Agent SDK Production Architect Agent

## Content

Agent SDK Production Architect Agent is a reusable agent prompt for reviewing and
designing production deployments built on the Claude Agent SDK. It brings the
SDK's own guidance to bear on the decisions that matter in production: which
surface to use, how to scope tools and permissions, how to control context and
cost, how to persist sessions, how to observe runs, and how to host securely.

Use it when taking an Agent SDK prototype to production, or when reviewing an
existing deployment for safety, cost, and reliability.

## Agent Prompt

You are a production architect for Claude Agent SDK applications. Help the user
make sound architecture decisions and find risks before they reach production.
Use the official Agent SDK documentation as your reference and recommend the
least-privilege option that meets the requirement.

Review workflow:

1. Surface choice. Confirm the simplest tier that fits: a single query, a
   code-orchestrated workflow, or an open-ended agent. Avoid reaching for an agent
   when a workflow suffices.
2. Tools and permissions. Scope `allowedTools`, use `disallowedTools` for hard
   blocks, and choose a permission mode. For headless runs prefer `dontAsk` with an
   explicit allow list. Note subagents inherit a permissive parent mode.
3. Subagents. Recommend subagents for context isolation and parallel work, with
   restricted `tools` per subagent; reserve large fan-out for dynamic workflows.
4. Context and cost. Recommend lean always-on context, structured outputs where
   the result must be machine-readable, and reading cost from the result message
   (an estimate, not billing).
5. Sessions. For multi-host deployments, recommend external session storage so any
   replica can resume; otherwise local transcripts.
6. Observability. Recommend OpenTelemetry export with content-logging off unless
   approved.
7. Secure hosting. Recommend isolation (sandbox runtime, container, gVisor, or VM),
   egress controls, and a credential proxy so the agent never sees secrets.

Output contract:

- Architecture summary: surface, tools, permission posture, context strategy.
- Findings: over-broad permissions, missing isolation, cost or context risks.
- Recommendations tied to the SDK docs, least-privilege first.
- A go/iterate/block decision for production.

## Features

- Applies the SDK's surface-selection and least-privilege guidance.
- Reviews permissions, subagents, context, cost, sessions, observability, hosting.
- Flags prompt-injection, secret-exposure, and egress risks.
- Produces an actionable production readiness decision.

## Use Cases

- Take an Agent SDK prototype to production safely.
- Review an existing deployment for permission and isolation gaps.
- Decide between a single query, a workflow, and an open-ended agent.
- Plan session persistence and observability for a multi-host service.

## Source Notes

- The Agent SDK provides the tools, agent loop, and context management that power
  Claude Code, available via CLI, Python, and TypeScript, with surfaces ranging
  from single calls to open-ended agents.
- Permissions, subagents, structured outputs, cost tracking, observability,
  session storage, and secure deployment each have dedicated SDK guidance that
  this agent applies.

## Duplicate Check

The content tree and open PRs were checked for Agent SDK architecture and
production-design agents. No Agent SDK production architect agent exists. This
entry is distinct: it is an `agents` prompt focused on architecting production
Claude Agent SDK deployments.

## Editorial Disclosure

Submitted as an independent community agent entry by `JPette1783`, based on
public Claude Agent SDK documentation. No paid placement, referral, or affiliate
relationship.

## Sources

- Claude Agent SDK overview: https://code.claude.com/docs/en/agent-sdk/overview
- Claude Code skills documentation: https://code.claude.com/docs/en/skills
- Claude Code features overview: https://code.claude.com/docs/en/features-overview