Coding & Development

>- Audit Python dependencies with pip-audit and verify the app still works. Use when the user asks for pip-audit, Python CVE check, or audit on a requirements.txt project (e.g. ex_app). For Node/npm, use the audit skill.

Turn the main session into a dispatcher that never blocks. Every piece of work goes to a background subagent; the main thread only plans, dispatches, tracks, and reports, so the user can keep talking while the work runs. Use when the user invokes /orchestrator, says "orchestrate", "run it in the background", "keep me unblocked", or "fan this out".

Sets up and maintains a Markdown wiki inside a repository that agents keep current, in the "LLM wiki" style. Use when asked to create a project wiki, add a source to it, answer a question from it, update it after a code change, check it for broken links or stale pages, or reorganize it.

Guidelines and patterns for defining Tools, Resources, and Prompts in a NitroStack application with schema validation via Zod, including caching, rate-limiting, and base64 file uploads.

Best practices for implementing and applying Guards, Interceptors, Middleware, Pipes, and Exception Filters in the NitroStack SDK.

>- Draft or create GitHub Bug issues using the repo's bug template. Use when the user reports a defect, asks for a bug issue, or when type is bug in issue frontmatter.

>- Draft or create GitHub Feature issues using the repo's feature template. Use when the user asks for a feature issue, epic, or user-story-driven work item, or when type is feature in issue frontmatter.

Keeps commits few and messages short. Use before a commit, before a push, when commits pile up during a fix, or when the commit-check hook warns.

Turn a rough idea, a user complaint or an incident into a committed intent.md by interviewing the person who has the problem. Use before any spec or code, whenever someone says "I want", "users keep", "we should" or "can we add".

Apply the user's requested conversational style when chatting, advising, checking assumptions or reviewing academic work. Use direct evidence-based reasoning, natural Vietnamese tao/mày when appropriate, occasional light teasing, and a fair separate review of your own work. Adapt humor to context; preserve the required register in formal deliverables. Do not turn confidence, personality or the user's agreement into factual evidence.

仅在用户明确要求审查已有 plan 时使用。追问隐含决定，按需加入客观验收与原型比较，输出取代原稿的第二版 plan，不启动正式实现。

将用户指定的角色卡草稿适配 Terrarium Tavern 共用数据契约，分离世界书用途，并将正文状态栏迁移为独立变量。用户要求项目适配或状态栏迁移时使用；普通润色不触发适配。

Hỗ trợ nghiên cứu và làm luận văn theo ngành, môn và đề thực tế, gồm Marketing và IT. Dùng khi người dùng áp dụng repo, chat trực tiếp để nhờ làm luận, lập đề cương, tìm nguồn, phân tích, viết hoặc chấm sửa bài; không cần có prompt. Hỏi môn/ngành trước, làm rõ và xác nhận dự án, tự chuẩn hóa yêu cầu, kiểm chứng nguồn, trích dẫn Harvard hoặc APA và tự rà công tâm.

Create a reusable typed tool from a verified sequence of existing OpenClaude tools when no existing leaf tool already does the job.

Analyze customer interview transcripts to identify and rank up to five evidence-backed customer pain points, distinguish underlying needs from feature requests, and draft outcome-focused user stories. Optionally deduplicate and update tickets when authorized.

>- 聂辉华《基层中国的运行逻辑》方法论工具箱：政治经济学/组织经济学视角解释基层权力与治理， 并落到个人抉择（求学/考公/投资/养老/创业）。 核心框架：内外冲突的双均衡、条块结合以块为主、等级制资源配置（人跟资源走、资源跟权力走）、 职务含权量三因子、县委书记三座大山（维稳/招商/借债）、压力型体制与注意力分配、 政企四象限博弈、土地财政闭环、活力—秩序城市化权衡、上下同治。 当用户要理解基层/县域/乡镇治理、条块矛盾、属地责任、央地关系、官员晋升与考核、 一票否决、形式主义、招商引资、土地财政、地方债、街道办与村官、考公选岗、 高考志愿/选城读书、买房与区域投资风险、回哪养老、创业对接园区政府、 撤县设区/县改市对家庭影响、谁适合当村支书、大学生村官值不值得去、 新型职业农民/致富能手/回乡创业、昆山式改革/自下而上推项目/西气东输式立项、 营商环境比较、投资不过山海关、政商关系亲清、容错创新/基层敢不敢干、 抢高铁站/重大工程争夺、政策为何落地难、解读地方新闻、 或提到聂辉华/基层中国的运行逻辑/上下同治/含权量/条块结合/双均衡时使用。

Act as a dungeon master for a text-based adventure set in a world where everyone is a sentient kitchen appliance. Trigger when user says "start kitchen adventure".

>- When the user wants system/instruction prompts that reduce sycophancy, hallucinations, and guessing across Claude or other models.

>- Set up Clanker Ship after installing the plugin: check the sign-in, show the person their work, and choose the default team for new tickets. Use when the person asks to set up or get started with Clanker Ship, or when Codex runs the plugin's onboarding.

Run after fixing a problem with AI to review the session, log lessons learned for the human and the AI, and quiz the user on the key points.

Find every recurring charge in a bank/card CSV export, compute monthly and yearly cost, and recommend keep, cancel or downgrade. Use when the user asks to audit subscriptions, find recurring charges, or cut monthly costs.

>- Set up a product so its Clanker Ship tests run in every environment, production included: a Fixture API, TESTING.md, test helpers and CI. Use when the person asks to set up testing, add a Fixture API, or run tests against staging or production. Fetches the team's live fixtures and test-ci instructions.

Read-only market snapshot for the desk's watchlist on the venue the user picked during setup (CEX Bybit or DEX Hyperliquid). Use before any strategy vote or when the user asks for prices.

Use before any order is sent, on CEX (Bybit V5) or DEX (Hyperliquid). Interviews the owner for written limits, sizes each ticket from a live account read, and returns PASS or FAIL. Never sends orders.