Agent Design Review

Design or review agents by identifying the success contract first, mapping the real execution path second, and changing the smallest layer that is actually causing failures.

Load only what applies:

Need	Read
Choose architecture or multi-agent shape	`references/principles.md`
Rewrite prompts or improve cache reuse	`references/prompt-and-caching.md`
Draft or rewrite an actual system prompt	`references/system-prompt-templates.md`
Draft a provider-specific prompt for OpenAI Responses or Anthropic tool use	`references/provider-specific-templates.md`
Improve tool calling, tool schemas, or final outputs	`references/tool-and-schema-design.md`
Draft or fix actual tool schemas	`references/tool-schema-examples.md`
Review loops, approvals, side effects, or trust boundaries	`references/runtime-and-guardrails.md`
Build evals or decide how to iterate	`references/evals-and-iteration.md`
Review classifier, matcher, router, extractor, ranker, or moderation agent	`references/classifier-agents.md`
Need examples of strong and weak output	`references/review-examples.md`

agent-design-review

Agent Design Review