Claude Skills
The open format is called Agent Skills and works in Claude Code, Codex, Cursor and other agents — most people know it as Claude Skills.
Every Agent Skill we could find on GitHub, deduplicated by content. 79 566 files from 1 758 authors, of which 61 913 are unique — the rest is the same skill repackaged into someone else's repository. For each one: what it weighs in tokens, whether it ships runnable scripts, and which MCP servers it needs.
Builds and refreshes domain-based index files for the Vaultr knowledge base. Use this skill whenever the user wants to build, rebuild, refresh, or update the knowledge index, create domain indexes, organize knowledge by domain, or index the knowledge base. Triggers on phrases like 'build knowledge index', 'rebuild index', 'refresh knowledge index', 'update domain indexes', 'index my knowledge base', 'regenerate index', or any request to create or maintain the knowledge index files.
Extract high-value verbatim quotes from a Vaultr note and save each as a short note with a backlink wikilink. Use this skill whenever the user wants to extract key quotes, highlight important passages, or "划重点" from a vault note. Trigger on phrases like "extract quotes", "highlight this note", "save important quotes from", "提取引文", "划重点", "摘录重点", or any request to pull out notable original text from a note and store it.
Extract and update personal memories from Vaultr notes into structured memory files. Use when the user wants to update their personal memory, extract memories from notes, run memory extraction, or refresh the personal memory base. Triggers on phrases like 'update my memory', 'extract memories from notes', 'run memory extraction', 'refresh personal memory', or any request to build or maintain a personal memory base from notes.
Use when a user needs通用中文参考文献、GB/T 7714-style bibliography entries, Chinese academic reference formatting, or BibTeX completion from Chinese or English literature titles, including 根据文献名补全参考文献、生成中文论文参考文献格式、format Chinese academic references.
Quick-understanding doc for ML/DL, AI4Science, and research codebases — task type, stack, architecture, and figure-worthy innovations. Use when the user wants repo analysis, 仓库分析, or to understand a codebase before figure planning.
Plan figures for an academic paper or document — map sections to figure types, count, priority, and aspect ratios. Use this skill whenever the user provides paper text, a PDF, an outline, section drafts, or asks what figures their paper needs, including "论文需要哪些图", "论文配图规划", "paper figure planning", "which figures should my paper have", "帮我规划配图". Produces a Figure Plan with must/strong/nice priorities and per-figure visual elements.
Generate publication-ready figure prompts in the modern pastel/airy ICLR/NeurIPS style with soft panels, tokens, pills, and rounded type. Use this skill whenever the user wants a pastel, airy, soft, or modern ML figure — including "pastel风格", "ICLR那种", "现代柔彩", "airy figure prompt", "token flow diagram", "2024-2025 conference style". Produces a layered English image prompt with P1/P2/P3 color schemes. For classic box-border CVPR/Nature diagrams, route to academic-figure-prompt instead.
Generate publication-ready figure prompts for image models (GPT-Image-2, Gemini NanoBanana, etc.) in the classic academic style. Use this skill whenever the user wants a box-border architecture diagram, framework/network/module figure, or JSON figure spec — including "生成框架图", "画架构图", "JSON配图规范", "academic figure prompt", "框架图JSON". Produces a structured JSON spec AND a 200-400 word English image prompt with icons, dimension labels, and panel grouping. For pastel/airy ICLR-style figures, route to academic-figure-prompt-pastel instead.
Make palette decisions for academic figures — choose between classic and pastel style families, then recommend a colorblind-safe scheme with exact hex values based on venue, domain, figure type, and module count. Use this skill whenever the user asks about colors, palettes, style for a figure, including "用什么配色", "推荐配色", "what palette for NeurIPS", "Nature Blue", "classic vs pastel", "色盲友好配色", or any color-related question about academic diagrams.
Extract and analyze architecture diagrams from academic PDFs or existing figure images. Use this skill whenever the user provides a PDF or image containing an architecture diagram and wants to extract, analyze, or redraw it — including "从PDF提取架构图", "架构图分析", "extract figures from pdf", "architecture diagram extraction", "analyze this diagram". Runs a local PDF extraction script and produces structured redraw parameters for downstream prompt skills.
Entry-point assistant for academic figure generation. Use this skill whenever the user wants to generate academic figures, paper diagrams, or architecture visualizations — including "帮我画图", "从仓库到配图走一遍", "完整论文配图工作流", "帮我分析这个仓库然后出图", or any end-to-end request from code/paper to figure. The assistant analyzes the input, presents a Figure Plan for user confirmation, then generates colors, prompts, and (when an image model is available) the final image. Routes repo-first (code) or paper-first (document/PDF) inputs.
产品宣传片制作总控 skill pack。用于从产品说明、官网、应用截图或 GitHub 仓库制作 60-90 秒宣传视频，按阶段完成 brief、storyboard、素材、HyperFrames 剪辑、BGM 设计和交付。当用户要做宣传片、产品视频、项目介绍视频、launch video、开源项目 promo、BGM 卡点或真实软件界面宣传片时使用。
宣传片两包素材生产。Pack A 使用当前可用图片生成能力生成产品图、概念图、UI 模拟图或风格补充图；Pack B 使用官网、应用截图、浏览器搜索、GitHub 或开源素材收集真实产品信号。当 storyboard 确认后进入。
产品宣传片创意简报。输入产品说明或 GitHub 链接，自动抓取产品信息，提炼核心卖点，推荐视觉风格和叙事结构。当用户提到"宣传片""promo""产品视频""做个视频介绍""项目宣传"时触发。
宣传片逐镜头分镜脚本。根据创意简报生成每个 Shot 的 7 维画面描述和详细 HyperFrames 提示词，是整条视频质量的关键。当用户确认 brief 后自动进入。
产品宣传片制作总控流程。串联 6 个 Skills，按阶段执行并暂停确认。从产品 URL 到带 BGM 成片 MP4 的全自动流水线。当用户提到"做个宣传片""promo""产品视频""项目介绍视频"时触发。
使用 HyperFrames 将所有素材按分镜时间码串联，添加文字层、动画、转场，渲染成完整宣传片 MP4。这是整个流程的核心渲染引擎。当素材确认后自动进入。
为产品宣传片生成高度贴合画面节奏的 BGM 方案与音乐生成 Prompt。读取 brief、storyboard、EDL、DESIGN 或成片，输出音乐风格、BPM、情绪曲线、卡点表、Mureka/Skywork Music Maker 英文 prompt、负面 prompt 和剪辑建议。当用户要“配乐”“BGM”“背景音乐”“音乐 prompt”“卡点”“按转场做音乐”时触发。
Deterministic fixture lane for the business-ops graph example.
Audit a sealed runx receipt for governance, comparing the authority a run exercised against what it was granted, and flag over-reach, ungated mutation, unrecorded refusals, or exposed secret material.
Governed runtime for agent skills: discover and install portable skills, run bounded skill graphs with explicit authority, and inspect signed receipts for what happened.
Answer one question strictly from a small, caller-supplied documentation corpus, with exact supporting quotations or an explicit account of what the corpus cannot answer.
Prepare a provider-side paid-call challenge and exact credential-verification handoff; forwarding requires a real settlement adapter.
Route one business signal through a replayable governed ops graph: classify, docs, release, work, outreach, spend, and proof, with consequential actions stopping at the right gate.
Run a standing team with a mandate, advanced one governed case-turn at a time: a fixed roster, a persistent objective, a multi-turn case, member dispatch under a scoped grant, escalation gates, a measurable done-check, and a sealed receipt trail.
Build a scoped brand voice packet from source material so downstream agents can write, review, and adapt content without inventing brand claims.
Turn governed source evidence into a citation-bound reader-facing draft, channel package, and provider-neutral publication handoff.
Audit exact npm dependency versions against OSV through Runx native HTTP and emit replay-verified evidence with no unverified findings.
Bind a provider price to a deterministic replay-safe challenge natively.
Prepare an exact provider-verifier request from an opaque credential reference natively.
Derive a provider-side price and bounded payment request natively.
Convert bounded mailbox and calendar evidence into a reviewable executive action packet.
Produce a decision-ready deep-research brief from bounded governed evidence and preserve every material source binding.
Govern provider-agnostic data reads and state transitions through declared data-source operations, not model-authored raw queries.
Prepare and explicitly file an evidence-bound payment dispute response using native receipt proof, approval, provider execution, and independent readback.
Produce a decision-ready ecosystem briefing from fresh governed evidence without claiming publication.
Turn validated evidence and operator intent into publication-ready drafts, deterministic channel packages, and provider-neutral handoffs.
Read and normalize Google Search Console performance, indexing, site, and sitemap evidence through any compatible Runx connector; plan and govern sitemap submission without coupling the skill to credential custody or a connector vendor.
Fetch one allowlisted external source, scrub personal data, obtain human authorization for the exact safe plan, and seal a provider-neutral outbound handoff without claiming delivery. Use when external content must cross a trust boundary through a separate delivery adapter.
Read and normalize GA4 properties, metadata, standard reports, and realtime reports through any compatible Runx connector, preserving the measurement caveats an operator needs before turning analytics into action.
Extract schema-validated JSON from messy HTML or text fixtures with digest-bound provenance.
Plan or execute a scoped GitHub issue, thread, or pull-request synchronization through a configured Runx Connect grant, with approval and readback on writes.
Classify a bounded support request, choose the safe next path, and draft a customer-ready reply only when a human-gated send is appropriate.
Advance one declared incident through a fixed roster, approval-bound communications planning, and receipt-backed resolution while canonical agency owns durable state.
Turn a noisy inbound request into a bounded intake artifact and an explicit next runx lane.
Govern a scafld-backed issue-to-PR lane with native scafld review and handoff surfaces.
Publish issue-to-PR outbox entries through the governed Rust thread-outbox-provider front.
Run existing scafld v2 lifecycle commands under runx governance.
Read, analyze, and draft high-signal GitHub issue responses from bounded provider evidence without silently mutating the repository.
Internal GitHub provider boundary for a prepared issue-to-PR outbox push.
Route one question or source event through a supplied knowledge catalog to validated source, owner, escalation, and follow-up skill references. Use for evidence discovery and ownership routing, not broad business-action fanout or answering the question itself.
Enrich a lead from supplied account signals and produce a consent-aware, evidence-bound outreach recommendation.
Route a validated lead to direct outreach planning, nurture planning, or a recorded hold, preserving why that path was chosen.
Turn one bounded meeting transcript into evidence-bound decisions, action items, a reviewable follow-up message, and provider-neutral task proposals without inventing owners, dates, approval, or live effects.
Verify the receipt ids behind a normalized authority-usage summary, compare that evidence with granted scopes, and propose the narrowest grant the evidence supports.
Govern a bounty-style messageboard from post through moderation, claim, delivery, acceptance, payout authorization, and trial take evidence.
Answer a cross-run audit question against the receipt ledger, returning matched receipts and a chain-verification result.
Route the mock provider-charge plan through canonical charge.
Route the mock refund plan through canonical receipt- and authority-bound refund.
Route the mock payment rail through canonical spend authority and finality.
Claude Skills — questions
Answers built from the skills we actually parsed.