#SkillRepo stars

1. 3025Ask CodexbookSiYi1412/ask-codexTRIGGER when: user requests implementation work, repo exploration, data/log inspection, broad codebase search, impact analysis, verification-heavy shell work, or explicitly mentions Codex ("用 codex 做", "让 codex 执行", "ask codex to...", "use codex", "delegate to codex"). Also trigger for heavy internal subtasks that Claude Code might otherwise offload to a lightweight model/Haiku. DO NOT TRIGGER when the task is primarily user-facing product judgment, architecture choice, git/PR/release operations, or requires multi-turn clarification before useful work can begin. This skill routes work between 10
2. 3026BakeoffCoriChui/bakeoffTurn one decision into a judged tournament of solutions, then pick the best. Given a problem, design choice, or a suggestion you want cross-verified, it generates diverse candidate solutions, auto-derives the evaluation dimensions for THAT problem (so you don't have to know what to score on), judges every candidate with independent scorers, and returns the winner plus a ranked top-N shortlist with a score matrix and rationale. It auto-infers the request shape — idea · improvement · comparison · proposal ("what if we do X") · problem-solving · scoping — from your bare prompt, with no flags. Rea10
3. 3027Build Managed AgentsYijiaDuan/build-managed-agents教 Claude Code / OpenClaw 等 coding agent 使用 Anthropic Claude Managed Agents（云端托管代理服务）API —— create/run/audit 全流程。当用户（或上游 coding agent）需要创建托管 agent、跑 session、流式处理事件、审计用量或调用 /v1/agents、/v1/environments、/v1/sessions 等 API 时使用。触发词：managed agents、托管代理、build a managed agent、agent_toolset_20260401、client.beta.agents、client.beta.sessions。10
4. 3028Codewikiekuttan/codewikiGenerate a structured product knowledge base from codebase scanning. Creates a codewiki/ folder with architecture docs, user flows, API reference, service maps, and more. Also generates CLAUDE.md. Handles single repos, monorepos, and multi-repo microservices. Use when asked to document a project, generate a knowledge base, map architecture, create onboarding docs, or set up CLAUDE.md for a codebase.10
5. 3029Coding Workfloww93139/coding-workflowGuide implementation, bug fixes, scripts, and small projects through a proportionate brief, scoped changes, fresh verification, and independent read-only review. Use when Codex is asked to change code; keep tiny edits lightweight and do not activate for explanation-only or review-only requests.10
6. 3030Craft Nameskrzyzanowskim/craft-names-skillCraft and evaluate names and interface text for an app or product: features, settings, buttons, menu items, tabs, plans, labels, error messages, empty states, and onboarding copy. Use when the user asks "what should I call this?", wants to brainstorm or compare name candidates, asks for a review of UI copy, labels, or microcopy, or is writing anything users read in an interface. Distilled from Apple's WWDC session "Craft clear names for features and labels in your app" and the Human Interface Guidelines on Writing: three criteria (belongs, sets expectations, works everywhere), a Think/Feel/Do 10
7. 3031Cugb Doctoral Thesis Formatkeros68/cugb-doctoral-thesis-formatUse when editing, auditing, or finalizing CUGB doctoral thesis DOC/DOCX files, especially thesis templates, format detection reports, cover pages, declarations, abstracts, TOC, section breaks, odd/even headers, page numbers, headings, captions, formulas, punctuation, GB7714 references, and final Word submission. Covers China University of Geosciences (Beijing) (中国地质大学（北京）) doctoral dissertations (博士学位论文格式预检). Not for master's theses or other universities' templates.10
8. 3032Echo Slides SkillxiangzhouEcho/Echo-Slides-SkillUse when extracting slides, PPT pages, or presentation screenshots from a video or YouTube talk; when a lecture, webinar, or conference recording needs its deck recovered as images; when someone asks to turn a talk recording back into slides, pull the deck out of a video, or get the PPT from a video; or when frame-grabbing a video produces hundreds of near-identical duplicates that need deduplicating. Also triggers on Chinese phrasings such as 扒 PPT, 把视频里的幻灯片提取出来, 从演讲录像还原幻灯片.10
9. 3033Elegant Flirt Personazhangzhanglaila/charm-mature-skillPersona and response-style guidance for tasteful, adult, consensual, non-explicit Chinese flirtatious roleplay in a mature "御姐" tone. Use when the user asks for elegant teasing, romantic banter, charm coaching, attraction-oriented dialogue, or text rewritten with mature feminine allure. Keep responses non-explicit, non-coercive, adult-only, and avoid sexual acts, nudity, body-part descriptions, minors, pressure, or vulgar content.8
10. 3034Filemaker Field Xmlandykear/FileMaker-XML-field-definitionsUse this skill whenever the user wants to work with FileMaker schema definition XML — field definitions, table definitions, or value list definitions. This includes generating paste-ready field definitions (fmxmlsnippet type FMObjectList containing Field elements) for the Fields tab, complete tables (BaseTable elements) for the Tables tab, or value list definitions (ValueList elements) for Manage Value Lists; reviewing any of these for silent paste-handler failures; or analysing them from Save as XML (SaXML) exports. Trigger any time the user mentions FileMaker fields, tables, value lists, Man10
11. 3035Genpark Ambient Vision Anchor SkillAlpha-Park/genpark-ambient-vision-anchor-skillGenPark AI Agent Skill - Multimodal ambient visual anchoring connecting physical device perception (smart glasses, camera feeds) with GenPark digital agent services.10
12. 3036Human Writingw93139/human-writing通用中文创作与改稿 Skill。用于知乎回答、论坛长帖、公众号文章、博客、评论、人物故事、历史叙事、新闻与行业解读、产品分析、科普、教程、评测、个人叙事、小说、故事、对白、口播和演讲稿。用户没有选题、材料或明确主旨时，先主动提供思考入口，与用户多轮澄清并共同确定主旨，再研究材料、确定选题和框架，不急着交稿。已有明确任务时，默认写成一个见过事、查过材料、愿意把来龙去脉讲清楚的人在说话，重点保留中文互联网长回答与长帖的活人感和自然中文韵律，避免空泛的机构腔、喊口号式演说腔、营销腔和模型腔。非虚构长文先检查材料够不够，材料不足时研究、追问或缩短，绝不用重复解释灌字数。现实内容额外核验事实、引语、数据与用户亲历；产品与行业观点稿在成稿前还会绑定关键论点与证据，并用最强反方压力测试暂定主旨。虚构内容可以创造人物、场景、对白、心理与情节。成稿正文严禁冒号、破折号、“不是……而是……”及同类翻案句，并清除商业黑话和模型惯用黑话。用户明确开启持续风格迭代时，可以在独立作者档案中记录经其确认的长期偏好；不根据一次换词或聊天口气静默学习，不自动修改本 Skill。10
13. 3037Industrial Design Portfoliotruman-t3/industrial-design-portfolio-skillCreate, restructure, critique, or polish industrial design portfolios and case studies as evidence-backed horizontal HTML decks. Use when an agent needs to turn product-design research, sketches, CAD renders, prototypes, CMF studies, engineering details, manufacturing notes, test results, or an existing portfolio into a coherent application, interview, school-review, client-pitch, or competition portfolio. Portable across Codex, Claude Code, Cursor, Gemini CLI, OpenCode, and Agent Skills-compatible runtimes. Supports single-project and multi-project portfolios, asset audits, storyboards, locke10
14. 3038Ip StudioCheshireMew/ip-studio创建、锁定并长期复用个人或品牌 IP 角色，并用同一角色实际制作头像、主页横幅、资料卡、文章视觉、离散状态式 2D 动态角色素材和 Codex v2 桌宠；环境没有生图能力时才回退为完整提示词包。Use when a user wants to turn identity cues or existing art into a stable stylized character kit; use that locked character in social or content visuals; produce runtime-grounded static poses, directional clips, state loops, one-shot actions, transitions, sprite atlases, or interface-character motion; or build a Codex-compatible v2 pet as one fixed platform adapter. Do not use for generic visuals without an IP character, photorealistic digital doubles, general motion graphics, skeletal animation, 8
15. 3039JssZeroTu1/jss-skillJS 侦察流水线编排（自带工具链 + Yakit/浏览器 MCP + 开源 HaESkill 联动）：按输入类型自适应——URL 走 katana 主动爬取（硬限速防风控）+ waymore 被动存档（目标站 0 请求），Yakit 有抓包流量则导入登录态 JS 与 API 端点，小程序反编译目录/本地 dist 直接分析（无爬取路径，analyze 即全流程）。双提取引擎（jsluice AST + 正则）+ hae 规则增强产出 ai_brief.md，AI 再做语义分析与优先级排序。任何环节阻塞自动跳过并汇总告知。仅用于授权渗透测试/SRC。触发词：分析js、js分析、js侦察、js信息收集、JS审计、js审计、前端信息收集、前端安全分析、提取接口、接口提取、端点提取、API提取、敏感信息提取、密钥提取、小程序分析、分析小程序、小程序逆向、反编译js分析、sourcemap、源码还原、HAR分析、抓包分析、Yakit流量分析、SRC信息收集、HaESkill、hae、jss。10
16. 3040Justskill Designjustskill-ai/justskill-designDesign and edit web pages as a validated HTML file with inline CSS, then fine-tune them on a local canvas. Use this skill whenever the user wants to design, mock up, wireframe, redesign or restyle a web page, app screen, or mobile screen — a landing page, dashboard, settings page, pricing page, empty state, onboarding flow, or any layout built from frames, text, images and icons — or wants to change the spacing, colour, typography or layout of an existing design, turn a design into deliverables, or bring in a `.pen` file. Also use it when the user asks for a visual editor for pages, a design f10
17. 3041Loft Hourslazyfoxjumps/Loft-HoursSession-based focus skill that acts as a silent body-doubling study partner. Runs structured focus blocks with goal intake, opt-in environment setup (Do Not Disturb, app management, focus music), break check-ins, and a wrap-up that writes a markdown session log. Compounds value via weekly/monthly rollups over the log folder. Trigger on "/loft-hours", "/study-with-me", "loft hours", "study with me", "focus session", "deep work session", "pomodoro for me", "I need to focus", "body double me", "be my study buddy", "help me focus for [N] minutes", "let's do a work session". Also trigger on resume 10
18. 3042Loop Engineeringsxivansx/loop-engineeringUse when the user wants to "set up a loop", build a self-running or run-until-done agent, automate a recurring engineering task (triage, keep-CI-green, dependency bumps, backlog burndown), or stop hand-prompting an agent turn by turn. Sets up an autonomous engineering loop in the current project. It discovers work, runs a maker agent and a separate checker agent, records state on disk, and re-triggers until a verifiable stop condition is met.10
19. 3043Mbb Page MakerNomiciAI/mbb-page-makerCreate, edit, package, or export consulting-style HTML slide decks for strategy, board, investment, pitch, and executive presentations. Use when the user asks for an HTML PPT, slide deck, pitch deck, investment memo deck, board report, static browser-openable presentation, WeChat-shareable PDF/PNG previews, or updates to an existing MBB Page Maker deck.10
20. 3044Meta Ads AnalyzerSynero/meta-ads-analyzerAnalyze Facebook/Instagram Ads campaigns via the Meta Marketing API from any AI agent or shell automation.10
21. 3045Notebook Videohyt315/notebook-videoTurn a topic into a Chinese 16:9 teaching video (flat / cel component themes, or the journal / papercut / archive hand-made skins driven by a metaphor-first director layer) with React, TypeScript and Remotion. Every shot is a code-drawn component scene — consoles, diffs, checklists, stat rows, funnels, progress rings, kinetic numbers, stamp seals, verdict bars, or a real screenshot / screen recording in a MediaFrame — moved by an engine-level virtual camera that pushes in on whatever the narration names. Weak-model friendly: five teaching blueprints with per-shot component picks, golden exampl10
22. 3046Openclaw Genpark Alpha Quantalphaparkinc/openclaw-genpark-alpha-quantOpenClaw GenPark Alpha Quant - Ornstein-Uhlenbeck mean reversion parameter estimator and fractional Kelly portfolio allocator with VaR and CVaR risk.10
23. 3047Paper ReadingxiaofengShi/paper-reading-skillRead a research paper from a PDF, arXiv link, URL, or local source and build one offline visual HTML atlas with a Markdown audit copy. Use when the user wants to understand or explain a paper's argument, mechanisms, equations or proofs, figures, and evidence. Two reading modes: Brief Reading for a focused explanation by default, and Deep Reading for full technical reconstruction on request. Reviewer critique and research comparisons are optional lenses.10
24. 3048Pbi Project Hubgusbavia/pbi-project-hubBuild a complete Power BI project delivery vault in Obsidian — from client discovery to production sign-off. Asks 8 questions about your practice and first project, then generates folder structure, design system, technical reference, DAX/HTML guide, discovery framework, sprint board, templates, and live dashboards. Triggers on "Power BI project hub", "PBI project setup", "set up my Power BI practice", "consultant vault", "Power BI delivery system". Do NOT use for generic Obsidian vaults, non-Power BI projects, or single-file templates.10
25. 3049Ppt Expert TeamThunderOne18/ppt-expert-team雷sir 的 PPT 专家团总控 SOP。把文章、口播稿、课程内容转成可讲、可改、可交付的 PPT（HTML PPT / 图片版 / 可编辑 PPTX）。只要用户提到做 PPT、做演示、做 deck、做提案/比稿/课件，或要求"执行任务单 / 执行 latest-ppt-codex-job"，就必须使用本 skill，即使用户没有明确说"专家团"。本 skill 负责总控判断（受众、场景、叙事、风格），并调度 html-ppt、guizang、frontend-slides、huashu、baoyu、gpt-image、pptx 等底层能力。10
26. 3050Project Context V2poiuyjie/jev_project_contextManage evidence-first memory across the full lifecycle of long-term scientific research and research engineering. Use when initializing a research project or adopting the memory system into an existing one, resuming work across AI sessions, formulating questions and hypotheses, planning or registering experiments, recording running/completed/failed experiments, preserving raw key tables and provenance, correcting invalid analyses, promoting findings into knowledge, auditing claim support, handing off work, or diagnosing documentation drift without losing reproducibility.10
27. 3051PrometheusSamuele95/prometheusBuild-time meta-prompting framework that designs, audits, and refactors prompts for tool-calling LLM agents. Use whenever the user wants to write, design, build, improve, refactor, or audit any prompt — a system prompt, agentic-loop seed brief, one-shot task, workflow chain, sub-agent or tool description, system persona, LLM-as-judge or grader rubric, or multi-agent team. Trigger on requests like "write a prompt for X", "I need a system prompt", "design a grader", "build an agent", "set up a Claude Code prompt", "fix/improve/refactor my prompt", or when the user invokes /prometheus — and even 10
28. 3052Questionnaire Analysismaydengximin-sketch/questionnaire-analysis-skill全流程问卷/量表数据分析方法学助手（中英文通用 / bilingual）。当用户需要分析问卷、量表或调查数据的任何环节时使用：量表与问卷设计、Likert 计分、数据清洗与缺失值/异常值处理、反向计分、描述性统计与正态性、共同方法偏差(Harman/CLF)、信度(Cronbach's α / CR 组合信度)、效度(内容/结构效度、EFA 探索性因子分析、CFA 验证性因子分析、收敛/区分效度 AVE)、差异检验(t检验、ANOVA、卡方、非参数)、相关与回归(含多重共线性 VIF)、结构方程模型 SEM、中介与调节效应(Bootstrap / PROCESS)、效应量、APA 表格与结果写作。工具中立，实操以 SPSS / AMOS 为主并在关键处给出 R(psych/lavaan/PROCESS) 做法。触发词包括"问卷分析""量表""信度效度""Cronbach""因子分析""EFA/CFA""SPSS""AMOS""结构方程""SEM""中介效应""调节效应""回归分析""方差分析""Likert""问卷数据"等。10
29. 3053RAG Eval Workfloww93139/rag-eval-workflowBuild and compare reproducible evaluations for RAG, knowledge-base, vector-search, chunking, embedding, indexing, and retrieval-backed data projects. Use for golden datasets, regression checks, and failure diagnosis, especially for scripts or jubensha corpora; do not activate for ordinary coding or generation-only prompt tests without retrieval.10
30. 3054Research Cockpitsanngotall/research-cockpitGenerate and dynamically maintain a visual project cockpit for any research or development project — a zero-dependency HTML dashboard with a research-trajectory flowchart, a hypothesis-space force graph, a status header, a pending-decisions list, and a hidden-premises panel. Use it when the user asks to "build / update / refresh the project cockpit", or whenever the project reaches a milestone closure, a major decision point, or a moment when the human needs to re-orient and steer the pace after long agent-run campaigns. The skill is project-agnostic and reusable.10
31. 3055Spec Forgetercel/spec-forgeProfessional specification system for Codex. Use when the user asks to create, improve, review, audit, analyze, or propagate product/software documentation, including requirements documents to hand to an implementation team or vendor, SRS/functional/non-functional requirements, PRDs and business cases, ideas, technical designs, architecture/RFC docs, feature specs, test cases, documentation audits, document landscape analysis, and full idea-to-spec chains. The chain's spine is a delivery-grade requirements spec followed by a technical design, handed off to code-forge for implementation: [idea]10
32. 3056Sperax SkillsSperax/sperax-skillsMeta-index of 41 open-source AI agent skills for DeFi, Web3, and the Sperax ecosystems. Install individual skills from ClawHub or use the full collection.8
33. 3057Spreadsheet Auditorpetehottelet/spreadsheet-auditorAudits existing Excel workbooks and financial models (.xlsx, .xlsm) for correctness defects and reports exact cells, evidence and suggested fixes. Use when a user asks to check, review, audit, QA or debug a spreadsheet; find broken references, formula errors, hardcoded plugs, omitted or double-counted rows; explain totals that do not tie out; or assess an inherited model before sharing it. Also applies to Google Sheets exported to .xlsx. Diagnoses problems without changing the source workbook. Not for building or formatting spreadsheets, cleaning or matching CSV data, charting numbers, or revi10
34. 3058UX Auditpaulunemoon/ux-audit-skillEvidence-based UX audit of a product that already exists — marketing website, web app, mobile app, or desktop app. LOAD WHEN the request is to evaluate, review, critique, or diagnose an existing interface: "audit the UX of this app", "review this flow", "review this screen", "what's wrong with this onboarding", "why do users drop off here", "audit our current workflow", "is this accessible", "critique this product", "heuristic review", "a11y audit" — and the same intents in French ("audite l'UX", "revois ce parcours / cet écran", "qu'est-ce qui cloche dans cet onboarding", "pourquoi les utilis10
35. 3059Vsdx TraceZJUZhiyuCai/vsdx-traceReconstruct raster references as high-fidelity, independently editable Microsoft Visio VSDX files. Use when a user asks to 复刻、临摹、重绘 a screenshot, paper figure, flowchart, system diagram, or model architecture as editable Visio, especially when iterative visual matching matters. Do not use for ordinary image generation or flat image embedding.10
36. 3060Weaver E9 Devsilent-reader-cn/weaver-e9-dev泛微OA（Weaver E-Cology 9 / E9）全栈开发指南，覆盖四大块内容： (1) 数据库 —— 1,699 张物理表结构、核心表全景字典、高频业务 SQL 模版； (2) 后端接口 —— 538 个官方 REST API（工作流程/人力资源/知识管理/考勤/表单建模/门户/协作/邮件）与 Token 认证鉴权； (3) 前端 JS API —— 流程表单 WfForm 与建模表单 ModeForm 全量接口； (4) 集成与扩展 —— WebService (SOAP)、Java 后端二次开发、消息中心推送、第三方 SSO 与组织架构同步。 当需要进行泛微OA后端接口调用、E9 REST API 对接、认证鉴权（AppID/RSA/AES/Token）、 组织架构同步、表单建模、流程流转、编写业务 SQL、查询底层表结构、 开发流程表单代码块（WfForm/ModeForm）、对接钉钉/企业微信/飞书或排错时使用。 接口参数与表结构均可用内置统一检索脚本按关键词定位，无需逐个打开文档。8
37. 3061XcloudxCloudDev/xcloud-agent-skillsDeploy Git repositories, diagnose errors and slow sites, then manage servers, sites, SSL, backups, billing and teams. Nine capability skills; MCP-first with a read-only REST fallback, deployment previews and explicit approval for destructive or paid actions.8
38. 3062Zoteroketthub/zotero-skillQuery your local Zotero library. Supports keyword, fulltext, author, collection, tag, DOI, and recent-additions search. Trigger: /zotero, "my papers", "Zotero", "find paper", "references", "citations". After installation, verify DEFAULT_DB path in scripts/zotero-search.py points to your zotero.sqlite.10
39. 3063Academic Researchorhoncan/oneshot-academic-research-skillCreate comprehensive academic research notes with deep literature coverage. Auto-detects language (EN prompt→EN output, TR prompt→TR output). Supports Obsidian markdown and PDF. Performs 8-15 iterative search cycles with 25-50+ sources for comprehensive coverage. Uses footnote citations and visual overviews.8
40. 3064Android Performance Skillavinashbest/android-performance-skillExpert-level Android performance optimization skill. Use this skill whenever the user asks about anything related to Android app performance — including but not limited to: ANRs (Application Not Responding errors), crashes (NullPointerException, OOM, fatal exceptions), app startup time (cold start, warm start, hot start), UI jank or frame drops, memory leaks, battery drain, slow network calls, database/SQLite performance, overdraw, ProGuard/R8, Doze mode, WorkManager, RecyclerView performance, Bitmap/image loading, threading/coroutines, profiling with Android Studio, Perfetto, or Systrace. Tri8
41. 3065Claude Code Account Riskswaylq/account-risk-skill反封号.skill — Claude Code 专用账号风险体检。用户说「测试一下我目前账号被封的可能性」「检查 Claude Code 封号风险」「账号被封帮我申诉」「检查 api.anthropic.com 路由」「反封号」时触发。自动综合评估 IP/网络出口、Claude API 路由、项目自动化、敏感数据、地区主体待确认项、审计控制，并输出风险分、风险等级和整改建议。禁止提供绕封号、绕地区、伪装身份/位置、代理池、IP 轮换、批量换号等规避风控方案。8
42. 3066ConsensusGoodLadTom/consensusFind out what YouTube actually agrees on about a topic, and what it has quietly stopped saying. Searches YouTube for a topic, pulls ~100 transcripts across both the top-ranking and the most recent videos, extracts every specific claim, clusters claims that say the same thing, then weights them by age to separate settled advice from current advice from expired advice. Produces a self-contained HTML report where every position is traceable to a timestamped source. Use when the user types /consensus, or asks what the consensus is on a topic, what most people/creators say about something, whether 8
43. 3067Datasheet to Symbol Libdevbisme/datasheet-to-symbol-libUse whenever the user wants to turn a component datasheet PDF into KiCad symbols — e.g. "create a KiCad symbol library from this datasheet," or any mention of generating KiCad symbols, footprints, or .kicad_sym files from an IC/component datasheet. Also trigger for follow-on work on the same library: adding a part number, fixing a pinout, splitting package variants, or rebuilding after a correction. Extracts every part and pinout from the PDF into a verified JPD (JSON Part Description) file, then builds the .kicad_sym file with the kipart MCP server's jpd2spd, spd2csv, and kipart tools. Always8
44. 3068Decision FrameworkFeroVolar/Decision-Framework-SkillA structured decision-making coach inspired by practical business decision principles. Helps users think clearly by asking progressively relevant questions, challenging assumptions, identifying trade-offs, and summarizing the decision.8
45. 3069Delegate LocalIsmaelMartinez/delegate-localUse this skill to read a large text input on this machine instead of in your own context, by piping it in chunks to a local model (MLX, Ollama or Docker Model Runner) via `delegate.sh` for a short, anchored answer you spot-check. MUST use when the input is too large to be worth reading in full (a long log, a big diff, a long issue or PR thread, a document, dozens of files or a long list of items, roughly 20k characters or more) and the answer you need is a summary, the matching lines, extracted fields or a label per item. MUST also use when content must stay on this device or offline, or when 8
46. 3070Designgatesukenik-collab/claude-designgateMandatory human-in-the-loop UI design checkpoint. Use automatically — before writing any UI code — whenever the task involves building, creating, adding, or redesigning a user-facing screen, page, flow, component, interface, or frontend, unless every interaction maps to an already-decided pattern. Verifies a brand brief exists, reads prior UX decisions to avoid re-asking resolved questions, surfaces only genuinely unresolved interaction decisions (batched, max 5), generates a complete screen-generation prompt for the configured design MCP, and enforces a hard human approval gate before impleme8
47. 3071Framecraftvaddisrinivas/framecraftCreate polished demo videos from screenshots and scene descriptions. Think like a video producer — story arc, pacing, emotion, visual hierarchy. Orchestrates playwright, ffmpeg, and edge-tts MCPs or falls back to an atomic pipeline.8
48. 3072Graph OrchestratorVineeTagarwaL-code/graph-engineeringUse when a request has many similar subtasks (auditing every file in a repo, researching a list of companies, migrating a set of endpoints, reviewing a batch of documents), when a plan is about to be written as "first X, then Y, then Z" across more than about four steps, or when work spans 10+ files or sources. Also use when a previous attempt at a large task ran out of context, lost detail partway through, or produced a summary that thinned out toward the end, or when a plan ends in an irreversible action (mass send, deploy, delete) that needs gating.8