migration help.
Is this AWS's own skill?
Yes. It lives in aws/agent-toolkit-for-aws, AWS's own repo, and the plugin marketplace there lists Amazon Web Services as owner.
How do I install it?
Use the install box above. In Claude Code the README installs the aws-core plugin with /plugin install aws-core@claude-plugins-official. The plugin adds 25 skills, an AWS MCP server config and a hook. The folder and skills CLI routes install less.
What do I need to run it?
AWS CLI v2 configured with valid credentials, boto3 1.34 or newer, and Bedrock model access enabled in your region. The SKILL.md says it uses the AWS MCP server when connected and falls back to the AWS CLI otherwise.
Will it create or change things in my AWS account?
Yes. Its references run commands that create knowledge bases, data sources, agents, IAM roles, CloudWatch alarms, budgets and AgentCore resources. The SKILL.md tells the assistant to present an overview first, explain each step, and confirm before deleting or overwriting. Use a development account.
Can it spend money?
Model calls and the resources it creates are billed to your AWS account. Its AgentCore Payments reference also lets an agent pay for x402 endpoints in USDC from an end user's wallet. It suggests starting on the Base Sepolia testnet, where tokens come free from a faucet.
Is it safe?
It can create and change AWS resources, IAM roles and payment setup in your account, so run it with a least-privilege role in a development account and read each command. The references call out risks such as PII staying in CloudWatch Logs. The plugin route also adds an MCP server and a hook, described in the permissions section.
Does it work in the Claude app?
The install box has no Claude app upload tab because the skill runs AWS CLI, boto3 and Python commands on your machine. The README documents plugins for Claude Code, Codex, Cursor and Devin.
Information
updated Nov 2, 2026
- Repository
- aws/agent-toolkit-for-aws
- Folder
skills/core-skills/amazon-bedrock
Sources
Docs, issue trackers and discussions this page is built from.
Show all 11Hide
- amazon-bedrock SKILL.md · github.com
- agent-toolkit-for-aws README · github.com
- aws-core plugin.json · github.com
- aws-core .mcp.json · github.com
- agent-toolkit-for-aws marketplace.json · github.com
- aws-core hooks.json · github.com
- secret-safety.py · github.com
- references/agentcore-payments.md · github.com
- references/agentcore-harness.md · github.com
- scripts/fetch_bedrock_agent.py · github.com
- references/bedrock-kb-migration/safety-and-iam.md · github.com