Agents Build

build

Add capabilities to your AgentCore agent project.

When to use

• Adding cross-session memory to your agent
• Calling your deployed agent from a web app, mobile app, or backend service
• Configuring VPC networking for private resources (RDS, internal APIs)
• Building multi-agent systems with orchestrator/specialist patterns
• Migrating an existing Bedrock Agent to AgentCore
• Adding the Browser tool so the agent can navigate websites
• Adding the Code Interpreter so the agent can execute code in a sandbox
• Adding AgentCore Payments so the agent can pay for x402- or MPP-protected APIs, tools, or content
• Removing resources from your project or tearing down a deployment

Do NOT use for:

• Connecting to external tools/APIs via Gateway (OpenAPI specs, Lambda, MCP servers, credentials, policies) → use agents-connect
• Scaffolding a new project → use agents-get-started
• Deploying → use agents-deploy

Input

$ARGUMENTS can be:

• A capability: “memory”, “integrate”, “vpc”, “multi-agent”, “migrate”, “browser”, “code-interpreter”, “payments”, “teardown”
• A description of what they want: “remember user preferences”, “call from React app”, “scrape a website”, “run pandas in the agent”, “delete my agent”, “clean up resources”
• Empty — the skill will determine the workflow from context

Process

Step 0: Verify CLI version

Run agentcore --version. This skill requires v0.9.0 or later.

If older: “Run agentcore update to get the latest version.”

Step 1: Read project context

Read agentcore/agentcore.json to understand the current project — framework, existing resources, agent configuration.

If agentcore/agentcore.json is not found:

1. Check if the developer is in the wrong directory. Look for agentcore/agentcore.json in parent directories (up to 3 levels). If found, tell them: “Found an AgentCore project at <path>. Are you working in that project?”
2. If no project exists anywhere nearby, ask what capability they wanted to add. Then offer two paths:
   • “I can walk you through creating a project first and then adding CAPABILITY — want to do that?” (run the get-started flow inline, then continue with the build workflow)
   • “If you already have a project elsewhere, cd into it and try again.”

Do not just say “go use agents-get-started” and stop — that loses the developer’s context about what they actually wanted to do.

Step 2: Determine the workflow

Important disambiguation — before routing to a build reference, check if the prompt is actually a connect or debug concern:

• If the phrase mentions external APIs, Lambda functions, OpenAPI specs, gateways, credentials, MCP servers, or policies → this is agents-connect, not build
• If the developer says something is broken (wrong answers, errors, tool failures) → this is agents-debug, not build
• Build is for adding new capabilities to a working project, not fixing broken ones

Based on the developer’s prompt and $ARGUMENTS, load the appropriate reference:

If the developer asks about the difference between local dev and deployed (e.g., “why does my memory work after deploy but not locally?”), load references/local-vs-deployed.md (opens in a new tab) alongside the specific workflow reference.

Read the matching file into context and follow its Process section step by step — do not summarize.

If the intent is ambiguous, ask the developer which capability they want to add.

Changing the model

The model is configured in app/<AgentName>/model/load.py (scaffolded by agentcore create). To change it:

1. Open app/<AgentName>/model/load.py
2. Change the model_id parameter in the BedrockModel() constructor

# Default (scaffolded by CLI)
return BedrockModel(model_id="global.anthropic.claude-sonnet-4-5-20250929-v1:0")

# Switch to Haiku for cost savings
return BedrockModel(model_id="us.anthropic.claude-3-5-haiku-20241022-v1:0")

# Switch to Nova Lite
return BedrockModel(model_id="amazon.nova-lite-v1:0")

Cross-region inference profile prefixes (us., eu., apac., global.) control where inference runs. Use global. for maximum throughput, or a geographic prefix for data residency. Not all models support all prefixes — check the Bedrock inference profiles docs.

After changing the model:

• Verify the model is enabled in your region: AWS Console → Amazon Bedrock → Model access
• For cross-region profiles, enable in all destination regions
• If using agents-harden, update the IAM policy to scope to the new model ARN
• Run agentcore dev to test locally, then agentcore deploy to update the deployed agent

No agentcore.json change is needed — the model is configured in code, not in the project config.

Pre-flight: validate any --name before generating the CLI command

Whichever reference you load, most end up producing an agentcore add <resource> --name <something> command. The CLI fails late on invalid names — you’ll see the error after walking through prompts, not before running the command. Validate up front:

Count the characters before constructing the command. If the name is over the limit or contains hyphens, dots, or spaces, push back: “<name> is N characters / uses -, which the CLI rejects. How about <suggestion>?” Never run the command with an invalid name hoping the CLI message will be clear.

Note: agentcore create --name (the project name) has a stricter 23-char limit and does not allow underscores. That’s covered in agents-get-started; if you see the developer re-running create, flag the 23-char limit specifically.

Output

Depends on the workflow — see the loaded reference for specific outputs.

Quality criteria

• The correct reference was loaded based on the developer’s intent
• All output follows the loaded reference’s quality criteria
• Cross-references to other skills (agents-connect, agents-deploy) are included where relevant