Tu agente se encuentra con una tarea que no puede hacer — leer datos del mercado de valores A, conseguir que un segundo modelo audite un diff, transcribir una entrevista, quitar una marca de agua, clonar el sistema de diseño de un sitio. En algún lugar existe un archivo de skill que lo resolvería. Encontrarlo es la parte molesta.
Este servidor pone un catálogo pequeño y seleccionado dentro del agente. Busca, lee los datos de entrega y te dice exactamente qué ejecutar. Los elementos gratuitos incluyen una URL de descarga directa; los de pago incluyen un precio y nada de discurso de ventas.
El catálogo es agentskill.nz. Es deliberadamente pequeño. Los directorios públicos de skills contienen cientos de miles de entradas sin selección — calidad media medida en torno a 6 sobre 12, aproximadamente un tercio con riesgo de inyección de prompts. Cada entrada aquí incluye qué hace, qué necesita para ejecutarse, con qué se probó y dónde se detiene.
Related MCP server: SkillFlow MCP Server
No hay paso de instalación. Apunta tu cliente MCP a él:
{
"mcpServers": {
"agentskill": {
"command": "npx",
"args": ["-y", "github:simonlin1212/agentskill-mcp"]
}
}
}Claude Code:
claude mcp add agentskill -- npx -y github:simonlin1212/agentskill-mcpSe compila en la primera instalación, así que la primera ejecución tarda unos segundos más que las siguientes. (Todavía no hay una publicación en npm; instalar desde el repositorio es la vía compatible hoy.)
Requiere Node 18+. Nada que configurar, sin clave de API, sin cuenta. El servidor no mantiene estado y no escribe nada en tu máquina.
Herramienta | Qué responde |
| "¿Hay algo para X?" Filtra por texto libre, módulo o nivel de precio. |
| "¿Qué recibo exactamente y cómo lo consigo?" Versión, tamaño del paquete, última actualización, repositorio de origen y la vía de obtención. |
| "¿Qué hay aquí?" Los cuatro módulos, sus recuentos y qué significan las dos formas de entrega. |
La búsqueda lee chino e inglés en un mismo índice, así que 去水印 y remove watermark llegan ambos a la misma
entrada. Escribas lo que escribas, los resultados incluyen ambos nombres.
Cuatro módulos — Finanzas, Comercio, Creador, General — que contienen dos formas:
Install Skill — un archivo de skill que carga el agente. Cópialo en el directorio de skills del agente.
Deploy App — una aplicación ejecutable con su propia interfaz. Despliégalo y úsalo junto al agente.
Aproximadamente la mitad de las entradas son gratuitas. Varias son la forma empaquetada de repositorios de código abierto con más de 16 000 estrellas combinadas en GitHub; cuando una capacidad se basa en un proyecto upstream, el upstream y su licencia se indican en la página del producto.
Es una herramienta de catálogo, no un espacio publicitario. Mantiene tres reglas:
Gratis significa gratis. Cuando el catálogo publica una URL de descarga, la herramienta la entrega directamente — sin cuenta, sin barrera. (Una entrada gratuita puede ocasionalmente no tener aún un archivo publicado; así lo indica.)
De pago indica el precio y se detiene. Sin urgencia, sin lenguaje de recomendación. El enlace de descarga de un elemento de pago se emite después del pago, en la página del pedido y por correo electrónico.
Concretamente: cualquier cosa que el catálogo no marque explícitamente como free se trata como de pago (cierre en fallo), se
elimina su campo de descarga, se extraen las URLs de sus campos de nombre, resultado, instalación, origen y
repositorio, y la salida ensamblada se
comprueba contra una lista blanca de rutas. Las únicas direcciones web que sobreviven son, en https://agentskill.nz
(opcionalmente bajo /zh): la raíz del sitio, /products/<handle>, /collections/<handle>,
/pages/<handle> y /cart o /cart/<variant>:<qty> — cada una coincidente como ruta completa, así que una ruta
más profunda como /products/x/download no califica. Cualquier otra cosa — otro host, http simple, un dominio
suplantado, /download?id=… — hace que el servidor se niegue a devolver ese registro.
Lo que esto no cubre: un nombre de host desnudo sin esquema (evil.example/x). Detectarlos significaría
tratar cualquier texto con forma de dominio como una dirección, y .md es un TLD real — SKILL.md en un manifiesto
de entrega se rechazaría como dominio, rompiendo todos los listados de pago. Estas respuestas son texto plano,
así que un nombre de host desnudo solo se convierte en enlace si algo posterior lo convierte en enlace.
La brecha que merece la pena nombrar: el catálogo se almacena en caché hasta cinco minutos. Un elemento que pasó a ser de pago dentro de esa ventana todavía se describe desde la instantánea anterior — incluida la URL de descarga directa que tenía mientras era gratuito.
Un comando de instalación es información, no autorización. get_capability — la única herramienta que devuelve
un comando de instalación — lo indica en cada llamada. Tu agente debería confirmar contigo antes de escribir en tu
sistema de archivos o gastar tu dinero.
La URL del catálogo se descubre desde el /llms.txt del sitio en tiempo de ejecución en lugar de estar codificada, así que el
servidor sigue funcionando cuando el sitio la rota.
npm install
npm run build
node test-smoke.mjs # drives the server over real MCP stdiotest-smoke.mjs habla el protocolo de red en lugar de importar el módulo, así que detecta errores de registro
y transporte que una prueba unitaria pasaría por alto. También verifica que ningún elemento de pago devuelva nunca una
URL de archivo.
Ver CHANGELOG.md.
Este servidor lee un catálogo y devuelve lo que dice. No evalúa si una capacidad se adapta a tu caso de uso y no ejecuta nada. Los archivos de skill son código que tu agente leerá y sobre el que actuará — revísalos antes de usarlos, igual que cualquier dependencia.
Si esto te ha ahorrado tiempo ☕
MIT
Autor: Simon Lin · X @linsizhen · Correo: simonlin0423@gmail.com
Full detail for one capability by id: version, package size, last update, source repository, licence-relevant facts, and exactly how to obtain it. Call this before installing anything.
| Name | Required | Description | Default |
|---|---|---|---|
| id | Yes | Capability id, e.g. 'a-stock-data'. Get it from search_capabilities. |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
Since no annotations are provided, the description carries full burden. It discloses the output contents (version, repository, license facts, etc.) and implies a read operation via the tool name 'get.' Does not explicitly declare read-only or mention side effects, but is fairly transparent.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
The description is a single, information-dense sentence: it lists the output contents and gives a clear usage directive. No filler or redundancy.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
Given the simple tool (1 param, no output schema, no annotations), the description covers purpose, usage, and output details. It does not explain potential errors or explicitly confirm read-only nature, but overall is sufficient for an agent to use it correctly.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema coverage is 100% for the single param with a description already stating 'Get it from search_capabilities.' The tool description does not add new parameter semantics beyond what the schema provides, meeting the baseline for high coverage.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description clearly states it provides 'full detail for one capability by id' and lists specific contents (version, package size, etc.), distinguishing it from sibling tools like search_capabilities which likely return a list.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
Explicitly advises to 'Call this before installing anything,' providing a clear when-to-use instruction. The schema also notes the id should come from search_capabilities, implying a prerequisite. No explicit alternatives or when-not cases, but the context is sufficient.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Overview of the catalog: the four modules, how many capabilities each holds, and what the two delivery forms mean.
| Name | Required | Description | Default |
|---|---|---|---|
No parameters | |||
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
No annotations are provided, so the description carries the full burden. It states the tool returns an overview of modules, capabilities, and delivery forms, but does not disclose whether the data is static, cached, or reflects real-time state. With no annotations, the description should clarify if this is a read-only, safe operation. A score of 3 reflects the description gives some useful behavioral context but lacks clarity on side effects or data freshness.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
The description is a single sentence of 18 words, very concise. It is front-loaded with the purpose ('Overview of the catalog') and then lists what it covers. It could be slightly more structured (e.g., bullet points implied), but for zero-param tool it's efficient.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
Given zero parameters and no output schema, the description must explain what the tool returns. It mentions 'the four modules, how many capabilities each holds, and what the two delivery forms mean,' which gives a clear idea of the output structure. However, without any annotations or output schema, the description should also clarify whether the output is a list, a summary, or a single object. The description is adequate but not fully complete.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
The schema has zero parameters and schema description coverage is 100%. Since there are no parameters to document, the description is not required to add anything. A baseline of 4 is appropriate because there is nothing to clarify.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description clearly specifies the verb (list/overview) and resource (catalog of modules, capabilities, delivery forms). It distinguishes from siblings: siblings 'get_capability' and 'search_capabilities' imply focused queries, while this tool provides a high-level catalog summary.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
The description implies this is an overview tool, but does not explicitly state when to use it vs siblings. For example, it doesn't say 'Use this for broad orientation; use get_capability for details on one module, or search_capabilities for filtered queries.' Guidance is implied but not explicit.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Search the AgentSkill catalog of installable AI-agent capabilities (skill files and small apps). Use when the user wants a capability the agent does not have — reading A-share/US market data, auditing code with a second model, transcribing audio, upscaling or de-watermarking images, cloning a site's design, deploying an overseas store, and so on. Returns name, module, form, price, version, size, and a direct download URL for free items.
| Name | Required | Description | Default |
|---|---|---|---|
| tier | No | Restrict to free or paid. | |
| limit | No | Max results, default 10. | |
| query | No | Free text. Matches name, description, repository, module and form, in Chinese and English. | |
| module | No | Restrict to one module. |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
With no annotations provided, the description carries the full burden for behavioral disclosure. It explains that the tool returns specific fields (name, module, etc.) and mentions a direct download URL for free items, implying paid items may behave differently. It doesn't cover query behavior or side effects, but for a non-destructive search tool, this is sufficient.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
The description is compact at three sentences, each earning its place: purpose, usage guidance with examples, return fields. Front-loaded with the core action. No redundant or filler content.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
For a search tool with 4 parameters and no output schema, the description is fairly complete. It covers purpose, when to use, what is returned. Could mention how parameters affect results but the schema handles that. Sibling tools fill the remaining gaps. Not missing anything critical.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema description coverage is 100% with each parameter already well-described (type, enum, constraints). The description adds overall context about free text matching and return fields but does not provide additional semantics beyond the schema. Baseline 3 is appropriate.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description clearly states the verb 'Search' and the resource 'AgentSkill catalog of installable AI-agent capabilities'. It specifies the tool's purpose: to find capabilities the agent doesn't have. This distinguishes it from siblings 'get_capability' (likely detail) and 'list_modules' (module listing).
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
The description explicitly tells when to use the tool: 'Use when the user wants a capability the agent does not have'. It provides concrete examples (e.g., market data, image upscaling). It doesn't explicitly state when not to use or name the sibling tools as alternatives, but the use case is clear enough.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
v0.1.0get_capabilitylist_modulessearch_capabilitiesScored across 3 tools
Each tool has a clearly distinct purpose: search_capabilities for searching, get_capability for full details by ID, and list_modules for a module overview. There is no functional overlap or ambiguity.
All tool names follow a consistent verb_noun pattern with snake_case (get_capability, search_capabilities, list_modules), making the set predictable and easy to understand.
With exactly three tools, the set is well-scoped for a read-only catalog exploration interface. Each tool earns its place and together they cover browsing, searching, and detail retrieval without being too few or excessive.
The tool surface fully covers the expected workflows for a capability catalog: listing modules for orientation, searching for capabilities, and retrieving full details including download information. There are no obvious gaps for the stated consumer-facing purpose.
Search and load AI agent skills. Free metadata; full content requires a Pro or Studio account.
Find AI agent skills: public previews and source links. Read-only discovery; no account required.
Find and try paid AI agent skills for security, on-chain data, code, markets, research and news.
Search, install and compose AI agent skills, loops and bundles from the LoopSkill marketplace.
