Agent design

How to write the AGENTS.md, skills, and tool descriptions that make your agent actually behave the way you want. Prescriptive content rather than reference — what patterns work, what failure modes to watch for, how to make the model use subagents and agentic_* wrappers efficiently.

The reference material these pages draw from lives in Concepts — Skills, Permissions, Context management.

System instructionsAGENTS.md patterns — role framing, do/don't lists, model-specific quirks. Includes the post-checkpoint loop case study from v2.0.

SkillsSKILL.md design — when to write a skill vs. an AGENTS.md rule, how to write a description that triggers, body structure, composability.

Subagents and wrappersGetting the model to actually use agentic_* wrappers and background subagents. The verify-with-bare-tool failure mode. Choreography patterns.

Cost efficiencyWhat moves the cost needle — model selection, context management compounding, prompt caching, output shape. Includes a decision tree.

Reading order

If you’re new to core-agent, walk these in order — each builds on the prior. Operators who’ve used core-agent for a while can pick whichever covers the specific friction they’re hitting; the pages cross-link generously.