加载中…
Craft reference for any text another LLM will read - a one-turn instruction to a subagent, a message that opens a long run, a prompt written into a file or code, a system prompt, a CLAUDE.md, a skill body, a hook. Use BEFORE writing any of them. Every text enters the reader's context at distance zero; what differs is the hold - how many turns after entry it must still be steering the reader. The hold picks the craft, not the kind of file and not the text's length. A one-turn hold wants terse, complete, said-once. A long hold keeps the body terse and hardens only the lines nothing will re-ask - boundaries, the stop condition, constraints that bite late. A whole-session hold against situations the writer cannot see buys redundancy, imagery, and rehearsed temptations. Reading the hold off the text's length or its filename is the known failure.
prompt 是 ChatHome Skills 收录的 技能 条目,来源为 GitHub Skill Search,质量分 58/100。 # Producing this artifact The craft for this medium lives in `references/craft.md`, relative to this skill's base directory. 1. Read `references/craft.md` befor. 作为 agent 技能,它把一组可复用的任务指令打包成 Codex 和 Claude Code 等 AI 编程 agent 可直接加载的格式,无需为该任务手写胶水代码。 该条目由 ChatHome Skills 的每日抓取从官方 registry 和 GitHub 发现查询中收录,每次运行都会重新核验仓库事实、新鲜度信号和安装元数据。 质量分综合了官方 registry 收录情况、可用安装命令、许可证完整性、仓库活跃度和 GitHub star 趋势,完整算法说明见来源页。安装命令已列于上方,可直接复制使用。 在将 prompt 加入生产 agent 工作流前,请核对上方的最近更新日期和仓库状态——下方事实快照是最近一次抓取的时点值。
codex skill install promptctl/laws运行此命令前,请先核对上游源码。
扫描范围: 安装元数据与已收录的文档正文。
基于安装元数据和有界技能正文的自动风险信号,不是完整代码安全审计,也不会递归审查源码或依赖。
# Producing this artifact The craft for this medium lives in `references/craft.md`, relative to this skill's base directory. 1. Read `references/craft.md` before writing anything.