A self-improving agent skill pipeline that prevents common AI coding agent failures. Includes 6 skills (plan-interview, intent-framed-agent, context-surfing, simplify-and-harden, self-improvement, agent-teams) and 5 subagents (simplify-auditor, harden-auditor, spec-auditor, context-monitor, self-improvement-logger). The pipeline catches scope drift, context degradation, security gaps, and repeated mistakes — each skill targets a specific failure mode and hands off cleanly to the next.