Skills
100,696 packaged capabilities and playbooks for agents. Install any of them with one command.
All
Vulnerability hunting gateway. Orchestration: identify targets and input surfaces → coverage-guided fuzzing → crash analysis → reverse-engineering and localization → report. Sub-skills: [[re-fuzzing]] [[re-crash-triage]] [[re-exploit]]. Trigger terms: vulnerability hunting, find vulnerabilities, fuzz, fuzz testing, AFL, libFuzzer, crash analysis, crash, vulnerability hunting, fuzzing
Turn supplied resources into practical reading notes: topics, interesting ideas, research leads, open questions, and homework todos. Use for articles, PDFs, docs, webpages, videos, transcripts, slides, meetings, books, RFCs, or talks. Do NOT use for code review or exhaustive transcripts.
Classify a batch of influencer replies into actionable categories (interested, negotiating, declined, needs info, ghosted) and generate a suggested next action for each. This skill should be used when triaging creator responses, classifying influencer replies, sorting outreach responses, categorizing creator DMs, reviewing batch replies, processing influencer inbox, prioritizing creator follow-ups, organizing outreach results, checking who replied to a campaign, figuring out which creators to follow up with, or cleaning up a messy outreach thread. For writing the initial outreach messages, see outreach-writer. For negotiating rates with creators, see rate-negotiation-playbook.
Build a standalone multi-page Markdown wiki package for a large software repository after reading and analyzing the whole codebase. Use when the agent is asked to create repository wiki documentation, onboarding docs, architecture guides, codebase tours, maintainer handbooks, or deep explanations of modules, core code paths, algorithms, design decisions, tradeoffs, tests, tooling, and operations for code learners or new maintainers. Optimized for large repositories with hundreds of thousands of lines of code where maintainers need broad coverage and deep subsystem documentation. The generated wiki must live in an independent package, usually `wiki/` or `maintainer-wiki/`, and must not be mixed into existing `docs/`, `website/`, or product documentation apps unless the user explicitly overrides this. Also supports optional Rspress/static documentation site setup when the user explicitly asks to publish or deploy the generated wiki. Supports Markdown output with Mermaid, Graphviz, and KaTeX where useful.
Reviews a technical design document or RFC adversarially in fresh-context read-only subagents, one per lens, (coverage, decisions, edge cases, consistency, risks, citations, scope) and returns one merged report with a verdict. Use when the user says "review this design doc", or when asked to review a design doc, RFC, or technical design. Not for code diffs; use reviewing-code.
When the user wants help with revenue operations, lead lifecycle management, or marketing-to-sales handoff processes. Also use when the user mentions 'RevOps,' 'revenue operations,' 'lead scoring,' 'lead routing,' 'MQL,' 'SQL,' 'pipeline stages,' 'deal desk,' 'CRM automation,' 'marketing-to-sales handoff,' 'data hygiene,' 'leads aren't getting to sales,' 'pipeline management,' 'lead qualification,' or 'when should marketing hand off to sales.' Use this for anything involving the systems and processes that connect marketing to revenue. For cold outreach emails, see cold-email. For email drip campaigns, see email-sequence. For pricing decisions, see pricing-strategy.
Analyzes and updates the `FROM` base images, Node headers, Go toolchain, and `rpms.lock.yaml` files in rhdh, rhdh-operator, and rhdh-must-gather; pins plugin-catalog `builder.Containerfile` / `.nvm/` / `konflux.additional-tags`; and bumps overlays `versions.json` `node` on `main` or a `release-1.10` branch. Use for weekly base-image maintenance, a UBI or RHEL bump, an RPM lockfile refresh, asking "which base images are out of date", node-v*-headers.tar.gz, catalog builder FROM, builder PipelineRun BASE_IMAGES, catalog release suffix, overlays Node version, UBI minor skew inside a Containerfile, or an OpenShift clients `N-el9-beta` / `4.21-el9-beta` RPM 404
Resolves the environment the other RHDH skills start from and hands back one JSON document: which RHDH repositories are checked out and where — rhdh, rhdh-operator, rhdh-plugins, rhdh-plugin-export-overlays, rhdh-plugin-catalog, rhdh-cli, rhdh-chart, rhdh-local, backstage and the rest — which tools are on PATH, and the target RHDH and Backstage versions with the source that produced them. Also owns the `rhdh` CLI behind workspace status, worklogs, and todos. Use for RHDH orientation, "where is my rhdh checkout", "which Backstage version goes with RHDH 1.10", `rhdh status`, `rhdh doctor`, `rhdh config`, `rhdh workspace`, `rhdh log`, `rhdh todo`, and for the read-only context another RHDH skill needs before it starts. Read-only — implementation work belongs to the domain skill that owns the requested outcome.
Migrate Riverpod from 2.0 to 3.0; automatic retry, paused listeners, legacy providers import, Ref simplification, FamilyNotifier removal, ProviderException, updateShouldNotify. Use when the user asks about Riverpod 3 migration, upgrading to Riverpod 3, or breaking changes in 3.0.
Set up the free Roblox game-creation skills on THIS machine - check which tools are installed (Python, Rojo, Wally, Lune, Selene, StyLua, Blender, the Roblox Studio MCP), walk the user through anything missing, then write each skill's personal LOCAL.md (folders, tool paths, rig file, VFX library location, env var names for keys) by asking the user and auto-detecting what it can, and finish with a smoke test. Use this the first time after installing the skills, whenever a skill says "LOCAL.md not found", when the user moves folders or installs a new tool, or when they ask "how do I set these skills up", "what do I need installed", or "why can't the skill find Blender/Rojo/my library".
Plans, creates, and optimizes digital marketing campaigns including content strategy, social media, email marketing, SEO, and AI visibility (GEO). Generates UTM parameters and tracking URLs. Helps develop go-to-market strategies, campaign messaging, content calendars, and performance measurement frameworks. Use when asked to "create a marketing campaign", "plan content strategy", "build email sequences", "set up UTM tracking", "analyze marketing metrics", "launch a product", "optimize for SEO", "improve AI visibility", "optimize for voice search", "set up IndexNow", "check accessibility for SEO", or "create social media previews". Supports both marketing beginners and experienced marketers needing templates.
Mines a session transcript, past agent sessions, or named sources such as PR review comments for durable learnings, then proposes approval-gated edits to skills, AGENTS.md, or docs and tracker issues. Use when the user says "retro", or when the user explicitly asks for a retrospective. Never infer from session end or friction.
Plan and, after explicit approval, execute runtime-behavior probes for local or live integrations. Use only when explicitly invoked to verify behavior that code review and normal tests cannot settle; define a controlled validation matrix, keep disposable JS probes outside tracked paths, and report observed evidence.
Review Rust changes on two axes at once — standards (idioms, ownership, errors, API surface, unsafe) and spec (does the change do what was asked) — with a Rust smell baseline layered on the repo lint configuration. Use when reviewing a Rust diff, pull request, or branch, when asked whether a change is ready to merge, or when a review needs to cover more than what clippy already reports.
Design Rust error types and panic policy — Result over unwrap, thiserror for libraries, anyhow for binaries, context that survives the call stack. Use when code calls unwrap or expect outside tests, when designing or refactoring an error enum, when choosing between thiserror and anyhow, or when the user asks how errors should be handled or propagated.
When the user wants to implement S&OP process, integrate demand and supply planning, balance business objectives, or facilitate executive planning meetings. Also use when the user mentions "S&OP," "IBP" (Integrated Business Planning), "demand-supply balancing," "executive S&OP," "monthly planning cycle," "consensus demand," "supply review," or "financial reconciliation." For detailed demand forecasts, see demand-forecasting. For capacity analysis, see capacity-planning.
Scene Feedback surveys for Decentraland SDK7 playtests: plan questions and timing, organize test rounds and find testers, integrate or debug the module, or analyze exported CSV ratings and comments. For audiovisual reactions to gameplay actions, use add-interactivity, audio-video or build-ui.
Transcribe a ScreenKite recording's mic audio, then remove BOTH silence gaps AND filler words (um, uh, ah, er, hmm) from the timeline in a single merged editTimeline cut. Use when the user asks to "clean up audio", "remove filler words", "remove ums and uhs", "cut dead air and fillers", or "full auto-cut" on a .skbundle project. Combines screenkite-transcription-cut silence removal with filler-word detection into one dry-run → confirm → apply flow. Do NOT use for B-roll overlay work (see use-screenkite-advanced-b-roll). Do NOT use when no .skbundle is open.
Turn a user-provided video into an Apple-style, scroll-driven website with smooth canvas video scrubbing, optimized WebP frame sequences, progressive loading, and responsive motion. Use for cinematic product reveals, landing pages, scrollytelling, image-sequence animation, and scroll-controlled video; also optimize generated frame assets.
【Internal suite skill · Users usually don’t need to call it directly in the full workflow; simply tell the orchestrator short-drama-director what you need, and it will route to this skill automatically】【漫剧老李 V7.5.0-SLIM suite · Subskill】Specialized engine for the asset-locking phase: extract three lists from the script (character appearance / props / settings) → produce according to material form (M0 appearance dossier paraphrase block / M1 image-generation prompts + image numbers + 《Material Patch Sheet》 · default / M2 batch image generation and locking), and manage spatial continuity of settings, single-instance locks for props, and crowd direction. This package does not hold images—the user generates them on external platforms; this engine must not request images or block downstream steps due to missing images. Trigger words: break down assets, asset extraction, asset catalog, image-generation prompts, character asset board, setting prompts, prop prompts, three lists, nine-grid setting image, character portrait, setting continuity, prop continuity errors, crowd direction, material form, patch sheet, I don’t have images. This skill is part of a suite; the complete workflow starts with the orchestrator short-drama-director.
[Internal suite skill · Users usually don’t need to invoke this directly in the full workflow; state your needs to the coordinator short-drama-director, which routes them here automatically] [漫剧老李 V7.5.0-SLIM suite · subskill] Director decision engine (storyboard selection criteria layer): situation mapping E01~E15 + 80/15/5 weighting mechanism + emotional anchor chart + A/B dual-state whitelist of composition pitfalls. Answers “what shot should this scene use?” and provides selection criteria for the storyboard engine. Triggers: situation mapping, what camera angle to use for this scene, emotional anchor, beat sheet, pitfalls, composition pitfalls, can we use a centered frontal shot, what shot should this use. This skill is part of the suite; enter the complete workflow through coordinator short-drama-director.
[Internal suite skill · Users usually don’t need to invoke this directly in the full workflow; state your needs to the coordinator short-drama-director, which routes them here automatically] [漫剧老李 V7.5.0-SLIM suite · subskill] Specialist engine for finished-video quality checks and continuity management: cross-shot/cross-episode consistency logs, editing transitions and Seedance 15-second segmentation gate, sound design templates, post-production music and sound design (BGM Cue Sheet/music search terms/AI music prompts/copyright checks), and automated output checks. **The consistency gate adjusts its judgment to the asset format** (M0/M1 compare against text records; only M2 performs image-level verification; this package does not hold images, so do not request images or make false verification claims). Triggers: review, quality check, continuity check, watch footage, shot-by-shot review, recheck after dialogue diagnosis, editing transitions, segmentation check, sound design, automated check, add BGM, post-production music, find music, Suno prompts, music copyright, asset format. This skill is part of the suite; enter the complete workflow through coordinator short-drama-director.
[Internal suite skill · Users usually don’t need to invoke this directly in the full workflow; state your needs to the coordinator short-drama-director, which routes them here automatically] [漫剧老李 V7.5.0-SLIM suite · subskill] Specialist engine for realistic live-action combat—designs action and produces Seedance 2.5/2.0 prompts for modern fighting, police chases, suspenseful assassinations, comedy fights, child-oriented fights, and small-group battles (1v2/1v3/2v2/3v3): R1 restrained realism mode, 11-link action dynamics chain, realistic combat move library (19 martial arts / 36 kicking techniques / 36 grappling techniques), three defensive states and large evasions, multi-person combat blocking, realistic force feedback, and compliance conversion. Triggers: live-action fighting, realistic combat, modern fighting, police chase, suspenseful assassination, comedy fighting, MMA, group battle, one against many, outnumbered fight. Suite members: coordinator short-drama-director (authority on system and feed formats) + sd-decision-engine (decisions) + sd-script-engine (scripts) + sd-asset-engine (assets) + sd-storyboard-engine (dramatic scene storyboards) + sd-combat-engine (general fight scenes) + sd-xuanhuan-engine (fantasy fight scenes) + sd-real-engine (realistic live-action · this skill) + sd-feed-engine (feeding) + sd-qc-engine (quality control).
【Internal suite skill · Users usually don’t need to call it directly in the full workflow; simply tell the orchestrator short-drama-director what you need, and it will route to this skill automatically】【漫剧老李 V7.5.0-SLIM suite · Subskill】Specialized engine for fantasy and xianxia fight scenes—generates fight-scene storyboards and prompts for fantasy cultivation, xianxia duels, supernatural powers, and high-energy Chinese animation: golden fight-scene granularity (2–3-second whole-second cuts, one action and one response per shot), two-way offense-defense power exchanges (preventing opponents from becoming punching bags), four spatial-closure protocols for finishing moves, a four-part scale kit and image geometry, 13 three-layer composite elements, a combined magic-and-martial-arts move library, and dual physical parameter libraries for realism and cultivation. Trigger words: fantasy combat, xianxia duels, spell clashes, ten thousand swords return to the origin, sword formations, magic formations, dharma forms, battles of gods and demons, ultimate finishing moves, xianxia cultivation. Suite members: orchestrator short-drama-director (authority for system locks and input format) + sd-decision-engine (decisions) + sd-script-engine (scripts) + sd-asset-engine (assets) + sd-storyboard-engine (dramatic-scene storyboards) + sd-combat-engine (general combat scenes) + sd-xuanhuan-engine (fantasy combat · this skill) + sd-real-engine (live-action realism) + sd-feed-engine (feeding) + sd-qc-engine (quality control).
The period feel comes from camera defects and one tiny everyday incident. Write the camcorder's autofocus hunting, exposure shifts and handheld shake as an explicit list, keep the story small, and the clip reads as a real old tape. Use when the user wants a Seedance 2.5 / 2.0 video prompt for this kind of clip (Early-2000s DV home video; 中文触发词: 早年 DV 家庭录像, DV, 家庭录像, VHS, 复古, 年代感), or asks how published cases of this type were prompted. Read references/cases.md before drafting.
One traveller moves through a place scene by scene, each scene carrying its own timecode, its own location and one short line. The polish is bought with film grain and golden-hour light. Use when the user wants a Seedance 2.5 / 2.0 video prompt for this kind of clip (Cinematic travel vlog montage; 中文触发词: 电影感旅行漫游, 旅拍, 城市漫游, 旅行), or asks how published cases of this type were prompted. Read references/cases.md before drafting
Self-improving skills toolkit that watches real agent sessions, detects missed triggers, grades execution quality, and improves skill packages through evals, replay, baselines, review, and post-deploy watch. Use when verifying or publishing a skill, improving instructions or routing, checking skill health, grading sessions, viewing the dashboard, ingesting agent histories, auditing installed skills, finding and temporarily using local skills for a task, managing the local Skill Library, scaffolding reusable workflow skills, or running autonomous improvement loops. Trigger whenever a user asks about skill performance, health, triggers, undertriggering, overtriggering, evolution, evaluation, or how their skills are doing, even if they do not name SelfTune explicitly.
Use when the user asks to "design an experiment", "build a predictive model", "run A/B test analysis", "perform causal inference", "engineer features", "evaluate model performance", "set up MLOps pipeline", "analyze time series", "calculate sample size", or "deploy a model to production". Expert data science covering statistical modeling, experimentation, causal inference, feature engineering, ML deployment, and advanced analytics with Python, R, and SQL.
New to MCP?
From what the protocol is to installing your first server, these answers cover most first questions.
An open protocol that defines how AI applications connect to external tools and data in a uniform way. Clients talk to the model; MCP servers provide tools, resources and prompt templates. They exchange standard messages, so one server can be reused by many clients.