AGENT SKILL

bedrock-guide

Build on AWS with Amazon Bedrock — managed foundation models, agents, knowledge bases, and guardrails.

License 许可
MIT
所属仓库
TanNhatCMS/jarvis
最近更新
2026-11-07
Stars
★ 0
Forks
1

01 前言 Front Matter

name
bedrock-guide
category
ai-research
description
Build on AWS with Amazon Bedrock — managed foundation models, agents, knowledge bases, and guardrails.

03 SKILL.md 正文

Overview

Amazon Bedrock is AWS's managed service for foundation models: access models from multiple providers (Anthropic, Meta, Mistral, Amazon's own, and others) through a unified AWS API, with managed capabilities for agents, knowledge bases (RAG), and guardrails — all inside your AWS account with IAM, VPC, CloudWatch, and regional controls. The pitch is enterprise GenAI on AWS terms: your data stays in your account boundary, access is IAM-governed, and billing flows through AWS.

For AWS-centric organizations, Bedrock is usually the default starting point — not necessarily because it's the best model API, but because it inherits your existing security, compliance, networking, and procurement posture. The models are good; the enterprise integration is the point.

The mental model: Bedrock brings foundation models into the AWS control plane. Evaluate it as AWS infrastructure with model capabilities, not as a model lab with AWS hosting.

When to use

  • AWS-centric organizations building GenAI features (leverage existing IAM, VPC, compliance).
  • Applications needing data to stay within your AWS account/region boundary.
  • Managed RAG via Knowledge Bases (OpenSearch/Aurora-backed retrieval without building it).
  • Managed agents with action groups for AWS-integrated tool use.
  • Guardrails for content filtering and PII redaction as a managed layer.
  • Procurement through AWS (enterprise agreements, private pricing).

Core concepts

  • Foundation models on Bedrock: multiple providers' models behind one API (Converse API for chat). Model access must be enabled per region — check availability in your operating regions first.
  • Cross-region inference: some models support routing across regions for throughput. Understand the data-residency implications for your compliance requirements.
  • Knowledge Bases: managed RAG — connect S3/data sources, Bedrock handles chunking, embeddings, and retrieval against OpenSearch Serverless or Aurora. The fastest compliant RAG on AWS.
  • Agents for Bedrock: managed agent orchestration with action groups (Lambda or OpenAPI-defined tools). Useful when your tools are already AWS-native.
  • Guardrails: managed content filters, denied topics, PII redaction, and grounding checks. A policy layer independent of the model — configure per use case.
  • IAM and VPC: model invocation governed by IAM policies; VPC endpoints keep traffic off the public internet. This is the enterprise control story — use it fully.
  • Provisioned Throughput: reserved model capacity for latency/throughput guarantees. For production SLOs beyond on-demand variability.
  • Model evaluation: managed eval jobs (automatic metrics + human workflows) for comparing models and prompts on your data.

Practical workflow

  1. Confirm regional availability. Check model availability in your required regions — Bedrock's catalog varies by region, and compliance may pin you to specific ones.
  2. Set up IAM properly. Least-privilege policies for model invocation, scoped per application. Use VPC endpoints where network isolation matters.
  3. Benchmark candidate models. Same eval set across Bedrock's model options via the Converse API. Include Amazon's own models — they're often cost-competitive.
  4. Build RAG with Knowledge Bases. For document-grounded apps: connect data sources, configure chunking, evaluate retrieval quality before building agents on top.
  5. Configure guardrails per use case. Content filters, PII handling, denied topics — tuned to the application's risk profile, not one-size-fits-all.
  6. Use agents where tools are AWS-native. If actions are Lambda functions and APIs in your AWS environment, Bedrock Agents reduce integration work.
  7. Evaluate and monitor. Run Bedrock eval jobs on model/prompt changes; monitor invocation metrics, guardrail interventions, and costs in CloudWatch.

Checklist for Bedrock production:

  • Model availability confirmed in operating regions.
  • IAM least-privilege; VPC endpoints where needed.
  • Knowledge Base retrieval quality evaluated.
  • Guardrails configured per use case.
  • Provisioned throughput sized if SLOs require it.

Common pitfalls

  • Region availability assumptions. Designing around a model unavailable in your compliance-required region. Check first, always.
  • Cross-region inference vs. data residency. Throughput routing across regions may violate data-residency requirements. Understand where data goes.
  • IAM over-permissioning. Broad bedrock:* policies. Scope invocation permissions per application and model.
  • Knowledge Base as magic. Expecting managed RAG to work without evaluating chunking and retrieval quality on your documents. Test retrieval first.
  • Guardrails as the only safety. Managed filters help; they don't replace application-level validation, prompt design, and human oversight where needed.
  • On-demand for strict SLOs. Shared on-demand capacity has variability. Provisioned throughput exists for guarantees — use it when SLOs demand.
  • Cost opacity. Bedrock usage across many models/apps without tagging and attribution. Tag resources; monitor per-application spend.
  • Treating Bedrock as model-agnostic. Models differ meaningfully; "we use Bedrock" isn't a model choice. Benchmark and choose deliberately.

04 同仓库更多 Skills

TanNhatCMS
TanNhatCMS/jarvis

web-scrape

Turn web pages into LLM-ready markdown via the Firecrawl API (renders JS, anti-bot handling). Use when browser_open/fetch fails on JS-heavy pages, or when you need clean markdown/structured extraction from a site. Requires FIRECRAWL_API_KEY.

★ 0
TanNhatCMS
TanNhatCMS/jarvis

design-taste-frontend

tasteskill v2, the anti-slop frontend skill. Art direction for websites, distilled from a study of 247 curated gallery sites (One Page Love, Awwwards Sites of the Day). Deals researched directions (idea, composition, type, palette, photographs, section blueprints) from a library, enforces a strict no-slop house style with a real-browser inspector, and runs a look-and-fix loop on screenshots. Use this whenever you build, design, redesign, restyle, polish or review a landing page, marketing site, portfolio, product page, one-pager, hero or any public-facing web page, in any stack (HTML/CSS, React, Next.js, Vue, Svelte, Astro, Tailwind) — also when the user says a page looks generic, templated, "AI-made", sloppy, or asks to make it look designed, premium, award-level or less like every other site. Not for dashboards or app screens.

★ 0
TanNhatCMS
TanNhatCMS/jarvis

test-driven-development

Use when implementing any feature or bugfix, before writing implementation code

★ 0
TanNhatCMS
TanNhatCMS/jarvis

systematic-debugging

Use when encountering any bug, test failure, or unexpected behavior, before proposing fixes

★ 0
TanNhatCMS
TanNhatCMS/jarvis

requesting-code-review

Use when completing tasks, implementing major features, or before merging to verify work meets requirements

★ 0
TanNhatCMS
TanNhatCMS/jarvis

brainstorming

You MUST use this before any creative work - creating features, building components, adding functionality, modifying behavior, or planning anything new, in software or out of it (a talk, a business, a renovation).

★ 0