What problem does it solve? Building a production-grade AI coding assistant involves far more than a model plus a tool-call loop. This Skill provides a Chinese-language pattern catalog for the harness around the agent — memory, skills, tool safety, context engineering, multi-agent coordination, and lifecycle extensibility — so engineers avoid the non-obvious failure modes that burn the most time. ## Core Features & Use Cases - Six Pattern Domains: Covers memory systems (layered instruction/auto/extracted memory), skill runtimes (lazy loading, budgeted discovery), tool registries and permission gates (fail-closed defaults, per-call concurrency classification), context engineering (select/write/compress/isolate), multi-agent orchestration (coordinator, fork, swarm), and lifecycle extensibility (hooks, task state machines, bootstrap sequencing). - Pitfall Guides: Each section lists counterintuitive failure modes, such as silent memory index truncation, memoized init hiding retry failures, and recursive fork cost explosion. - Use Case: An engineer designing a custom agent runtime reads the permission-gate pattern to implement a single authorization chokepoint with allow/deny/ask semantics before shipping any auto-approval mode. ## Quick Start Ask the agent to explain how to design a layered memory system with a bounded index and two-step save invariant for your coding assistant.
