Skills
100,696 packaged capabilities and playbooks for agents. Install any of them with one command.
All
Guide for adding new LLM models to Letta Code. Use when the user wants to add support for a new model, needs to know valid model handles, or wants to update the model configuration. Covers models.json configuration, CI test matrix, and handle validation.
Use when the user is addressing pull request review comments locally and asks for help evaluating, implementing, or drafting responses to reviewer feedback - requires technical rigor and verification, not performative agreement or blind implementation
Google Ads deep analysis covering Search, Performance Max, Display, YouTube, and Demand Gen campaigns. Evaluates 80 checks across conversion tracking, wasted spend, account structure, keywords, ads, and settings. Use when user says Google Ads, Google PPC, search ads, PMax, Performance Max, or Google campaign.
Landing page quality assessment for paid advertising campaigns. Evaluates message match, page speed, mobile experience, trust signals, form optimization, and conversion rate potential. Use when user says landing page, post-click experience, landing page audit, conversion rate, or landing page optimization.
Product photography enhancement for ad creatives using banana-claude image generation. Takes a product image and generates 5 professional photography styles for ad use: Studio, Floating, Ingredient, In Use, and Lifestyle. Requires banana-claude (v1.4.1+) with nanobanana-mcp. Triggers on: product photo, product photography, photoshoot, enhance product image, product shoot, product photos for ads, generate product photos, studio shot, lifestyle photo.
Strategic paid advertising planning with industry-specific templates. Covers platform selection, campaign architecture, budget planning, creative strategy, and phased implementation roadmap. Use when user says ad plan, ad strategy, campaign planning, media plan, PPC strategy, or advertising plan.
TikTok Ads deep analysis covering creative quality, tracking, bidding, campaign structure, and TikTok Shop. Evaluates 28 checks with emphasis on creative-first strategy, safe zone compliance, and Smart+ campaigns. Use when user says TikTok Ads, TikTok marketing, TikTok Shop, Spark Ads, Smart+, or TikTok campaign.
Govern an AG2 multi-agent network — identity (`Passport`, `Resume`), per-agent `Rule` with two top-level blocks `access` (`AccessBlock`) and `limits` (`LimitsBlock`, which nests `RateBlock` + `InboxBlock`), the swappable `HubArbiter` / `RuleBasedArbiter` access-&-routing seam, `AuthAdapter` / `AuthRegistry` registration, channel-level `Expectation`s with `audit` / `notify_channel` / `auto_close` violation handlers, the hub's append-only audit log and `AUDIT_KIND_*` constants, live `HubListener` / `BaseHubListener` observability plus `Hub` `on_*` hooks and `register_sweeper`, and task observation via `agent.task(...)` + `TaskMirror` (updates `Resume.observed` for peer ranking). Use when the user needs rate limits, access policy, SLAs, compliance trails, live metrics/alerting, capability-driven peer ranking, or to inspect what actually happened on the network. Load this after `ag2-network-quickstart`. For the agent-side surface (custom handlers, views, LLM tools, `HumanClient`) see `ag2-network-tools-and-views`.
Shape what an AG2 network agent perceives and which actions its LLM can take. Covers the five auto-injected LLM-facing tools that ship via `NetworkPlugin` (`delegate`, `peers`, `channels`, `tasks`, `context`) plus the adapter-owned `say`; replacing the default handler with `agent_client.on_envelope(callback)` (gateways, headless workers); the `ViewPolicy` Protocol with the built-in `FullTranscript`, `WindowedSummary(recent_n=N)`, and `NamedWindowedSummary(recent_n=N)` views plus how to write a custom view; peer discovery via skill markdown (`skill_md=`, `parse_skill_frontmatter`, `hub.set_skill`); the `Envelope` wire format with the `EV_*` event taxonomy (`EV_TEXT`, `EV_PACKET`, `EV_CHANNEL_*`, `EV_EXPECTATION_VIOLATED`, `EV_TASK_CANCEL_REQUEST`, `EV_TASK_CANCELLED`), `audience` and `visible_to` semantics, `Priority`, `causation_id`, and how to send raw envelopes with custom event types via `agent_client.send_envelope(...)`. Use when the user wants to customise the LLM's network surface, write a custom envelope handler, build a gateway / headless worker, or wire peer discovery.
Build an AG2 network `workflow` channel — the orchestrated N-party adapter driven by a declarative `TransitionGraph`. Use when the user needs conditional handoffs, multi-step pipelines, context-aware routing, feedback loops, or is migrating from the classic `GroupChat` + `Agent.handoffs` pattern. Covers `TransitionGraph`; factories `TransitionGraph.sequence(...)` / `.round_robin(...)`; built-in targets (`AgentTarget`, `RoundRobinTarget`, `StayTarget`, `RevertToInitiatorTarget`, `TerminateTarget`); built-in conditions (`Always`, `FromSpeaker`, `ToolCalled`, `ContextEquals`); the typed `Handoff` return for dynamic routing; channel-scoped context via `set_context`; `register_target` / `register_condition` for custom subclasses; the packet execution model and idempotent-tool requirement; eight cookbook patterns (pipeline, hierarchical, escalation, feedback loop, triage, and more); and side-by-side migration from classic `GroupChat`. Load this after `ag2-network-quickstart`.
Plan and deliver repository-aware AI-assisted software work through Agile Iteration Method using intent discovery, PO-owned Epics, TDO-owned User Stories, cross-Epic delivery planning, explicit gates, review, validation, and user-owned acceptance. Use for discussing product direction with repository context, creating and refining Epics and User Stories, planning delivery selections, implementing work, reviewing delivery, configuring project specialists, calibrating repositories and sharing local learning, consolidating knowledge across local AIM projects, controlling AIM modes and cost profiles, opening and controlling local AIM UI, and continuing AIM runs
AISBench Benchmark - AI model evaluation tool for Ascend NPU. Supports accuracy evaluation (service/local models on text, multimodal datasets), performance evaluation (latency, throughput, stress testing, steady-state, real traffic simulation), vLLM/Triton inference services, 15+ benchmarks (MMLU, GSM8K, MMMU, docvqa, ocrbench_v2, etc.), multi-turn dialogue, Function Call (BFCL), and custom datasets.
Creating algorithmic art using p5.js with seeded randomness and interactive parameter exploration. Use this when users request creating art using code, generative art, algorithmic art, flow fields, or particle systems. Create original algorithmic art rather than copying existing artists' work to avoid copyright violations.
Use when generating videos from images with DashScope Wan 2.7 image-to-video model (wan2.7-i2v). Use when implementing first-frame video generation, first+last frame interpolation, video continuation, or audio-driven video synthesis via the video-synthesis async API.
One-click comprehensive cryptocurrency analysis. Parallel subagents collect five dimensions — price, news sentiment, sector comparison, market environment, and project fundamentals — then cross-analyze and output an HTML report (with 24-hour quotes and 7-day trends). Trigger words: 分析BTC, analyze ETH, 比特币怎么样, SOL值得买吗.
Andrej Karpathy's thinking frameworks and way of expressing ideas. Systematically distilled from 20+ blog posts, 16 in-depth interviews and 100+ X posts, extracting 6 core mental models, 8 decision heuristics, complete Chinese output adaptation and a quick reference of classic phrasings. Use as a thinking advisor, analyzing the reliability of AI technology, learning methods, industry trends and product design from Karpathy's perspective. Activates when the user mentions "from Karpathy's perspective", "what would Karpathy think", "Karpathy" or "karpathy mode". Also applies to: Software 2.0/3.0 discussions, vibe coding topics, neural network training, AI hype assessment and the limits of LLM capabilities. It can also trigger when the user merely says "from an engineering realism angle", "march of nines", "building is understanding" or "jagged intelligence". Do not trigger when the user is simply asking ordinary AI-related questions — activate only when they explicitly want Karpathy-style thinking frameworks.
Omniscient APEX Ecosystem development skill. Triggers: apex code, omnihub development, tradeline build, aspiral feature, apex bug, fix apex, apex architecture, omnidash component, triforce guardian, man mode, apex security, apex test, armageddon test, apex deploy, apex optimize, semantic translation, web2 web3 bridge. Produces: zero-drift, first-pass success code for APEX OmniHub, TradeLine 24/7, aSpiral, and all connected applications. Compatible with all LLMs.
Write in-depth article reviews/interpretations based on the original text. Use this skill when the user provides an article, blog post, WeChat official account article, or any long-form content and asks for a review, interpretation, reading response, or derivative content. Applies to: (1) in-depth interpretation of technical articles, industry analyses, year-end summaries, etc., (2) distilling an article's core points and restating them in plain language, (3) generating derivative content for social media distribution.
Answer questions about the Supabase docs app (apps/docs) using documented architecture, build pipeline, and review-pattern notes, and apply feature-design principles (codebase reuse, coding minimalism) when proposing or critiquing changes. Use when the user asks "how does X work in the docs app?", "where does Y live?", "is this approach OK for the docs app?", or before writing non-trivial changes under apps/docs/ — especially anything touching the MDX pipeline, markdown generation, content components, federated docs, or contributor-facing authoring patterns. Can answer architecture questions with Mermaid diagrams when helpful.
Use ast-grep (sg) for AST-aware code search and rewrite across 25 languages. Trigger for structural code matching or deterministic codemods: find every function/call/class/import shaped like X, rewrite console.log to logger.info, strip `as any`, migrate require() to import, find empty catch blocks or missing await, and scan/apply YAML rules. Prefer this over rg/grep when the target is syntax shape rather than text; use rg for string contents, comments, filenames, or regex-style byte searches.
Guide for writing ast-grep rules to perform structural code search and analysis. Use when users need to search codebases using Abstract Syntax Tree (AST) patterns, find specific code structures, or perform complex code queries that go beyond simple text search. This skill should be used when users ask to search for code patterns, find specific language constructs, or locate code with particular structural characteristics.
Write correct async Rust — runtime choice, Send and Sync bounds, cancellation safety, blocking work inside async contexts, and shared state across tasks. Use when writing or reviewing async code, when a future is held across an await point, when the user hits a Send bound error on a spawned task, when a runtime stalls or deadlocks, or when the user asks about tokio, select!, or spawn_blocking.
MySQL and MariaDB specifics for Atlas: dev database and URL scope (schema vs server), character sets and collations, users, roles and grants, enum and set columns, generated columns, foreign-key indexes, functional and prefix indexes, triggers and stored programs, table copies and MY lint checks, partial migration failures (no transactional DDL), GET_LOCK, managed services (RDS, Aurora, Cloud SQL, Azure), schema and migration tests, and custom lint rules. Use with the atlas skill whenever Atlas manages a MySQL or MariaDB database
Guided onboarding of a repository onto Atlas, one verified stage at a time: plan the units and the workflow, define the schema as code, set up the migration workflow, add CI with the Atlas Registry, deploy to staging, and promote to production. Each stage ends on a check the agent runs itself, such as a dry run with no changes or a deployment recorded in Atlas Cloud. Use when the user asks to onboard, adopt, or set up Atlas for a project or a team, or to resume an onboarding
Attack OAuth 2.0 and OpenID Connect flows — enumerate endpoints from the OIDC discovery document, break redirect_uri validation with path traversal, open-redirect chaining, subdomain and regex weakness, and %2F/@ parser tricks, exploit missing state (callback CSRF) and absent or downgraded PKCE, steal codes and tokens via open redirectors and referer leakage, replay and inject authorization codes across clients, escalate scope and bypass consent, confuse access_token with id_token, and take over accounts through "Sign in with X" email trust and device-code consent phishing. Use when you see /authorize, /oauth/token, response_type, redirect_uri, client_id, code= or state= parameters, a "Sign in with Google/Microsoft/GitHub" button, or an OIDC discovery document at /.well-known/openid-configuration.
Audits a third-party or freshly cloned repo before install or run for install hooks, auto-run files, CVEs, typosquats, and malicious code, graded A-F with a safe-to-run verdict. Use when the user says "is this repo safe to run?", or when asked whether a clone is safe to run, or to audit or scan a third-party repo. Not for security review of a diff or PR; use reviewing-security.
Audits a test suite, or a named part, for low-value tests and marks each retain, fix, consolidate, or delete with verified removal evidence. Read-only. Use when the user says "audit the tests", or when asked to audit tests for redundant, junk, or low-value tests. Not for code complexity; use auditing-complexity.
Integrate Avonni components into the code of an LWC component (HTML, JS, CSS). Covers MCP tool usage, attribute/event/type/slot resolution, styling hooks, and validation — not general LWC creation. Trigger when: user asks to add, use, or style an Avonni component in their code; when writing or editing LWC that references avonni-* tags; when the user mentions a component name that starts with "avonni-".
Scandit Barcode Capture (`BarcodeCapture`) in React Native projects—the low-level, full-control single-barcode scanning mode (BarcodeCapture + DataCaptureView + BarcodeCaptureOverlay), without the pre-built SparkScan UI. Use for integration, symbology configuration, result handling, viewfinder and feedback customization, SDK version migration, replacing a third-party scanner (react-native-vision-camera, expo-camera), or troubleshooting.
Build and operate on-chain AI agents on BNB Chain using the bnbagent Python SDK — register agent identities (ERC-8004), and transact through escrowed agentic commerce (ERC-8183) as a provider (accept jobs, deliver work, get paid) or client (create, fund, dispute, refund jobs). Also covers x402 micropayment signing. Use for anything involving BNB Chain agent identity, agent-to-agent paid jobs, BSC escrow jobs, or ERC-8004/ERC-8183/x402.
Brand-first landing page designer — interviews the user to discover brand identity (adjectives, colors, typography, shape language), then generates and iterates on a polished landing page via Stitch with deployment-ready HTML output. Preferred over frontend-design for standalone landing/marketing pages where the user hasn't established visual direction yet. TRIGGER when: user asks to "create/design/build a landing page", "make a homepage for my project/product/service", "build a marketing page", or wants to promote an app/side project. Especially when they haven't defined brand colors, fonts, or visual style — the guided brand interview is the core value. DO NOT TRIGGER when: user has a specific design mockup to implement, wants a dashboard or app UI, needs component-level frontend work (buttons, forms, navbars), is building a multi-page application, or is restyling an existing page with known design tokens. Use frontend-design for those cases.
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.