ai-agent-skills

Monitor Codex CLI session status, detect timeouts, errors, and completion. Use when user asks about Codex session status, wants to check if Codex is still running, needs to debug a stalled session, or mentions "codex monitor", "session status", "codex 状态", "session 断了吗".

Manage long-running, multi-hour AI agent tasks using durable project memory — a structured 4-file system (SPEC, PLAN, RULES, STATUS) that prevents drift, ensures verification, and keeps work coherent across extended sessions. Use when the user asks to build something complex from scratch, run a multi-milestone project, delegate large autonomous coding work, set up a long-running agent session, create a project with 5+ milestones, or mentions "long horizon", "长任务", "长时间运行", "多里程碑", "自主编码", "autonomous coding", "durable memory". Based on OpenAI's Codex 25-hour design-tool experiment documented at https://developers.openai.com/cookbook/examples/codex/long_horizon_tasks.

双 Agent 全面审查修复框架。Antigravity + Codex MCP 独立审查 → 交叉验证 → 协作修复。内置 Codex session 监控、权限管控、双向状态追踪、清理策略。触发词："Ralph Loop"、"raphaloop"、"全面审查"、"零容错审查"、"/ralph-loop"。适用场景：代码审查、数据质量检查、发布前终验、跨 Agent 协作修复。