Skills
97,222 packaged capabilities and playbooks for agents. Install any of them with one command.
Development
INVOKE THIS SKILL when routing a LangGraph agent with a decision model (TypeSafe Jev, SemIf) instead of an LLM, or when auditing an existing agent for LLM calls that only produce a routing decision. Covers langchain-typesafe Noul/Choice/Score, reading answers correctly, threshold design, and LangSmith Gateway wiring.
Automatically implement all descendant sub-issues under a parent issue in parallel using worktrees, while respecting dependency order, then review before push, create PRs, monitor CI, and bring them to a mergeable state. Use for “implement the issue tree in parallel,” “implement all sub-issues under this issue,” “implement the whole tree in parallel,” or “automatically develop the issue hierarchy.” Divide work into per-issue planning (Plan: session inheritance model) and implementation (Implement: sonnet). Review before push; push, create PRs, and start CI only after review passes. Specify external checks explicitly with args.externalChecks as {"app", "context"} pairs. Use [] to confirm “none” and avoid unnecessary waiting; if unspecified, automatic merge stops; the legacy slug-only format fails closed for automatic merge. Opt in to automatic squash merge with autoMerge: true and explicit externalChecks, including trusted context declarations for all Apps. This requires merge-exec self-fetch revalidation and an actual check of server-side branch protection. By default, autoMerge is false and execution stops at a mergeable state for a human to merge. You can also delegate to a server-side workflow sample (see upstream docs/implement-issue-tree/auto-merge-sample.yml) and branch protection. Control execution order with parallelism (parallel) and dependencies (dependsOn). For strict serialization by phase, use phaseGate: true (opt-in): serialize the direct children of the root (phase parents) in sub-issues list order, and do not start the next phase until all descendants of the previous phase are merged or closed. By default, phaseGate is false and only the current post-order priority applies. To see in advance how many issues can be started and how many will be blocked, use dryRun: true (returns an estimate without writing anything). See implement-issue for implementing a single issue and implement-review-pr for PR reviews.
Read and write large cuPyNumeric arrays to HDF5 with Legate's parallel, distributed HDF5 I/O (legate.io.hdf5: to_file, from_file, from_file_batched). Use when a developer needs to save a cuPyNumeric array to an .h5/.hdf5 file, load an HDF5 dataset into a distributed cuPyNumeric array, read a large HDF5 dataset in chunks, hand arrays to an HPC pipeline as a single file, or accelerate HDF5 disk I/O with GPUDirect Storage (GDS). Do not use it for Parquet/cuDF/raw-binary or other sharded/custom layouts (see the cupynumeric-parallel-data-load skill), Zarr or object-store/S3 output, .npz or pickled archives, plain h5py without cuPyNumeric, or pure array compute such as FFT, matmul, or reductions.
Test-driven development via red-green-refactor with an auditable commit protocol. Use when implementing any feature, bugfix, refactor, or behavior change before writing implementation code, or when the user mentions TDD, red-green-refactor, or test-first development.
Integrate Salesforce B2B Commerce open source components from GitHub into B2B Commerce stores. Use when users mention "integrate open code components", "open source B2B commerce", "add open code components", "forcedotcom/b2b-commerce-open-source-components", or want to add open source commerce components to their store. Copies all components and labels so they become available in Experience Builder.
When the user's app or update was rejected by Apple App Review or Google Play Review and they need to diagnose why, fix it, and resubmit fast. Use when the user mentions "app rejected", "App Review rejection", "guideline violation", "Apple rejected my app", "Google Play rejected", "Play policy violation", "Resolution Center", "metadata rejection", "binary rejection", "guideline 2.1", "guideline 4.3", "guideline 5.1.1", "Sign in with Apple required", "Apple ID rejection", "Play Store suspension", "appeal", "I need to respond to App Review", or "expedited review". For pre-submission listing health, see aso-audit. For metadata-only fixes, see metadata-optimization.
Creates, structures, and reviews technical documentation following the Diátaxis framework (tutorials, how-to guides, reference, and explanation pages). Use when a user needs to write or reorganize docs, structure a tutorial vs. a how-to guide, build reference docs or API documentation, create explanation pages, choose between Diátaxis documentation types, or improve existing documentation structure. Trigger terms include: documentation structure, Diátaxis, tutorials vs how-to guides, organize docs, user guide, reference docs, technical writing.
Elite UX/UI & Advanced GSAP Motion Engineer. Enforces Python-driven true randomization for layout variance, strict AIDA page structure, wide editorial typography (bans 6-line wraps), gapless bento grids, strict GSAP ScrollTriggers (pinning, stacking, scrubbing), inline micro-images, and massive section spacing.
Run a spec-driven agent loop where coding tasks live as markdown specs that move through inbox → active → archive, get implemented by Claude Code or Codex, and pass a review gate before they count as done. Use when the user mentions "loop factory", a "spec-driven loop", an "agent factory", wants repeatable/reviewable agent work, or when a repo has a factory/specs/inbox or factory/specs/active directory. Also covers installing and scaffolding the loop-factory CLI into a project.
SOLID principles, design patterns, DRY, KISS, and clean code fundamentals. Use when reviewing architecture, checking code quality, refactoring, or discussing design decisions. Triggers on “review architecture,” “check code quality,” “SOLID principles,” “design patterns,” or “clean code”
Use when building .NET 10 or C# 14 applications; when using minimal APIs, modular monolith patterns, or feature folders; when implementing HTTP resilience, Options pattern, Channels, or validation; when seeing outdated patterns like old extension method syntax
Recommend and customize Megatron Bridge library and benchmark recipes for a user's model, GPU count, hardware, sequence length, and pretrain/SFT/PEFT goal. Use when selecting a starting recipe, comparing library and benchmark configs, resizing parallelism for a GPU allocation, or distinguishing convergence changes, semantics-preserving execution tuning, and benchmark-only shortcuts.
One-click contribution flow for OpenDesign (nexu-io/open-design) — even for non-coders. Pick one of four cards (ship a Skill or Design System you made with OD; translate docs; fix a typo / write a blog; report a bug), the agent validates and opens a PR (or issue) for you. Trigger words contribute to open design, ship my OD skill, ship my OD design system, translate OD docs, report an OD bug, od-contribute.
Package and build custom AI models with Cog for deployment on Replicate. Use when creating a cog.yaml or predict.py, defining model inputs and outputs, loading model weights at setup time, building Docker images for ML models, serving locally with cog serve or cog predict, or porting a HuggingFace, GitHub, or ComfyUI model to run on Replicate. Trigger on phrases like "build a model", "package a model", "create a Cog model", "wrap a model", "containerize an AI model", "predict.py", "cog.yaml", "BasePredictor", or "Cog container", and when referencing cog.run, github.com/replicate/cog, or github.com/replicate/cog-examples. Covers GPU and CUDA setup, pget for fast weight downloads, async predictors with continuous batching, streaming outputs, and cold-boot optimization for image, video, audio, and LLM models. For pushing built models to Replicate, see publish-models. For running existing models, see run-models.
Validate and use selective and full activation recompute in Megatron Bridge to reduce GPU memory usage at the cost of extra compute. Use for activation memory OOMs or regressions involving recompute_granularity, recompute_num_layers, recompute_modules, recompute_method, selective recompute, full recompute, or activation checkpointing.
Add a new cuTile GPU kernel operator to TileGym. Covers dispatch registration in ops.py, cuTile backend implementation, __init__.py exports, test creation, and benchmark in tests/benchmark. Use when adding, creating, or implementing a new cuTile operator/kernel in TileGym, or when asking how to register a new cuTile op.
Version-aware guide for configuring and running Apollo Router for federated GraphQL supergraphs. Generates correct YAML for both Router v1.x and v2.x. Use this skill when: (1) setting up Apollo Router to run a supergraph, (2) configuring routing, headers, or CORS, (3) implementing custom plugins (Rhai scripts or coprocessors), (4) configuring telemetry (tracing, metrics, logging), (5) troubleshooting Router performance or connectivity issues, (6) securing the graph with JWT, declarative field-level authorization directives, or persisted-query safelisting, (7) managing router.yaml as version-controlled config with CI/CD validation.
Full Sentry SDK setup for React Router Framework mode. Use when asked to "add Sentry to React Router Framework", "install @sentry/react-router", or configure error monitoring, tracing, profiling, session replay, logs, or user feedback for a React Router v7 framework app.
Elite website image-to-code skill for Codex. For visually important web tasks, it must first generate the design image(s) itself, deeply analyze them, then implement the website to match them as closely as possible. In Codex, it must prefer large, readable, section-specific images instead of tiny compressed boards, generate fresh standalone images for sections or detail views instead of cropping old ones, avoid lazy under-generation, avoid cards-inside-cards-inside-cards UI, and keep the hero clean, spacious, readable, and visible on a small laptop.
Convert single-node scripts to multi-node Slurm sbatch jobs and debug common multi-node failures. Covers srun-native vs uv run torch.distributed approaches, container setup, NCCL timeouts, OOM sizing for MoE models, and interactive allocation
Design UIs in Pencil (.pen files) and generate production code from them. Use when working with .pen files, designing screens or components in Pencil, or generating code from Pencil designs. Triggers on tasks involving Pencil, .pen files, design-to-code workflows, or UI design with the Pencil MCP tools.
Update the skill list and repository structure tree in CLAUDE.md based on changes since the previous update commit (tracked in _/.last-update-docs). Use when adding a new skill, changing .claude/agents/ or .claude/rules/, or when asked to “update the docs” or “update CLAUDE.md.” Also includes guidance to consult when asked to strengthen code comments or documentation comments.
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.