Skills
96,770 packaged capabilities and playbooks for agents. Install any of them with one command.
Testing & quality
Execute QA test plans on Holoscan Sensor Bridge hardware. Reads a user-provided test document, filters tests based on the user's setup, determines which tests can run automatically, executes them with pass/fail evaluation, and produces a structured test results report.
Problem scoping with Is/Is-Not for 8D D2 problem description, CAPA investigation, or hypothesis elimination. Defines the precise boundary by contrasting what IS observed vs what IS NOT — eliminating hypotheses that don't fit the pattern. A Ford-originated automotive technique used in every structured quality investigation.
Plan and audit localization quality programs for mobile apps, games, SaaS, web, desktop, store listings, lifecycle messages, support content, legal copy, RTL, truncation, pluralization, cultural fit, in-context review, pseudo-localization, LQA, release gates, and market feedback. Use when translation must become launch-quality product experience.
Format and lint markdown and MDX files using markdownlint with comprehensive rule knowledge for automated and manual fixes. Use when working with .md or .mdx files, formatting documentation, linting markdown/MDX, or when user mentions markdown issues, formatting problems, or documentation standards.
Plan, configure, and chain repo-native Nemotron customization steps into single-step or multi-step pipelines: curation, translation, SFT/PEFT (AutoModel or Megatron-Bridge), pretraining/CPT, RL alignment (DPO/RLVR/GRPO/RLHF), BYOB/MCQ benchmarks, checkpoint conversion, ModelOpt optimization, env profiles, and evaluation of trained checkpoints or existing/hosted endpoints. Use when a request names a Nemotron step or workflow, or asks to clean, translate, train, fine-tune, align, convert, optimize, evaluate, or compose these into a pipeline. Do NOT use for frontend/dashboard/visualization work, generic ML advice, billing/access, or non-Nemotron coding tasks.
Generates code coverage reports for OpenHarmony C/C++ based on UT cases, supporting full and incremental coverage analysis; requires connecting to the test device via hdc. Use when: (1) the user requests full code coverage for a subsystem/component/module/test suite/test case; if a report path is provided, parse it as the --output path; (2) the user requests incremental code coverage for a component/module/test suite/test case, requiring git diff files; if a report path is provided, parse it as the --output path. Note: incremental coverage does not support subsystem-level analysis. Activate when tests need to be run by connecting to an OpenHarmony device
Security review requiring judgment about exploitability: injection and input handling, authentication and session management, authorization, secret storage and cryptography, and Agentic AI risks, against OWASP Top 10:2025 and ASVS 5.0. Use when user says 'review for security', 'is this secure', 'review this endpoint for injection', 'check for SQL injection or XSS', 'review auth/authorization', 'how are we storing secrets', 'check how we store secrets in this service', 'is this crypto correct', or when writing cryptography, session management, or AI agent code. Judges reachability and impact in this codebase; /clouddrove:appsec owns the deterministic checks a tool can answer (lockfile CVEs, missing headers, wildcard CORS).
Reproduce a paper's method in minimal, direct PyTorch when no official code is available, to check whether the method works as claimed. Use whenever the user says "reproduce this paper", "implement this paper's method", "there's no code for this paper", "I want to check if this paper actually works", "turn this algorithm/equation into code", or shares a paper/arXiv link and wants a working implementation of its core mechanism. The output is a small, single-file, readable implementation plus staged verification gates (shape checks → overfit tiny batch → baseline comparison at toy scale) and an explicit assumptions ledger for everything the paper left underspecified. Do NOT use for designing training recipes for your own experiments (pytorch-training-recipe / jax-training-recipe), auditing an existing released artifact (reproducibility-audit), or structuring a research repo (research-codebase).
Run a PDCA cycle, Plan Do Check Act improvement cycle, or structured improvement project. Guides through problem analysis, piloting, verification, and standardisation. Distinguished from 8D: PDCA is for proactive improvement initiatives, 8D is for reactive defect response. Use for process optimisation, lessons learned implementation, and quality objectives.
Run an independent, report-only overall audit of a delivered design document or multi-story feature. Verify story completion, assembled behavior, design alignment, documentation, and relevant tests, then report whether blocking findings remain.
Production Part Approval Process (PPAP) — verify PPAP submission level, audit all 18 elements, check completeness for customer approval, prepare PSW. Use when a supplier needs to submit parts for approval, when reviewing a PPAP package, or when determining which PPAP level is required. Covers AIAG PPAP 4th edition with Ford, BMW, VW, and Stellantis OEM-specific requirements.
Audit a recently written Playwright test as a fresh pair of eyes. Reviews only the most recent uncommitted changes via git diff, checks compliance with the project's Playwright scripting rules, architecture, and POM conventions, and reports findings before optionally applying fixes.
Use when writing tests, designing test coverage, handling defect fixes, performing root-cause analysis, setting up CI verification, or configuring commit-time hooks. Also use when creating test matrices or choosing between unit/integration/E2E tests. Do not use for general development or linting setup.
Use when the user wants to adversarially stress-test a guardrail, classifier, prompt, or API they own or are authorized to test, to surface the distinct ways it fails. Generates adversarial inputs, runs them through the target and a ground-truth oracle, logs every target-vs-oracle disagreement as a failure de-duplicated by technique class, and loops until rounds stop surfacing new classes. Produces a catalogue of distinct, reproducible failures — the attacker half of a find→fix setup. Not for patching the target, and not for attacking systems the user does not own or have permission to test.
Review module boundaries, coupling, public APIs, and refactors for concrete complexity and unnecessary abstraction. Use for structural reviews or requests such as "this file does too much", "hard to test", "split this file", or "the same change touches many places". Exclude ordinary bug fixes, formatting, and naming-only edits that do not affect structure.
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.