AI & ML

finchipaiorg Skill

IRP review memos fail when the agent assesses the updated plan against general regulatory standards without using prior incident materials as evidence of specific gaps the new plan must close, and without assessing whether governance materials and audit findings impose additional requirements the plan must satisfy.

pyyding Bundle

Generate or update an agent skill (SKILL.md) from a CLI script by inspecting its `--help` output and subcommands. Use when the user wants to "make a skill out of" a CLI, "wrap" a binary as a skill, "turn `foo --help` into a skill," "regenerate the skill for `<cli>` now that I added new commands," or otherwise package or refresh a command-line tool's interface as a reusable agent skill so future agents can drive it without re-discovering its flags.

finchipaiorg Skill

DPA redline reviews lose rigour when the agent treats playbook positions as optional and fails to cross-reference the broader commercial agreement when assessing risk of counterparty changes.

finchipaiorg Skill

Incident summary memos fail when the agent does not reconcile conflicting accounts across multiple incident-related documents into a single authoritative incident narrative with discrepancies explicitly flagged.

mathieu0905 Bundle

Evaluate, compare, calibrate, and rank research papers or LLM/foundation-model technical reports from PDFs, extracted text, or paper titles/topics. Use when Codex is asked to score papers on innovation/value/rigor, find related comparison papers, compare model technical reports, use public reviews for calibration, analyze historical score drift, create tier lists, or write image-generation prompts summarizing rankings.

roanbrasil Skill

Use when designing, running, or improving evaluations for LLM applications — covers eval types, datasets, metrics, LLM-as-judge, tools (LangSmith, Ragas, PromptFoo, DeepEval, Braintrust), and CI integration

roanbrasil Skill

Expert Spring AI (1.x) skill — invoke when building AI-powered Java/Kotlin applications with Spring Boot, including chat integrations, RAG pipelines, tool/function calling, streaming, advisors, embedding, multimodal, or swapping LLM providers without code changes.

roanbrasil Skill

Implement Google's Agent-to-Agent (A2A) protocol for agent interoperability — invoke when the user is building multi-agent systems, agent registries, agent delegation, or integrating A2A with Spring AI or Python frameworks.

persimmon-automation-labs Skill

Prompt hygiene and Claude output-validation audit for Persimmon stack. Checks prompt centralization, prompt-injection defenses, structured output with tools or Zod-parsed JSON, citation grounding, prompt caching, model routing, idempotent persistence, and eval harness coverage. Triggers on "prompt review", "ai review", "claude audit", "pre-launch", "before shipping".

kyohwang Bundle

Use the ClawHub CLI to search, install, update, and publish agent skills from clawhub.com. Use when you need to fetch new skills on the fly, sync installed skills to latest or a specific version, or publish new/updated skill folders with the npm-installed clawhub CLI.

kyohwang Bundle

Use mcporter to inspect MCP server configs and run direct MCP tool calls with explicit, auditable command patterns.

kyohwang Bundle

Audit candidate Agent Skills for malicious or unsafe patterns before admission. Use when reviewing skills in proposals/, before promoting to skills/, or when re-validating existing skills for security regressions.

kyohwang Bundle

使用 xiaohongshu-mcp HTTP 服务执行小红书图文发布。适用于“先审后发”流程：健康检查、登录检查/二维码登录、按 pages 目录组装内容、owner 审核后发布、回传结果与失败诊断。

roanbrasil Skill

Expert Rust systems programming — ownership, lifetimes, async, traits, unsafe, and the idiomatic Rust mental model

aastroza Bundle

Write and revise documentation that supports code creation and technical projects: READMEs, project guides, engineering blog posts, setup docs, agent instructions, release notes, and technical narratives. Use when the user asks for clearer docs, better README prose, less AI-sounding writing, stronger documentation structure, or a writing pass grounded in real project work.

roanbrasil Skill

Expert guidance for LLM fine-tuning decisions, PEFT methods (LoRA/QLoRA), dataset preparation, training, evaluation, and deployment — invoke when the user asks about training, adapting, or customizing a language model.

persimmon-automation-labs Skill

ai-rag-retrieval

persimmon-automation-labs Skill

ai-prompt-library

roanbrasil Skill

Use when setting up tracing, monitoring, logging, or cost tracking for LLM applications — covers LangSmith, LangFuse, Arize/Phoenix, metrics, alerting, and prompt versioning

roanbrasil Skill

Expert LangChain patterns for Python (v0.2/v0.3+) — invoke when building RAG pipelines, agent chains, LCEL expressions, tool-calling agents, memory-backed conversations, or migrating from LangChain v0.1.

persimmon-automation-labs Skill

Design and review Prisma/Postgres schemas with rigor — independent of domain. Snapshot mutable values onto historical records, complete lifecycle/state enums (Postgres enums, not string constants), join tables for hidden many-to-many, FK/cardinality integrity, money-as-integer-cents, createdAt/updatedAt on every model, and a per-field rationale. Includes the validation 'lenses' to review an existing data model in a spec for missing/extra models and fields, plus gap-bucket classification. Use when designing a Prisma schema or reviewing one. Trigger keywords: schema design, data model, Prisma model, review schema, missing table, enum, snapshot price, join table, cardinality.

persimmon-automation-labs Skill

Scaffolds the admin CRUD layer for the Persimmon stack — NextAuth v5 session + middleware route guard, role model, list/detail/edit/create/delete via Server Components + Server Actions, server-side pagination/search/filter, and dashboard KPI cards + recharts. Use when building an admin panel, internal tool, back-office, or CRUD section for an entity. Trigger keywords: admin panel, CRUD, back-office, dashboard, internal tool, role guard, pagination, recharts.

persimmon-automation-labs Skill

Prisma + pgvector schema and query patterns for RAG on the Persimmon stack. Use when designing vector-indexed tables, writing embedding pipelines, choosing HNSW vs IVFFlat, picking embedding dimensions (OpenAI/Voyage/Cohere), writing cosine-distance retrieval queries, or wiring hybrid search (BM25 + vector). Covers the `Unsupported("vector(N)")` Prisma trick, raw `$queryRaw` retrieval, chunk table design, and migration workflow on Railway Postgres.

persimmon-automation-labs Skill

Pre-delivery QA orchestrator for Persimmon client projects. Runs all review-* skills in parallel (security, performance, type-safety, data-layer, prompt-output), aggregates findings by severity, and produces a ship / don't-ship verdict. Triggers on "final review", "pre-delivery QA", "before shipping", "before we ship", "ship check", "launch check".

roanbrasil Skill

Expert LangGraph skill for stateful multi-agent workflows — invoke when building graph-based agents, human-in-the-loop flows, checkpointed resumable pipelines, supervisor/swarm multi-agent systems, or any LLM workflow that requires cycles or conditional routing.

roanbrasil Skill

Production AI safety patterns for LLM applications — invoke when the user is building guardrails, defending against prompt injection, handling sensitive data, auditing tool execution, or red-teaming an LLM system.

agent-spm-registry Bundle

A test skill published via the MCP server

roanbrasil Skill

LLM cost optimization for production systems — invoke when reducing API spend, designing caching strategies, routing requests, or budgeting token usage

roanbrasil Skill

MCP (Model Context Protocol) server development — invoke when building, debugging, or extending an MCP server with tools, resources, or prompts in any language

rsaccone Skill

Apply SwiftUI best practices: predictable state, small views, testable logic, and correct navigation patterns.

rsaccone Skill

Production-grade best practices for ActivityKit and Live Activities including architecture, token lifecycle, updates, concurrency, UI, and testing.

vathivis Skill

Verify and triage review findings without code changes; use when a second-agent review summary is pasted and you must confirm or refute each finding before acting. Use main-agent or highest-quality reviewer verification; do not use explorer subagents for this review work.

findddx Bundle

用于 Codex 通过可见浏览器和 Chrome DevTools MCP 执行授权范围内的知网论文批量检索、候选整理、题名精确回查、详情页 PDF 优先下载、下载文件归档与问题清单输出；当用户要求自动搜集并下载 CNKI PDF 论文时使用。

roanbrasil Skill

Advanced prompt engineering for production LLM applications — invoke when designing, debugging, or optimizing prompts for any model or use case

douglac Skill

Skill abrangente pra operar o ERP Conta Azul via Conta Azul MCP — vendas, faturamento, contas a pagar/receber, fluxo de caixa, cadastros de pessoas/produtos/serviços, contratos, parcelas/quitações e NF-e. Use sempre que o usuário perguntar ou pedir algo sobre a operação financeira/comercial da empresa dele na Conta Azul (vendas, clientes, fornecedores, produtos, contas, parcelas, notas fiscais). Orquestra as 35 ferramentas contaazul_* (read + write) do servidor MCP remoto em https://api.mcp.ai/contaazul.

tarzelf Bundle

Use when an AI agent needs to search live hotel rates, quote a price comparison against Booking.com-style public pricing, create a tracked Stripe checkout link, and earn a 10-20% commission per order (3% on free keys). Loads the Travaso MCP commerce layer into Claude, Cursor, Codex, Hermes Agent, or any backend that speaks MCP. End-user trigger phrases: 'find a hotel', 'check live rates', 'compare to public price', 'send me a checkout link', 'book me a room'. Use when the user wants the agent to sell hotels, not just recommend them.