AI & ML

cyz26 Skill

Use when repeated AgentKB lessons should become playbooks, ADRs, skills, or context packs.

cyz26 Skill

Use when failures, corrections, or review findings should become AgentKB knowledge.

cyz26 Skill

Use when a repository should be configured for AgentKB project problem capture.

marvin-lijunzan Bundle

Design, redesign, and review high-quality interfaces for building and structural digital-twin applications. Use for dashboards, monitoring consoles, sensor telemetry, 3D model views, alarms, and desktop delivery. Default desktop shell is Electron; default test build is a portable, no-install package.

marvin-lijunzan Bundle

RAG-powered multi-paper knowledge base for academic literature: cross-paper QA, literature corpus management, topic accumulation, and evidence retrieval.

marvin-lijunzan Skill

Autonomous scientific researcher agent: orchestrates comprehensive literature review surveys, hypothesis-driven computational experiments, and publication-ready technical report writing.

marvin-lijunzan Bundle

WeChat personal protocol gateway and remote PC automation controller. Listens to WeChat group and direct messages, routes instructions to Antigravity/MCP agents by group topic, executes local tasks, and replies with files/results.

ferroxlabs Skill

Generate review + photo + video prompts at the right post-delivery moment by product type. Inputs: product type (consumable / durable / cosmetic / apparel), platform. Output: prompt-timing recommendation (D+3 / D+7 / D+14 by category), template copy, incentive structure (compliant with platform anti-incentive rules), photo-prompt CTA. Trigger phrases: 'review prompt', 'UGC prompt', 'ask for a photo review', 'request video testimonial', '/commerce ugc-prompts <product-type>'.

jpoindexter Skill

The inclusive-design philosophy and practice — Microsoft's 3 principles + Persona Spectrum, the social model of disability as mismatch, and deep cognitive, neurodivergent, sensory, cultural, economic, and age inclusion patterns that go beyond WCAG conformance.

jpoindexter Skill

iOS layout & spacing for iPhone — the 8pt grid convention + 4pt sub-grid, real Apple layout numbers (16/20pt system margins, safe-area insets, 44/49/50pt bar heights), the 44pt touch target, building dense low-scroll layouts with LazyVGrid/Grid + GridItem (2-up/3-up rules), SwiftUI's adaptive default spacing, and the iOS 26 floating-chrome layout model. Use when laying out any iPhone screen, choosing margins/gutters, or building a grid.

jpoindexter Skill

Decision framework for making a model do a task — choosing between in-context learning, RAG, fine-tuning (LoRA/QLoRA/DPO), and distillation. Maps the gap you have (knowledge / behavior / capability / cost) to the right lever, names when each is the WRONG tool, and covers data requirements, eval, catastrophic forgetting, and combining methods.

jpoindexter Skill

Use when serving or optimizing LLM inference in production — diagnosing or improving TTFT/TPOT/throughput, choosing batching strategy, sizing GPUs, picking vLLM/TensorRT-LLM, or debugging low GPU utilization, TTFT spikes, and OOM. Covers prefill vs decode, the roofline, continuous batching, PagedAttention, chunked prefill, disaggregation, and FlashAttention.

jpoindexter Skill

Reference-grade guide to caching in LLM inference — provider prompt caching (Anthropic cache_control breakpoints, OpenAI automatic prefix caching, Gemini implicit/explicit), semantic caching, and KV-cache internals & management (PagedAttention/vLLM, RadixAttention/SGLang, eviction, quantized KV, memory pressure, multi-tenant safety). Concrete numbers, formulas, failure modes.

jpoindexter Skill

Reference-grade guide to production LLM observability and cost attribution — distributed traces and spans for multi-step agents, OpenTelemetry GenAI semantic conventions, the metrics/percentiles that matter, drift detection with online evals, and per-feature/tenant/user cost attribution via span metadata. Real tooling (LangSmith, Langfuse, Phoenix, Helicone, OTel), tables, do/don't, and the failure modes that leave you blind.

jpoindexter Skill

Reference-grade guide to routing LLM requests across models by cost/latency/quality, escalating hard tasks, and surviving provider failures with multi-provider fallback, circuit breakers, hedging, and honest degraded-mode UX. Use when choosing which model serves a request, building a router/cascade, adding fallback for outages/ratelimits/timeouts, or designing what the user sees when you degrade.

jpoindexter Skill

Reference-grade guide to LLM safety engineering and multi-tenant isolation for AI engineers and security teams — the untrusted-input threat model (OWASP LLM Top 10), direct and indirect prompt injection and why prompting alone can't solve it, defense in depth (privilege separation, dual-LLM, constrained action space, human approval), data-leakage and exfiltration prevention, permission boundaries where the system authorizes (not the model), and cross-tenant isolation across cache, vector index, and memory.

jpoindexter Skill

Reference-grade guide to designing beyond the screen — combining touch/visual/voice/audio/haptic modalities, conversational and LLM-agent UX, voice UI principles and error recovery, sound design, iOS/Android haptics, and TV/automotive/wearable/AR contexts.

jpoindexter Skill

Reference-grade guide to the engineering AROUND the model — the control loop, tools, memory, retries, and budgets (the harness) plus context as a dynamically-assembled token budget — covering lost-in-the-middle, context rot, just-in-time retrieval, compaction, sub-agent isolation, prompt-cache-stable ordering, and the agent-loop and eval patterns that actually determine reliability.

jpoindexter Skill

Reference-grade guide to evaluating LLM and RAG systems — golden/regression/adversarial eval sets, LLM-as-judge and its biases, retrieval metrics (recall@k, MRR, nDCG), grounding/faithfulness/attribution, RAGAS-style scoring, eval-set construction from production traces, and the CI gates that stop silent regressions.

jpoindexter Skill

Reference-grade guide for AI engineers building autonomous and tool-using agents — the plan→act→observe loop and its failure physics, the five budgets every agent must set (iteration, tool-call, token, wall-clock, cost), termination and no-progress detection, runaway prevention (circuit breakers, kill switch), the three guardrail layers (input/output/action), verify-before-acting, bounded sub-agent orchestration, reflection cost, determinism-where-possible, checkpoint/resume, and the canonical failure modes with concrete budgets and pseudocode.

jpoindexter Skill

Reference-grade guide to shrinking and speeding up LLMs without retraining from scratch — numeric formats (FP8/INT8/INT4), PTQ methods (GPTQ, AWQ, SmoothQuant, bitsandbytes NF4, GGUF k-quants), KV-cache quantization, speculative decoding (Medusa/EAGLE/n-gram), and distillation — with concrete numbers, when each fits, and the quality cliffs.

jpoindexter Skill

Reference-grade guide to getting reliable structured output and tool/function calls out of LLMs — JSON mode vs constrained/grammar-guided decoding, response_format json_schema and tool-schema coercion, Zod/Pydantic validation at the boundary, the validate→repair→escalate→fallback loop, tool-argument validation, idempotency for safe retries, fallback chains, and evals. Use when an LLM must return parseable JSON, fill typed tool arguments, or drive deterministic downstream code.

jpoindexter Skill

Capstone synthesis for shipping LLM systems — the full production failure taxonomy (symptom, root cause, detection, mitigation, owner skill), the four-axis inference-stack tradeoff map (latency/quality/cost/reliability), a production-readiness checklist, and incident response for systems that fail silently with no stack trace. Use when hardening an LLM app for production, designing observability/evals/fallbacks, debugging a quality regression with no exception, or making a model/serving/RAG tradeoff decision.

johnqtcg Bundle

Design, maintain, and execute reliable end-to-end tests for critical user journeys with Agent Browser as first choice for exploration and Playwright as the preferred code path for suites and CI. Use for E2E strategy, journey coverage, flaky test triage, artifact collection, CI gating, regression prevention, and browser automation tasks.

johnqtcg Bundle

REST API contract designer and reviewer. ALWAYS use when designing new endpoints, reviewing existing API contracts, planning API versioning, or standardizing error models. Covers resource modeling (URL/naming), HTTP method semantics, status code selection, error model consistency, pagination/filtering/sorting, idempotency keys, concurrency control (ETag/If-Match), object-level authorization (IDOR prevention), rate limiting, backward compatibility assessment, and OpenAPI-ready output. Use even for "just add an endpoint" — inconsistent APIs compound into integration nightmares that are extremely expensive to fix after clients depend on them.

johnqtcg Bundle

Transcribe a specified local video or audio file into cleaned final `.txt`, `.pdf`, or `.docx` transcripts using speech recognition with Apple Silicon GPU acceleration and optional LLM-based proofreading. Use when the user wants text extracted from a local media file path such as `.mp4`, `.mov`, `.mkv`, `.webm`, `.mp3`, `.m4a`, or `.wav`, and the output language should follow the spoken language in the media automatically. Prefer this skill for local-file transcription workflows that should produce cleaned transcripts with natural paragraphs, deterministic Chinese cleanup, and simplified Chinese output for Chinese speech. LLM proofreading is available but off by default.

johnqtcg Bundle

Orchestrate a US-stock investment analysis — classify sector archetype, fetch SEC filings, dispatch a tiered fan-out of six vertical equity-research agents (business model, earnings quality, balance sheet, management, industry, peer comparison) over a validated JSON findings contract, then synthesize a buy/hold/sell verdict with Bull/Base/Bear target ranges using sector-aware thresholds and an auditable probability-assignment procedure. Each verdict can persist to an opt-in JSON-Lines log so later analyses reckon with the prior view. Use when the user asks "should I buy AAPL", "analyze Microsoft", "is NVDA a good buy now", "美股目标价 / 估值 / 投资标的分析", "evaluate GOOGL", or wants any full US-equity workup on a named ticker. NOT for trading signals, technical analysis, options, crypto, ETFs, or macro/sector calls — single-stock fundamental analysis only. NOT for A-shares, HK-shares, or non-US listings — the framework is keyed to SEC filings and US GAAP and refuses them.

creatorland Skill

The front door to the Creatorland Data MCP. A short guided interview that learns who the user is, how familiar they are with AI tools and MCPs (1-3), and what they're trying to achieve — then routes them into the right workflow and skill chain from this pack, at the right level of hand-holding, and saves a workspace profile so no future session re-asks. Run this FIRST on a fresh install or new workspace, before any other skill in this plugin. Also use when the user says "get started", "set me up", "onboard me", "intake", "what can this do for me", "which skill should I use", "help me figure out my workflow", "I don't know where to start", or invokes the plugin with no specific ask. 0 credits by itself; hands off to other skills for the real work.

dauquangthanh Bundle

Drafts detailed module / API / data-model / integration design for the active feature, under docs/design/<feature>/. Uses the architecture intake + ADRs + spec as inputs. Does not commit to implementation timelines. Use when designing technical implementation, planning module structure, or before coding starts on a new feature.

dauquangthanh Bundle

Drafts a project status report (RAG status, accomplishments, next steps, budget variance, open blockers, and escalated risks) from facts the user supplies and upstream artefacts. Produces a dated report file the user can share with stakeholders; does not send it. Use when preparing a project status update, creating a weekly or monthly report, or communicating project health to stakeholders.

fbakiyev Skill

Применяйте при проектировании, реализации или диагностике ответов с опорой на найденные источники; определяйте авторитетность корпуса, разбиение и поиск, права на источники, цитирование и различайте ошибки поиска и генерации.

fbakiyev Skill

Применяйте при оценке или сравнении ML, LLM, RAG, систем ранжирования и прогнозирования; задавайте сопоставимые условия, знаменатели, срезы, неопределённость, проверку автоматического оценивания и основания для приёмки.

nexadevapp Skill

Creates entity model documents with Mermaid.js ER diagrams and attribute tables defining entities, relationships, data types, and validation rules. Use when the user asks to "create an entity model", "design a data model", "draw an ERD", "define database schema", "model entities", or mentions entity-relationship diagram, ER diagram, database design, or data modeling.

nexadevapp Skill

Creates Prisma schema models and generates database migrations from the entity model. Use when the user asks to "create a migration", "update the schema", "set up database tables", "write a Prisma migration", or mentions schema migration, DB migration, database versioning, or Prisma schema changes.

nexadevapp Skill

Lead Requirements Engineer that systematically elaborates all use cases through structured analysis, probing questions, and iterative refinement. Processes use cases in thematic clusters, applies CRUD matrix analysis, actor-goal completeness, exception path probing, MoSCoW prioritization, and traceability checks. Spawns a Product Owner sub-agent to answer clarification questions — only genuinely ambiguous decisions reach the human. Updates requirements, entity model, and use case diagram as living documents. Use when the user asks to "engineer requirements", "elaborate use cases", "analyze use cases", "validate use case completeness", "deep-dive on requirements", "prioritize requirements", or mentions requirements engineering, use case elaboration, MoSCoW, or completeness analysis.

grasscaograss Bundle

Build and sharpen a project's domain model. Use when the user wants to pin down domain terminology or a ubiquitous language, record an architectural decision, or when another skill needs to maintain the domain model.