Claude Skills by hoplogic

github.com/hoplogic

17 skillsA× 13B× 42 installs24 views

TutorialsA**Goal**: take one of your own **natural-language skills** (a SKILL.md written in pure prose, also called a prose skill in English docs), translate it into a HopSpec specification (a **hopskill**), run it, and feel the difference between the two. **Prerequisite**: complete tutorial 1 (engine and skill installed). If you've read tutorials 03-05, your audit of the translation output (step 3) will go much deeper. > Note for Codex users: the translator `/hopbuild` is currently available only on t...ai-agentsgobashVotes: 0GitHub stars: 15
TutorialsA**目标**：拿一个你自己的**自然语言 skill**（纯散文写的 SKILL.md，英文文档中亦称 prose skill），翻译成 HopSpec 规约（**hopskill**），跑起来，体会两者的差别。 **前提**：完成教程 1（引擎和 skill 已装好）。读过 03-05 三篇的话，翻译产物的审计（第 3 步）会做得深得多。 > Codex 用户注：翻译器 hopbuild 目前仅 CC 载体提供（对 /hop 说"把 xxx 翻译成 spec"即分派到它）——在 CC 里完成翻译即可，产物 spec 两载体通用。ai-agentsgobashVotes: 0GitHub stars: 15
CodexARun an existing HopSpec file (reuse mode — Codex itself is the reasoning LLM via CLI, zero API key): confirm parameters, answer intervention points, present the final state by contract. Dispatched by $hop; users do not invoke it directly. Ad-hoc tasks without a spec go through $hop.ai-agentsgobashVotes: 0GitHub stars: 15
CodexBHopSpec turns a task into a structured spec (goal, steps, verification points) that the hopjit engine executes step by step — progress is on the books, checkpoints are enforced by the engine, and runs resume across sessions. This skill is the single entry point for four things: (1) complex tasks that need controllable, reliable execution (multi-step, progress worth keeping, needs verification, must resume after interruption) — upgrade them into a HopSpec run by the engine; one-line questions ...ai-agentspythongoVotes: 0GitHub stars: 15
DriverAHopSpec 是把一件活写成带目标、步骤、核验点的结构化规约,交给 hopjit 引擎按步执行：进度有账、把关由引擎强制、中断可续。/hop 是它的唯一入口,管四件事：①复杂的、需要可控可靠保障的任务（多步、要留进度、要核验、中断后要接得上）——升格为 HopSpec 交引擎管;一句话能答完的对话照常直接答,不进 hop。②跑一份现成的 HopSpec 文件。③构建、翻译、修订 HopSpec（把自然语言 skill 翻成 spec、改一份已有 spec 的几处）。④用户说"继续上次的活/接着干"这类接续话语时接上手头活的账。用户明说"走 MCP / standalone"时按独立模式跑（spec 在 hopjit server 进程里由它自己的 LLM 执行,本对话只做启动与介入）。ai-agentspythongoVotes: 0GitHub stars: 15
DriverB跑一份已经写好的 HopSpec 文件（复用模式——本对话当推理 LLM，hopjit CLI 管执行状态，零 API key）：参数确认、介入点问人、终态按契约呈现。由 /hop 分派调用，用户不直接调；现场的活不写 spec，走 /hop 主路。ai-agentspythonbashVotes: 0GitHub stars: 15
Skills DirectorySkills Directory Look up any of 103,619 scanned skills — grade, score, and every finding. Free.Check a skill
FixturesA每月对全部供应商执行对账审计，产出审计报告并把确认无误的付款单提交财务系统。ai-agentsVotes: 0GitHub stars: 15
FixturesA收到退款请求后核对订单信息；按金额分两档处理——200 元以下自动核验通过后直接退款（退款打出后不可撤回）；200 元及以上需客服主管人工批准后才退款。约束：任何退款打出前必须有对应档位的放行结论。ai-agentsVotes: 0GitHub stars: 15
FixturesA对收件箱里的每封未读邮件判断是否紧急；把紧急邮件的发件人和主题汇总成一份清单发送给值班群（发送后撤不回）。ai-agentsVotes: 0GitHub stars: 15
Multi SkillA把本周的周报文件整理归档：读取周报草稿，按 references/gate.md 的归档门禁核对齐全性，核对通过后把周报写入归档目录（写入后不可修改）。归档细节见 references/gate.md。ai-agentsVotes: 0GitHub stars: 15
FixturesA对收到的每个客户工单，分析其类别和紧急度；全部分类完成后汇总成处理计划；处理计划须经主管批准；批准后给每个高优先级工单的客户发送回复邮件。约束：处理计划必须覆盖所有工单。ai-agentsVotes: 0GitHub stars: 15
FixturesA汇总本周的项目进展记录草拟周报；核验周报必须覆盖所有进行中的项目且数字与记录一致，不达标就修订重来；对写多少细节拿不准时询问负责人偏好；核验通过后把周报发布到团队频道。ai-agentsVotes: 0GitHub stars: 15
HopbuildA由 /hop 分派调用（用户要把自然语言 skill 翻译成 HopSpec 时）；用户不直接调。把一个自然语言 skill 翻译成结构化的 HopSpec 规约（hopskill），执行纪律由 HopJIT 引擎结构化强制。 Trigger: "把 skill 转成 spec", "生成 hopskill", "skill 转 hopspec", "translate skill to hopspec", "hopskill 构建", "/hopbuild".ai-agentspythongoVotes: 0GitHub stars: 15
Hopbuild2A实验件——只在用户点名"hopbuild2"时由 /hop 分派调用，不自动选；日常翻译走 hopbuild。【实验性——日常翻译请用 /hopbuild；本 v2 仍在真机淬炼中，仅显式点名时使用】把一个自然语言 skill 翻译成结构化的 HopSpec 规约（hopskill）——递归等价分拆形态：判定序（顺序→分支→循环→原子）逐层消解 NL 节点，对齐门/终审门作者把关，产物带档位（hop=全结构化,双模式皆稳;mixed=含未尽原子——standalone 由引擎带工具面的 LLM 循环执行,所需引擎外能力须经 Tools 段声明,声明齐即可跑;复用模式 caller 能力面开放天然宽容）。 Trigger: "hopbuild2", "/hopbuild2", "用 hopbuild2 翻译", "递归分拆构建", "skill 转 hopspec v2", "构建 hopskill v2".ai-agentsnodeVotes: 0GitHub stars: 15
HopfixA由 /hop 分派调用（用户要改一份已有 HopSpec 的几处时）；整篇重写走 hopbuild。对既有 HopSpec 做定向修正——把修正意见（或 deep-validate 检查报告）变成节点级定向编辑：只改点名的地方，改完机械核对+呈人确认才写回（自动留快照可回滚）；结构性变更如实拒并指路重新翻译，不硬修。 Trigger: "hopfix", "/hopfix", "修一下这个 spec", "这个 spec 有几处要改", "按报告修正 spec", "定向修正".ai-agentsVotes: 0GitHub stars: 15
OpencodeBRun an existing HopSpec file (reuse mode — opencode itself is the reasoning LLM via CLI, zero API key): confirm parameters, answer intervention points, present the final state by contract. Dispatched by the hop skill; users do not invoke it directly. Ad-hoc tasks without a spec go through hop.ai-agentspythongoVotes: 0GitHub stars: 15
OpencodeBHopSpec turns a task into a structured spec (goal, steps, verification points) that the hopjit engine executes step by step — progress is on the books, checkpoints are enforced by the engine, and runs resume across sessions. This skill is the single entry point for four things: (1) complex tasks that need controllable, reliable execution (multi-step, progress worth keeping, needs verification, must resume after interruption) — upgrade them into a HopSpec run by the engine; one-line questions ...ai-agentspythongoVotes: 0GitHub stars: 15