Skills
98,067 packaged capabilities and playbooks for agents. Install any of them with one command.
Development
Use when a finished feature branch needs integration or disposition — produces a local merge, pull request, kept branch, discard, or block with reviewer-readable commits and pull-request text. Not for reviewing an existing PR (inspect-change) or cutting a version (cut-release).
INVOKE THIS SKILL when implementing human-in-the-loop patterns, pausing for approval, or handling errors in LangGraph. Covers interrupt(), Command(resume=...), approval/validation workflows, and the 4-tier error handling strategy.
Invoke this skill when your LangGraph needs to persist state, remember conversations, travel through history, or configure subgraph checkpointer scoping. Covers checkpointers, thread_id, time travel, Store, and subgraph persistence modes.
Audits Dart and Flutter package dependency licenses using the Very Good CLI packages_check_licenses MCP tool, flags non-compliant or unknown licenses, and produces a compliance summary report. Use when the user says "check licenses", "license audit", "check dependency licenses", "license compliance", "review package licenses", "are our dependencies compliant", "scan for license issues", or "pre-release license check". Use it especially when the request asks for a compliance verdict without a scan, as in "read the licenses off this pubspec", "confirm we're compliant", "just tell me if these packages are safe to ship", "I'd rather not run anything", or "which of these are GPL", because refusing to certify compliance from a dependency list is the call this skill governs. A pasted pubspec is a trigger, not a substitute for the audit.
Audit, plan, scaffold, and safely migrate research project structures across greenfield, existing Git/non-Git repositories, and monorepo subroots. Use for project folders, repository cleanup, source inventory, move maps, naming and storage policy, Python/R/mixed/LaTeX profiles, template provenance, conflict review, applied-move evidence, or rollback. Existing projects are read-only until the user authorizes exact action IDs bound to a plan digest. Preserve uncommitted and untracked work, symlinks, submodules, and memory-pm's .light content. This is an off-DAG local tool: do not emit findings or invent a STAGE_GATES/ROUTES connection.
Design or modernize a software system from an evidence-backed current-state inventory through quality attributes, architecture options, API and schema contracts, migration/rollback plans, ADRs, and verification. Use for greenfield or existing monoliths, modular monoliths, services, system/API/database design, schema migration review, data-flow reliability, tenant/PII controls, or architecture evolution. Existing systems stay read-only until the user selects an option and authorizes exact mutations. Unknown facts stay UNKNOWN. This is an off-DAG engineering skill: do not emit findings or invent STAGE_GATES, ROUTES, stages, or back-edges.
Runs LLM inference on CPU, Apple Silicon, and consumer GPUs without NVIDIA hardware. Use for edge deployment, M1/M2/M3 Macs, AMD/Intel GPUs, or when CUDA is unavailable. Supports GGUF quantization (1.5-8 bit) for reduced memory and 4-10× speedup vs PyTorch on CPU.
Guides fine-tuning of large language models with LLaMA-Factory, covering the WebUI no-code interface, training across 100+ supported models, quantized QLoRA at 2/3/4/5/6/8-bit precision, and multimodal model support. Use when setting up or configuring a LLaMA-Factory fine-tuning run, choosing a quantization level for QLoRA training, using the WebUI instead of writing training code, fine-tuning a multimodal model, or debugging a LLaMA-Factory training error. Not for general-purpose inference serving or for other training frameworks; use the matching skill for those.
Use when creating or iterating on a project after the Make App requirements proposal is confirmed. Decide whether to generate the My Approvals navigation by default based on business entry, edit-and-save, or approval-processing capabilities. Also covers BPM, bpmDefinitionKey, approval creation fields, save and submit, submit and save, process status, and approval tags in detail titles. Handles generation conditions, runtime initiation configuration, saving and first initiation, and status display in business lists and details. Does not cover cancel/recall, installing the My Approvals npm package or implementing the task center, approval processing, approval history, the process designer, or Console approval management.
Translate NI-DAQmx C function signatures into correct calldaqlib MATLAB calls. Use when the user mentions calldaqlib, any DAQmx* C function name (DAQmxCfgSampClkTiming, DAQmxExportSignal, DAQmxSetStartTrigRetriggerable, DAQmxGetSampClkRate, etc.), or needs NI-DAQmx functionality not exposed by the daq object. Covers: projection rules (taskHandle implicit, output pointers become return values, buffer-size args, enum-as-string), multi-task semantics, start() clobber, device-level empty-daq workaround, dictionary returns, string getter placeholder, array getter placeholder. Trigger contexts: low-level DAQ, escape hatch, retriggerable acquisition, shared timebase, export signal, connect terminals, external sample clock, trigger configuration, NI driver, C function translation.
Migrates existing Nango TypeScript createSync implementations from nango.lastSyncDate, legacy incremental syncType, and non-resumable full refreshes to checkpoint-based syncs. Use when updating customer Nango sync code to define checkpoint schemas, call getCheckpoint/saveCheckpoint/clearCheckpoint after every batchSave (including inside paginate loops), test dryruns with --checkpoint, and fix deletion handling for checkpointed incremental or full syncs.
This skill should be used when the user asks to train, debug, scale, or improve ML models. PROACTIVELY activate for: (1) PyTorch, TensorFlow/Keras, JAX, Flax, Hugging Face Trainer/Accelerate training loops, (2) distributed training, DDP/FSDP/DeepSpeed, TPU/GPU setup, (3) mixed precision AMP/bf16, gradient accumulation, checkpointing, seeding, (4) overfitting, imbalance, loss functions, regularization, LR schedules, warmup, (5) memory optimization, gradient checkpointing, offloading, quantization-aware training. Provides: reproducible training best practices across deep learning and classical ML.
next-intl internationalization for 6 locales (pt-PT, en, tr, es, fr, de) with locale-prefixed routing, useTranslations/getTranslations patterns, and message file structure. Use when adding translations, creating localized pages, implementing language switchers, or handling locale routing.
Use Opper from application code and over HTTP, and as the source of truth for Opper platform concepts: the gateway, control plane, models, the OpenAI / Anthropic / Google / OpenResponses-compatible endpoints under /v3/compat, structured output, server-side tools (web search), roundtable, streaming, tracing, and building agents. Use whenever the user writes code against Opper in any language (the OpenAI, Anthropic or Google Gen AI SDK, or an agent framework, pointed at /v3/compat), asks about Opper concepts, models, endpoints or raw HTTP, or migrates from OpenAI / OpenRouter / Anthropic, off Opper's legacy /v3/call (opper.call), or off the discontinued `opperai` SDK, even if they only say "Opper". For media generation (images, audio, video, OCR), files, and realtime voice, use `opper-multimodal`. For endpoint signatures, fetch the live spec at https://api.opper.ai/v3/openapi.yaml
Optimizes audio memory, CPU cost, and playback quality through import settings and mixer configuration. Use when asked to reduce audio memory or Audio Mixer CPU cost, change clip Load Types, sample rates or codecs, or force 3D audio to mono.
Provides comprehensive guidance for Parcel bundler including zero-configuration setup, asset handling, hot module replacement, and production builds. Use when the user asks about Parcel, needs to set up a build tool quickly, or work with Parcel's automatic configuration.
PavedPath Code is the code-focused edition of PavedPath. It helps agents solve software engineering problems by finding proven implementation paths from GitHub repositories, issues, pull requests, discussions, code examples, release notes, and open-source evidence.
Fine-tunes LLMs with Hugging Face PEFT, using LoRA, QLoRA, IA3, AdaLoRA, prefix tuning, and prompt tuning so that under 1% of parameters are trained. Covers rank, alpha, and target module selection, loading and merging adapters, multi-adapter serving, and integration with TRL SFTTrainer, Axolotl, and vLLM. Use when fine-tuning 7B-70B models on limited GPU memory, when running QLoRA on a single 24GB GPU, when choosing LoRA rank and alpha, when merging or swapping adapters on one base model, or when debugging CUDA OOM or an adapter that does not apply. Not for full fine-tuning of small models under 1B parameters or cases needing all weights updated.
Create a PlanWeave package-shaped plan draft from a fuzzy goal, sparse requirements, codebase context, or a user idea before a formal Plan Package exists. Use when the user asks to make, draft, design, break down, or plan PlanWeave work without an existing PRD, roadmap, issue set, or strong source plan. Materialize a Plan Package only when the user explicitly asks.
Decomposition methodology for pm agent — turns an approved ARCH document into a Beads task list with explicit dependencies, time-boxes, and acceptance criteria. The pipeline can only orchestrate work it can see; this skill defines what "seeable work" looks like.
Use when creating an approachable, self-contained HTML review aid for a pull request; explaining what changed, why it matters, how it works, and how it fits into the broader system; turning PR diffs, commits, tests, and architecture context into a local `.pr-review/` HTML page for reviewers; or helping reviewers understand complex code changes without dumping the full diff.
Use when adding or fixing types on Prisma `Json`, `String`, `String[]`, `Int`, or `Float` fields with prisma-json-types-generator, including `/// [Type]` and `/// ![Type]` comments, enum-like literal unions, namespace vs inline types, TypedSQL result columns and parameters, and generated types that still show `JsonValue`, `unknown`, or a plain scalar.
Use when adding service-wide Spring Boot 3 Actuator, Micrometer metrics, distributed tracing, OpenTelemetry export, structured logging, health probes, dashboards, or alerts. Combine with ai-observability only for Spring AI-specific token, prompt, and model telemetry.
Enhance user prompts by analyzing project context (code structure, dependencies, conventions, existing patterns). Use when users provide brief development requests that would benefit from project-specific context to generate more accurate, contextually-aware prompts.
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.