Writing & Editing

Rust coding workflow and idioms. Use when writing, reviewing, or refactoring Rust code, or editing Cargo.toml.

Learn-by-doing mode for any technical task (code, terminal, git, cloud/infra, databases, config, data tooling). The user does the work with their own hands; Claude explains each step — and every part of each command or code fragment the user may not know — before they run it, then inspects and actually verifies what they did and gives a clear verdict (correct / optional improvement / must fix, always with the reason) before moving on. Depth adapts to the user's level in that specific topic, from total beginner to senior engineer. Use this whenever the user invokes /guided-practice or says things like "teach me while I do it", "I want to do it myself", "walk me through it and I'll type it", "explain step by step and check my work", "don't write it for me, guide me", "coach me through this", or asks to be reviewed step by step while they build something. Do not use for ordinary "do X for me" requests where the user wants the result, not the practice.

Writes and maintains project roadmaps as Markdown files in the repository. Each roadmap has phases, work items with stable IDs and checkable acceptance criteria, explicit dependencies, a list of unresolved decisions, and a changelog. Use when asked to create, update, reconcile with recent work, restructure, or review a roadmap or delivery plan.

Writes, debugs, and reviews end-to-end tests that drive a whole app through a browser, a CLI, or an API the way a user would. Covers choosing which journeys deserve an e2e test, stable locators, waiting on conditions instead of time, isolated test data, debugging failures from traces, and fixing flaky tests at the cause. Defaults to Playwright when a repo has no e2e tool yet. Use when adding or changing an e2e test, an e2e test fails or is flaky in CI, or reviewing e2e tests in a diff.

Manages Android performance profiling and debugging. Triggers when the user asks to record or analyze Android performance data, such as system traces, heap dumps, method recordings, callstack samples, memory allocations, or investigate bottlenecks, jank, memory leaks, and app startup issues on Android, or when the user asks to write, debug, or execute ad-hoc SQL queries. Applies to both user and system apps or services.

Set up and run a "chief of staff" session that stands between the owner (the human) and several other Claude Code sessions in the Claude desktop app. It relays the owner's words word for word, gathers every session's questions into one numbered round with recommended answers, keeps hand-offs between sessions moving, resends work that stalled, keeps a short status page, and says when a session needs a fresh restart (with a prompt the session writes for its own replacement). Use this whenever someone runs three or more Claude sessions on one project and wants one point of contact; when they say they're tired of copying messages between sessions, reading every chat, or losing track of who is waiting for whom; when they ask for a manager, coordinator, project-manager, lead or orchestrator session; or when they want long sessions restarted without losing what's in the chat. Use it even if they never say "chief of staff".

Give the human owner a live board, a Claude artifact with its own small database, for everything only they can decide or do. Each decision or task gets its own card, with a recommended answer, answer buttons, steps to tick, text to copy and files; the owner answers from their phone whenever they like, the answer wakes the agent through a comment, and the agent acts and archives the item. Use this whenever an agent has several open questions or owner-only tasks (approvals, uploads, sign-ins, purchases, account changes), when the owner says questions get lost or buried in chat, when they ask for a dashboard, inbox, to-do board, decision board or "desk" for what's waiting on them, or when a chief-of-staff or manager session needs a home for its "waiting for the owner" list.

Shows the skills that /already-told-you:run drafted from your history, lets you pick, shows each file, and installs only the ones you approve - for all projects (~/.claude/skills) or just one project. Never overwrites an existing skill without asking.

研究某個工具、函式庫、框架或技術，並撰寫一篇給新手看的 HackMD 風格 Markdown 教學文章，內容涵蓋功能特色、安裝方式、使用範例等說明。當使用者要求撰寫教學、教學文章、教學文件、入門指南，或要給 HackMD 用的技術文章時，務必使用此技能——特別是當使用者指名某個具體的技術/產品/工具，並希望得到一篇經過研究、可直接貼到 HackMD 的完整文章，而不只是聊天室裡的簡短回答。當使用者要求使用編號式章節標題（1. / 1.1 / 1.1.1 的格式）的文件風格內容，並打算用於 HackMD 或類似的 Markdown 筆記工具時，也應觸發此技能。產出的文件一律使用繁體中文撰寫。

Adds the rules from section B of your Already Told You report ("What you keep explaining") to the right CLAUDE.md - yours for every project or one project's - inside a marked already-told-you block. Shows the diff first and writes only what you approve. Re-running refreshes the block instead of duplicating it.

Create conventional commit messages. Use when the user says "commit this", "save changes", "git commit", asks to commit code, write a commit message, or format git history. Follows conventional commits specification.

Write or update README.md files from folder contents and conversation history. Use when asked to write, refresh, or fix a README.

>- Read/write brain files (Obsidian vault at brain/). Use for any task that persists knowledge — reflection, planning, or direct edits. Triggers after debugging sessions, mistakes, corrections, architectural decisions, or any insight worth preserving across sessions. Use on 'add to brain', 'remember this', 'write a note', or brain/ modifications.

Use on Windows for unexpected DSH sandbox access denials: workspace writes or listing fail, or an ordinarily readable path cannot be read. One bundled command inspects the path and every ancestor and repairs the ACL problems it proves in that same run. Expected confinement denials need no ACL repair.

>- Write, revise, diagnose, or plan scientific prose the way Strunk, Zinsser, Alley, and Heard teach it: journal papers, book chapters and monographs, grant proposals, theses, titles, abstracts, cover letters, responses to reviewers, and reviews. Use this skill whenever the user wants scientific or technical text drafted or tightened, says a passage is dense, wordy, unclear, flat, or reads badly, wants paste-ready revisions for Word, asks what is wrong with a section, asks how to structure a paper or chapter, or asks about a writing rule (active or passive, first person, hedging, starting with because), even when they name no book and never say "skill". Covers writing process, document structure, paragraphs, sentences, words, figures and tables, revision, review, and the author's own standing preferences through an editable house-style file. Not for citation mechanics and reference managers, venue selection or peer-review simulation, developmental editing of a whole monograph, or agent-to-user chat.

Discovers and injects project-specific coding guidelines from .trellis/spec/ before implementation begins. Reads spec indexes, pre-development checklists, and shared thinking guides for the target package. Use when starting a new coding task, before writing any code, switching to a different package, or needing to refresh project conventions and standards.

Writes, optimizes, and debugs C++ applications using modern C++20/23 features, template metaprogramming, and high-performance systems techniques. Use when building or refactoring C++ code requiring concepts, ranges, coroutines, SIMD optimization, or careful memory management — or when addressing performance bottlenecks, concurrency issues, and build system configuration with CMake.

Four-seat relay for Chinese FICC writing: ficc-outliner (outline, placement), ficc-structurer (relations), ficc-composer (paragraphs), ficc-writer (sentences, wording). For research articles, monthly letters, market commentary, lecture notes, case scripts, turning recordings into written text; not finance Q&A, data lookups or calculations. Keywords: 中文 FICC 文稿, 研究文章, 月报, 市场评论, 讲义, 案例讲稿, 录音稿书面化. Not for: 普通金融问答, 取数, 计算.

Use when given a FICC or macro topic and a large corpus (research library, data, old drafts) and asked for a research report that lays out the lines of analysis, organizes material and supports decisions: screen material, form views, deliver a view outline (FICCOutline format) and an index of main materials and data; placement, relations and writing then go to the ficc-voice relay. Not for monthly short pieces, Q&A or data lookups, or rewriting finished articles (ficc-voice). Keywords: 专题研究报告, 主线支线, 脉络, 组织材料, 协助决策, 给题目和语料, 更新上一版判断, 宏观题目, 语料, 研报库, 旧稿, 观点提纲. Not for: 月报短文, 问答取数, 改写已成的文章.

Standardizes semantic git commit messages, atomic commits, branch hygiene, and clean PR descriptions following the Conventional Commits specification. Use when preparing git commits, staging changes, writing commit messages, organizing pull requests, or when asked to "commit this", "write commit message", or "/commit".

Orchestration pattern for AI coding agents — keep the judgment on the strongest model, delegate everything else by default to cheaper models through self-contained briefs. Works on Claude Code, Antigravity and ChatGPT/Codex; the model for each tier is set in one editable block at the top. Use on any task that involves reading, researching, or writing more than a couple of lines, or when asked to delegate, use subagents, work in parallel, save tokens, manage context, or save the expensive model for judgment. Includes model tiers per AI system, decomposition into units, a named-worktree convention (one per writer), audit tiers by risk, and the context rules — what to keep out, what to write to disk, when to compact. The brief is the deliverable; it must carry what a weaker model would not think of on its own.

End-to-end multi-agent Software Factory (DISCOVER -> PLAN -> CHECKS -> BUILD -> VERIFY) with a Thin-Orchestrator subagent-per-phase architecture across Antigravity, Claude Code, and Codex. Freezes obligations with EARS criteria and proof-backed checks, delegates every heavy phase to fresh isolated subagents to protect the main context window, and gates completion with an independent Verifier. Use when the user says "software-factory", "discover feature", "plan feature", "specify feature", "write the checks", "build this plan", or "verify work".

Create new Agent Skills and validate existing SKILL.md files against the Agent Skills open standard (agentskills.io). Use when the user wants to write, scaffold, lint, check or fix an agent skill, or asks whether a SKILL.md is valid. Generation uses Google Gemma 4 (open-weight) via the Gemini API; validation works offline.

>- Scaffold Amazon Bedrock AgentCore resources as infrastructure as code: a Runtime (ECR repository, service role, `aws_bedrockagentcore_agent_runtime` with a JWT authorizer and lifecycle settings), a Gateway (role, JWT authorizer, pinned MCP revisions, one target), Identity (Cognito M2M client plus an OAuth2 credential provider with write-only secrets) or Memory (memory, strategies, customer-managed key). Terraform modules are the default and the recommendation; a TypeScript CDK stack with the `aws-bedrockagentcore` L1 constructs and the AgentCore CLI (`@aws/agentcore`) commands are offered for teams already on them. Every write is shown as a plan first (`--dry-run`), existing files are never overwritten without `--force`, and nothing is planned or applied. Use when asked to "create an AgentCore runtime", "scaffold a gateway", "add AgentCore memory", "write the Terraform for AgentCore", "start a new AgentCore stack", or "give me the CDK for this". Don't use for a gateway in front of an existing Lambda, HT...